[Gluster-devel] [RFC PATCH 1/1] bd: Add FOPS for overload operations

[M. Mohan Kumar]

From: "M. Mohan Kumar" <address@hidden>

Add FOPs for overload operations instead of relying on setxattr interfaces.

Signed-off-by: M. Mohan Kumar <address@hidden>
---
 xlators/storage/bd/src/bd-helper.c |  14 +++-
 xlators/storage/bd/src/bd.c        | 144 ++++++++++++++++++++++++++++++++++++-
 xlators/storage/bd/src/bd.h        |   1 +
 3 files changed, 153 insertions(+), 6 deletions(-)

diff --git a/xlators/storage/bd/src/bd-helper.c 
b/xlators/storage/bd/src/bd-helper.c
index 63e26d8..45af163 100644
--- a/xlators/storage/bd/src/bd-helper.c
+++ b/xlators/storage/bd/src/bd-helper.c
@@ -51,6 +51,9 @@ bd_local_free (xlator_t *this, bd_local_t *local)
                 fd_unref (local->fd);
         else if (local->loc.path)
                 loc_wipe (&local->loc);
+        if (local->dfd)
+                fd_unref (local->dfd);
+
         if (local->dict)
                 dict_unref (local->dict);
         if (local->inode)
@@ -634,6 +637,7 @@ bd_clone (bd_local_t *local, bd_priv_t *priv)
         bd_gfid_t     source       = {0, };
         bd_gfid_t     dest         = {0, };
         void         *bufp[IOV_NR] = {0, };
+        uuid_t        gfid         = {0, };
 
         vec = GF_CALLOC (IOV_NR, sizeof (struct iovec), gf_common_mt_iovec);
         if (!vec)
@@ -648,15 +652,19 @@ bd_clone (bd_local_t *local, bd_priv_t *priv)
         }
 
         uuid_utoa_r (local->loc.gfid, source);
-        uuid_utoa_r (local->dloc->gfid, dest);
+        if (local->dfd)
+                uuid_copy (gfid, local->dfd->inode->gfid);
+        else
+                uuid_copy (gfid, local->dloc->gfid);
+
+        uuid_utoa_r (gfid, dest);
 
         gf_asprintf (&spath, "/dev/%s/%s", priv->vg, source);
         gf_asprintf (&dpath, "/dev/%s/%s", priv->vg, dest);
         if (!spath || !dpath)
                 goto out;
 
-        ret = bd_create (local->dloc->gfid, local->size,
-                         local->bdatt->type,  priv);
+        ret = bd_create (gfid, local->size, local->bdatt->type,  priv);
         if (ret)
                 goto out;
 
diff --git a/xlators/storage/bd/src/bd.c b/xlators/storage/bd/src/bd.c
index 17a9a5f..d2a3b00 100644
--- a/xlators/storage/bd/src/bd.c
+++ b/xlators/storage/bd/src/bd.c
@@ -1151,9 +1151,14 @@ bd_offload_getx_cbk (call_frame_t *frame, void *cookie, 
xlator_t *this,
                 goto out;
         }
 
-        STACK_WIND (frame, bd_offload_setx_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->setxattr,
-                    local->dloc, local->dict, 0, NULL);
+        if (!local->dfd)
+                STACK_WIND (frame, bd_offload_setx_cbk, FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->setxattr,
+                            local->dloc, local->dict, 0, NULL);
+        else
+                STACK_WIND (frame, bd_offload_setx_cbk, FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->fsetxattr,
+                            local->dfd, local->dict, 0, NULL);
 
         return 0;
 
@@ -2225,6 +2230,139 @@ err:
         return 0;
 }
 
+/*
+  splice: Copies data  between two  file descriptors. It transfers up to len
+  bytes of data from the file descriptor fd_in to the file descriptor fd_out.
+  FIXME: Check if destination file is on the same brick by checking
+  LINKTO xattr
+ */
+static int
+bd_splice (call_frame_t *frame, xlator_t *this, fd_t *fd_in, off_t offset_in,
+           fd_t *fd_out, off_t offset_out, size_t len, unsigned int flags,
+           dict_t *xdata)
+{
+        bd_attr_t  *bdatt_in  = NULL;
+        bd_attr_t  *bdatt_out = NULL;
+        bd_local_t *local     = NULL;
+        int         op_errno  = 0;
+
+        bd_inode_ctx_get (fd_in->inode, this, &bdatt_in);
+        if (!bdatt_in) {
+                op_errno = EINVAL;
+                gf_log (this->name, GF_LOG_WARNING, "source fd is not a BD");
+                goto out;
+        }
+
+        if (offset_in + len > bdatt_in->iatt.ia_size) {
+                op_errno = EINVAL;
+                gf_log (this->name, GF_LOG_WARNING, "Attempt to copy beyond "
+                        "source disk capacity");
+                goto out;
+        }
+
+        bd_inode_ctx_get (fd_out->inode, this, &bdatt_out);
+        if (bdatt_out) {
+                op_errno = EINVAL;
+                gf_log (this->name, GF_LOG_WARNING, "destination fd is already 
"
+                        "mapped to BD");
+                goto out;
+        }
+
+        if (!IA_ISREG (fd_out->inode->ia_type)) {
+                op_errno = EINVAL;
+                gf_log (this->name, GF_LOG_WARNING, "destination fd is not a "
+                        "regular file");
+                goto out;
+        }
+
+        /* set bd xattr */
+
+        local = bd_local_init (frame, this);
+        BD_VALIDATE_MEM_ALLOC (local, op_errno, out);
+
+        local->inode = inode_ref (fd_in->inode);
+        local->fd = fd_ref (fd_in);
+        local->dfd = fd_ref (fd_out);
+        local->offload = BD_OF_CLONE;
+
+        local->bdatt = CALLOC (1, sizeof (bd_attr_t));
+        BD_VALIDATE_MEM_ALLOC (local->bdatt, op_errno, out);
+
+        STACK_WIND (frame, bd_offload_getx_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->getxattr,
+                    &local->loc, BD_XATTR, NULL);
+
+out:
+        BD_STACK_UNWIND (splice, frame, -1, op_errno, NULL);
+        return 0;
+}
+
+/*
+  reflink: Create a COW file(BD) 'dloc' for given 'loc'
+
+  FIXME: Check if destination file is on the same brick by checking
+  LINKTO xattr
+
+  http://lwn.net/Articles/331808/
+*/
+static int
+bd_reflink (call_frame_t *frame, xlator_t *this,
+            loc_t *loc, loc_t *refloc, dict_t *xdata)
+{
+        bd_attr_t  *bdatt_in = NULL;
+        bd_attr_t  *bdatt_out = NULL;
+        bd_local_t *local     = NULL;
+        int         op_errno  = 0;
+
+        bd_inode_ctx_get (loc->inode, this, &bdatt_in);
+        if (!bdatt_in) {
+                op_errno = EINVAL;
+                gf_log (this->name, GF_LOG_WARNING, "source file is not a BD");
+                goto out;
+        }
+
+        bd_inode_ctx_get (refloc->inode, this, &bdatt_out);
+        if (bdatt_out) {
+                op_errno = EINVAL;
+                gf_log (this->name, GF_LOG_WARNING, "destination file is "
+                        "already mapped to BD");
+                goto out;
+        }
+
+        if (!IA_ISREG (refloc->inode->ia_type)) {
+                op_errno = EINVAL;
+                gf_log (this->name, GF_LOG_WARNING, "destination file is not a 
"
+                        "regular file");
+                goto out;
+        }
+
+        local = bd_local_init (frame, this);
+        BD_VALIDATE_MEM_ALLOC (local, op_errno, out);
+
+        local->inode = inode_ref (loc->inode);
+        loc_copy (&local->loc, loc);
+        local->inode = inode_ref (loc->inode);
+
+        local->dloc = CALLOC (1, sizeof (loc_t));
+        BD_VALIDATE_MEM_ALLOC (local->dloc, op_errno, out);
+
+        loc_copy (local->dloc, refloc);
+        local->offload = BD_OF_SNAPSHOT;
+
+        local->size = bd_get_default_extent (this->private);
+
+        local->bdatt = CALLOC (1, sizeof (bd_attr_t));
+        BD_VALIDATE_MEM_ALLOC (local->bdatt, op_errno, out);
+
+        STACK_WIND (frame, bd_offload_getx_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->getxattr,
+                    &local->loc, BD_XATTR, NULL);
+
+out:
+        BD_STACK_UNWIND (reflink, frame, -1, op_errno, NULL);
+        return 0;
+}
+
 /**
  * notify - when parent sends PARENT_UP, send CHILD_UP event from here
  */
diff --git a/xlators/storage/bd/src/bd.h b/xlators/storage/bd/src/bd.h
index f59bc6a..0fc08fc 100644
--- a/xlators/storage/bd/src/bd.h
+++ b/xlators/storage/bd/src/bd.h
@@ -150,6 +150,7 @@ typedef struct {
         bd_offload_t offload;
         uint64_t     size;
         loc_t       *dloc;
+        fd_t        *dfd;
 } bd_local_t;
 
 /* Prototypes */
-- 
1.7.11.7