Whamcloud - gitweb
Branch HEAD
[fs/lustre-release.git] / lustre / mdd / mdd_object.c
index 9846cb4..c2b1d7e 100644 (file)
@@ -304,7 +304,11 @@ static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
 static int mdd_object_print(const struct lu_env *env, void *cookie,
                             lu_printer_t p, const struct lu_object *o)
 {
-        return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p", o);
+        struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);
+        return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p(open_count=%d, "
+                    "valid=%x, cltime=%llu, flags=%lx)",
+                    mdd, mdd->mod_count, mdd->mod_valid,
+                    mdd->mod_cltime, mdd->mod_flags);
 }
 
 static const struct lu_object_operations mdd_lu_obj_ops = {
@@ -385,6 +389,7 @@ out:
 /** mdd_path() lookup structure. */
 struct path_lookup_info {
         __u64                pli_recno;        /**< history point */
+        __u64                pli_currec;       /**< current record */
         struct lu_fid        pli_fid;
         struct lu_fid        pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
         struct mdd_object   *pli_mdd_obj;
@@ -473,7 +478,12 @@ static int mdd_path_current(const struct lu_env *env,
                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
         }
 
-        /* Verify that our path hasn't changed since we started the lookup */
+        /* Verify that our path hasn't changed since we started the lookup.
+           Record the current index, and verify the path resolves to the
+           same fid. If it does, then the path is correct as of this index. */
+        spin_lock(&mdd->mdd_cl.mc_lock);
+        pli->pli_currec = mdd->mdd_cl.mc_index;
+        spin_unlock(&mdd->mdd_cl.mc_lock);
         rc = mdd_path2fid(env, mdd, ptr, &pli->pli_fid);
         if (rc) {
                 CDEBUG(D_INFO, "mdd_path2fid(%s) failed %d\n", ptr, rc);
@@ -497,9 +507,15 @@ out:
         return rc;
 }
 
+static int mdd_path_historic(const struct lu_env *env,
+                             struct path_lookup_info *pli)
+{
+        return 0;
+}
+
 /* Returns the full path to this fid, as of changelog record recno. */
 static int mdd_path(const struct lu_env *env, struct md_object *obj,
-                    char *path, int pathlen, __u64 recno, int *linkno)
+                    char *path, int pathlen, __u64 *recno, int *linkno)
 {
         struct path_lookup_info *pli;
         int tries = 3;
@@ -520,7 +536,7 @@ static int mdd_path(const struct lu_env *env, struct md_object *obj,
                 RETURN(-ENOMEM);
 
         pli->pli_mdd_obj = md2mdd_obj(obj);
-        pli->pli_recno = recno;
+        pli->pli_recno = *recno;
         pli->pli_path = path;
         pli->pli_pathlen = pathlen;
         pli->pli_linkno = *linkno;
@@ -529,7 +545,6 @@ static int mdd_path(const struct lu_env *env, struct md_object *obj,
         while (tries-- && rc == -EAGAIN)
                 rc = mdd_path_current(env, pli);
 
-#if 0   /* We need old path names only for replication */
         /* For historical path lookup, the current links may not have existed
          * at "recno" time.  We must switch over to earlier links/parents
          * by using the changelog records.  If the earlier parent doesn't
@@ -538,12 +553,13 @@ static int mdd_path(const struct lu_env *env, struct md_object *obj,
          * We may ignore this problem for the initial implementation and
          * state that an "original" hardlink must still exist for us to find
          * historic path name. */
-        if (pli->pli_recno != -1)
+        if (pli->pli_recno != -1) {
                 rc = mdd_path_historic(env, pli);
-#endif
-
-        /* return next link index to caller */
-        *linkno = pli->pli_linkno;
+        } else {
+                *recno = pli->pli_currec;
+                /* Return next link index to caller */
+                *linkno = pli->pli_linkno;
+        }
 
         OBD_FREE_PTR(pli);
 
@@ -1209,7 +1225,8 @@ static int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
         unsigned int qnids[MAXQUOTAS] = { 0, 0 };
         unsigned int qoids[MAXQUOTAS] = { 0, 0 };
         int quota_opc = 0, block_count = 0;
-        int inode_pending = 0, block_pending = 0;
+        int inode_pending[MAXQUOTAS] = { 0, 0 };
+        int block_pending[MAXQUOTAS] = { 0, 0 };
 #endif
         ENTRY;
 
@@ -1253,20 +1270,17 @@ static int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
                         mdd_quota_wrapper(la_copy, qnids);
                         mdd_quota_wrapper(la_tmp, qoids);
                         /* get file quota for new owner */
-                        lquota_chkquota(mds_quota_interface_ref, obd,
-                                        qnids[USRQUOTA], qnids[GRPQUOTA], 1,
-                                        &inode_pending, NULL, 0, NULL, 0);
+                        lquota_chkquota(mds_quota_interface_ref, obd, qnids,
+                                        inode_pending, 1, NULL, 0, NULL, 0);
                         block_count = (la_tmp->la_blocks + 7) >> 3;
                         if (block_count) {
                                 void *data = NULL;
                                 mdd_data_get(env, mdd_obj, &data);
                                 /* get block quota for new owner */
                                 lquota_chkquota(mds_quota_interface_ref, obd,
-                                                qnids[USRQUOTA],
-                                                qnids[GRPQUOTA],
-                                                block_count, &block_pending,
-                                                NULL, LQUOTA_FLAGS_BLK,
-                                                data, 1);
+                                                qnids, block_pending,
+                                                block_count, NULL,
+                                                LQUOTA_FLAGS_BLK, data, 1);
                         }
                 }
         }
@@ -1319,14 +1333,10 @@ cleanup:
         }
 #ifdef HAVE_QUOTA_SUPPORT
         if (quota_opc) {
-                if (inode_pending)
-                        lquota_pending_commit(mds_quota_interface_ref, obd,
-                                              qnids[USRQUOTA], qnids[GRPQUOTA],
-                                              inode_pending, 0);
-                if (block_pending)
-                        lquota_pending_commit(mds_quota_interface_ref, obd,
-                                              qnids[USRQUOTA], qnids[GRPQUOTA],
-                                              block_pending, 1);
+                lquota_pending_commit(mds_quota_interface_ref, obd, qnids,
+                                      inode_pending, 0);
+                lquota_pending_commit(mds_quota_interface_ref, obd, qnids,
+                                      block_pending, 1);
                 /* Trigger dqrel/dqacq for original owner and new owner.
                  * If failed, the next call for lquota_chkquota will
                  * process it. */
@@ -1565,7 +1575,8 @@ static int mdd_object_create(const struct lu_env *env,
         struct mds_obd *mds = &obd->u.mds;
         unsigned int qids[MAXQUOTAS] = { 0, 0 };
         int quota_opc = 0, block_count = 0;
-        int inode_pending = 0, block_pending = 0;
+        int inode_pending[MAXQUOTAS] = { 0, 0 };
+        int block_pending[MAXQUOTAS] = { 0, 0 };
 #endif
         int rc = 0;
         ENTRY;
@@ -1575,9 +1586,8 @@ static int mdd_object_create(const struct lu_env *env,
                 quota_opc = FSFILT_OP_CREATE_PARTIAL_CHILD;
                 mdd_quota_wrapper(&ma->ma_attr, qids);
                 /* get file quota for child */
-                lquota_chkquota(mds_quota_interface_ref, obd, qids[USRQUOTA],
-                                qids[GRPQUOTA], 1, &inode_pending, NULL, 0,
-                                NULL, 0);
+                lquota_chkquota(mds_quota_interface_ref, obd, qids,
+                                inode_pending, 1, NULL, 0, NULL, 0);
                 switch (ma->ma_attr.la_mode & S_IFMT) {
                 case S_IFLNK:
                 case S_IFDIR:
@@ -1589,9 +1599,8 @@ static int mdd_object_create(const struct lu_env *env,
                 }
                 /* get block quota for child */
                 if (block_count)
-                        lquota_chkquota(mds_quota_interface_ref, obd,
-                                        qids[USRQUOTA], qids[GRPQUOTA],
-                                        block_count, &block_pending, NULL,
+                        lquota_chkquota(mds_quota_interface_ref, obd, qids,
+                                        block_pending, block_count, NULL,
                                         LQUOTA_FLAGS_BLK, NULL, 0);
         }
 #endif
@@ -1659,18 +1668,14 @@ unlock:
 out_pending:
 #ifdef HAVE_QUOTA_SUPPORT
         if (quota_opc) {
-                if (inode_pending)
-                        lquota_pending_commit(mds_quota_interface_ref, obd,
-                                              qids[USRQUOTA], qids[GRPQUOTA],
-                                              inode_pending, 0);
-                if (block_pending)
-                        lquota_pending_commit(mds_quota_interface_ref, obd,
-                                              qids[USRQUOTA], qids[GRPQUOTA],
-                                              block_pending, 1);
+                lquota_pending_commit(mds_quota_interface_ref, obd, qids,
+                                      inode_pending, 0);
+                lquota_pending_commit(mds_quota_interface_ref, obd, qids,
+                                      block_pending, 1);
                 /* Trigger dqacq on the owner of child. If failed,
                  * the next call for lquota_chkquota will process it. */
                 lquota_adjust(mds_quota_interface_ref, obd, qids, 0, rc,
-                              FSFILT_OP_CREATE_PARTIAL_CHILD);
+                              quota_opc);
         }
 #endif
         return rc;
@@ -1841,6 +1846,7 @@ static int mdd_close(const struct lu_env *env, struct md_object *obj,
                      struct md_attr *ma)
 {
         struct mdd_object *mdd_obj = md2mdd_obj(obj);
+        struct mdd_device *mdd = mdo2mdd(obj);
         struct thandle    *handle;
         int rc;
         int reset = 1;
@@ -1864,27 +1870,53 @@ static int mdd_close(const struct lu_env *env, struct md_object *obj,
         /* release open count */
         mdd_obj->mod_count --;
 
-        if (mdd_obj->mod_count == 0) {
+        if (mdd_obj->mod_count == 0 && mdd_obj->mod_flags & ORPHAN_OBJ) {
                 /* remove link to object from orphan index */
-                if (mdd_obj->mod_flags & ORPHAN_OBJ)
-                        __mdd_orphan_del(env, mdd_obj, handle);
+                rc = __mdd_orphan_del(env, mdd_obj, handle);
+                if (rc == 0) {
+                        CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
+                               "list, OSS objects to be destroyed.\n",
+                               PFID(mdd_object_fid(mdd_obj)));
+                } else {
+                        CERROR("Object "DFID" can not be deleted from orphan "
+                                "list, maybe cause OST objects can not be "
+                                "destroyed (err: %d).\n",
+                                PFID(mdd_object_fid(mdd_obj)), rc);
+                        /* If object was not deleted from orphan list, do not
+                         * destroy OSS objects, which will be done when next
+                         * recovery. */
+                        GOTO(out, rc);
+                }
         }
 
         rc = mdd_iattr_get(env, mdd_obj, ma);
-        if (rc == 0) {
-                if (mdd_obj->mod_count == 0 && ma->ma_attr.la_nlink == 0) {
-                        rc = mdd_object_kill(env, mdd_obj, ma);
+        /* Object maybe not in orphan list originally, it is rare case for
+         * mdd_finish_unlink() failure. */
+        if (rc == 0 && ma->ma_attr.la_nlink == 0) {
 #ifdef HAVE_QUOTA_SUPPORT
-                        if (mds->mds_quota) {
-                                quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
-                                mdd_quota_wrapper(&ma->ma_attr, qids);
-                        }
+                if (mds->mds_quota) {
+                        quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
+                        mdd_quota_wrapper(&ma->ma_attr, qids);
+                }
 #endif
-                        if (rc == 0)
-                                reset = 0;
+                /* MDS_CLOSE_CLEANUP means destroy OSS objects by MDS. */
+                if (ma->ma_valid & MA_FLAGS &&
+                    ma->ma_attr_flags & MDS_CLOSE_CLEANUP) {
+                        rc = mdd_lov_destroy(env, mdd, mdd_obj, &ma->ma_attr);
+                } else {
+                        rc = mdd_object_kill(env, mdd_obj, ma);
+                                if (rc == 0)
+                                        reset = 0;
                 }
+
+                if (rc != 0)
+                        CERROR("Error when prepare to delete Object "DFID" , "
+                               "which will cause OST objects can not be "
+                               "destroyed.\n",  PFID(mdd_object_fid(mdd_obj)));
         }
+        EXIT;
 
+out:
         if (reset)
                 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
 
@@ -1897,7 +1929,7 @@ static int mdd_close(const struct lu_env *env, struct md_object *obj,
                 lquota_adjust(mds_quota_interface_ref, obd, qids, 0, rc,
                               quota_opc);
 #endif
-        RETURN(rc);
+        return rc;
 }
 
 /*
@@ -2032,6 +2064,8 @@ static int mdd_dir_page_build(const struct lu_env *env, struct mdd_device *mdd,
                         memcpy(ent->lde_name, name, len);
 
                         result = mdd_append_attrs(env, mdd, attr, iops, it, ent);
+                        if (result == -ESTALE)
+                                goto next;
                         if (result != 0)
                                 goto out;
                 } else {
@@ -2054,6 +2088,8 @@ static int mdd_dir_page_build(const struct lu_env *env, struct mdd_device *mdd,
 
 next:
                 result = iops->next(env, it);
+                if (result == -ESTALE)
+                        goto next;
         } while (result == 0);
 
 out:
@@ -2157,8 +2193,8 @@ static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj,
         return rc;
 }
 
-static int mdd_readpage(const struct lu_env *env, struct md_object *obj,
-                        const struct lu_rdpg *rdpg)
+int mdd_readpage(const struct lu_env *env, struct md_object *obj,
+                 const struct lu_rdpg *rdpg)
 {
         struct mdd_object *mdd_obj = md2mdd_obj(obj);
         int rc;
@@ -2215,6 +2251,24 @@ static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
         return next->do_ops->do_object_sync(env, next);
 }
 
+static dt_obj_version_t mdd_version_get(const struct lu_env *env,
+                                        struct md_object *obj)
+{
+        struct mdd_object *mdd_obj = md2mdd_obj(obj);
+
+        LASSERT(mdd_object_exists(mdd_obj));
+        return do_version_get(env, mdd_object_child(mdd_obj));
+}
+
+static void mdd_version_set(const struct lu_env *env, struct md_object *obj,
+                            dt_obj_version_t version)
+{
+        struct mdd_object *mdd_obj = md2mdd_obj(obj);
+
+        LASSERT(mdd_object_exists(mdd_obj));
+        return do_version_set(env, mdd_object_child(mdd_obj), version);
+}
+
 const struct md_object_operations mdd_obj_ops = {
         .moo_permission    = mdd_permission,
         .moo_attr_get      = mdd_attr_get,
@@ -2232,5 +2286,7 @@ const struct md_object_operations mdd_obj_ops = {
         .moo_readlink      = mdd_readlink,
         .moo_capa_get      = mdd_capa_get,
         .moo_object_sync   = mdd_object_sync,
+        .moo_version_get   = mdd_version_get,
+        .moo_version_set   = mdd_version_set,
         .moo_path          = mdd_path,
 };