Exemplo n.º 1
0
/*
 * VBR: save parent version in reply and child version getting by its name.
 * Version of child is getting and checking during its lookup. If
 */
static int mdt_reint_unlink(struct mdt_thread_info *info,
                            struct mdt_lock_handle *lhc)
{
        struct mdt_reint_record *rr = &info->mti_rr;
        struct ptlrpc_request   *req = mdt_info_req(info);
        struct md_attr          *ma = &info->mti_attr;
        struct lu_fid           *child_fid = &info->mti_tmp_fid1;
        struct mdt_object       *mp;
        struct mdt_object       *mc;
        struct mdt_lock_handle  *parent_lh;
        struct mdt_lock_handle  *child_lh;
        struct lu_name          *lname;
        int                      rc;
	int			 no_name = 0;
	ENTRY;

        DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
                  rr->rr_name);

        if (info->mti_dlm_req)
                ldlm_request_cancel(req, info->mti_dlm_req, 0);

        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
                RETURN(err_serious(-ENOENT));

	if (fid_is_obf(rr->rr_fid1) || fid_is_dot_lustre(rr->rr_fid1))
		RETURN(-EPERM);
        /*
	 * step 1: Found the parent.
         */
	mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
	if (IS_ERR(mp)) {
		rc = PTR_ERR(mp);
		GOTO(out, rc);
	}

	parent_lh = &info->mti_lh[MDT_LH_PARENT];
	lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
	if (mdt_object_remote(mp)) {
		mdt_lock_reg_init(parent_lh, LCK_EX);
		rc = mdt_remote_object_lock(info, mp, &parent_lh->mlh_rreg_lh,
					    parent_lh->mlh_rreg_mode,
					    MDS_INODELOCK_UPDATE);
		if (rc != ELDLM_OK)
			GOTO(put_parent, rc);

	} else {
		mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
				  rr->rr_namelen);
		rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
				     MDT_LOCAL_LOCK);
		if (rc)
			GOTO(put_parent, rc);

		rc = mdt_version_get_check_save(info, mp, 0);
		if (rc)
			GOTO(unlock_parent, rc);
	}

	/* step 2: find & lock the child */
	/* lookup child object along with version checking */
	fid_zero(child_fid);
	rc = mdt_lookup_version_check(info, mp, lname, child_fid, 1);
	if (rc != 0) {
		/* Name might not be able to find during resend of
		 * remote unlink, considering following case.
		 * dir_A is a remote directory, the name entry of
		 * dir_A is on MDT0, the directory is on MDT1,
		 *
		 * 1. client sends unlink req to MDT1.
		 * 2. MDT1 sends name delete update to MDT0.
		 * 3. name entry is being deleted in MDT0 synchronously.
		 * 4. MDT1 is restarted.
		 * 5. client resends unlink req to MDT1. So it can not
		 *    find the name entry on MDT0 anymore.
		 * In this case, MDT1 only needs to destory the local
		 * directory.
		 * */
		if (mdt_object_remote(mp) && rc == -ENOENT &&
		    !fid_is_zero(rr->rr_fid2) &&
		    lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
			no_name = 1;
			*child_fid = *rr->rr_fid2;
		 } else {
			GOTO(unlock_parent, rc);
		 }
	}

	if (fid_is_obf(child_fid) || fid_is_dot_lustre(child_fid))
		GOTO(unlock_parent, rc = -EPERM);

        mdt_reint_init_ma(info, ma);

	/* We will lock the child regardless it is local or remote. No harm. */
	mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
	if (IS_ERR(mc))
		GOTO(unlock_parent, rc = PTR_ERR(mc));

        child_lh = &info->mti_lh[MDT_LH_CHILD];
        mdt_lock_reg_init(child_lh, LCK_EX);
	if (mdt_object_remote(mc)) {
		struct mdt_body	 *repbody;

		if (!fid_is_zero(rr->rr_fid2)) {
			CDEBUG(D_INFO, "%s: name %s can not find "DFID"\n",
			       mdt_obd_name(info->mti_mdt),
			       (char *)rr->rr_name, PFID(mdt_object_fid(mc)));
			GOTO(put_child, rc = -ENOENT);
		}
		CDEBUG(D_INFO, "%s: name %s: "DFID" is another MDT\n",
		       mdt_obd_name(info->mti_mdt),
		       (char *)rr->rr_name, PFID(mdt_object_fid(mc)));

		if (!mdt_is_dne_client(req->rq_export))
			/* Return -EIO for old client */
			GOTO(put_child, rc = -EIO);

		if (info->mti_spec.sp_rm_entry) {
			struct lu_ucred *uc  = mdt_ucred(info);

			if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
				CERROR("%s: unlink remote entry is only "
				       "permitted for administrator: rc = %d\n",
					mdt_obd_name(info->mti_mdt),
					-EPERM);
				GOTO(put_child, rc = -EPERM);
			}

			ma->ma_need = MA_INODE;
			ma->ma_valid = 0;
			mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
			rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
					NULL, lname, ma, no_name);
			GOTO(put_child, rc);
		}
		/* Revoke the LOOKUP lock of the remote object granted by
		 * this MDT. Since the unlink will happen on another MDT,
		 * it will release the LOOKUP lock right away. Then What
		 * would happen if another client try to grab the LOOKUP
		 * lock at the same time with unlink XXX */
		mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
				MDT_CROSS_LOCK);
		repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
		LASSERT(repbody != NULL);
		repbody->fid1 = *mdt_object_fid(mc);
		repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
		GOTO(unlock_child, rc = -EREMOTE);
	} else if (info->mti_spec.sp_rm_entry) {
		rc = -EPERM;
		CDEBUG(D_INFO, "%s: no rm_entry on local dir '%s': rc = %d\n",
		       mdt_obd_name(info->mti_mdt), (char *)rr->rr_name, rc);
		GOTO(put_child, rc);
	}

	/* We used to acquire MDS_INODELOCK_FULL here but we can't do
	 * this now because a running HSM restore on the child (unlink
	 * victim) will hold the layout lock. See LU-4002. */
	rc = mdt_object_lock(info, mc, child_lh,
			     MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
			     MDT_CROSS_LOCK);
	if (rc != 0)
		GOTO(put_child, rc);

        mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
                       OBD_FAIL_MDS_REINT_UNLINK_WRITE);
        /* save version when object is locked */
        mdt_version_get_save(info, mc, 1);
	/*
	 * Now we can only make sure we need MA_INODE, in mdd layer, will check
	 * whether need MA_LOV and MA_COOKIE.
	 */
	ma->ma_need = MA_INODE;
	ma->ma_valid = 0;
	mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);

	mutex_lock(&mc->mot_lov_mutex);

	rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
			mdt_object_child(mc), lname, ma, no_name);

	mutex_unlock(&mc->mot_lov_mutex);

	if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
		rc = mdt_attr_get_complex(info, mc, ma);
	if (rc == 0)
		mdt_handle_last_unlink(info, mc, ma);

        if (ma->ma_valid & MA_INODE) {
                switch (ma->ma_attr.la_mode & S_IFMT) {
                case S_IFDIR:
			mdt_counter_incr(req, LPROC_MDT_RMDIR);
                        break;
                case S_IFREG:
                case S_IFLNK:
                case S_IFCHR:
                case S_IFBLK:
                case S_IFIFO:
                case S_IFSOCK:
			mdt_counter_incr(req, LPROC_MDT_UNLINK);
                        break;
                default:
                        LASSERTF(0, "bad file type %o unlinking\n",
                                 ma->ma_attr.la_mode);
                }
        }

        EXIT;
unlock_child:
	mdt_object_unlock(info, mc, child_lh, rc);
put_child:
	mdt_object_put(info->mti_env, mc);
unlock_parent:
	mdt_object_unlock(info, mp, parent_lh, rc);
put_parent:
	mdt_object_put(info->mti_env, mp);
out:
        return rc;
}
Exemplo n.º 2
0
/*
 * VBR: save parent version in reply and child version getting by its name.
 * Version of child is getting and checking during its lookup. If
 */
static int mdt_reint_unlink(struct mdt_thread_info *info,
                            struct mdt_lock_handle *lhc)
{
        struct mdt_reint_record *rr = &info->mti_rr;
        struct ptlrpc_request   *req = mdt_info_req(info);
        struct md_attr          *ma = &info->mti_attr;
        struct lu_fid           *child_fid = &info->mti_tmp_fid1;
        struct mdt_object       *mp;
        struct mdt_object       *mc;
        struct mdt_lock_handle  *parent_lh;
        struct mdt_lock_handle  *child_lh;
        struct lu_name          *lname;
        int                      rc;
        ENTRY;

        DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
                  rr->rr_name);

        if (info->mti_dlm_req)
                ldlm_request_cancel(req, info->mti_dlm_req, 0);

        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
                RETURN(err_serious(-ENOENT));

        /*
	 * step 1: Found the parent.
         */
	mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
	if (IS_ERR(mp)) {
		rc = PTR_ERR(mp);
		GOTO(out, rc);
	}

	if (mdt_object_obf(mp))
		GOTO(put_parent, rc = -EPERM);

	parent_lh = &info->mti_lh[MDT_LH_PARENT];
	lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
	if (mdt_object_remote(mp)) {
		mdt_lock_reg_init(parent_lh, LCK_EX);
		rc = mdt_remote_object_lock(info, mp, &parent_lh->mlh_rreg_lh,
					    parent_lh->mlh_rreg_mode,
					    MDS_INODELOCK_UPDATE);
		if (rc != ELDLM_OK)
			GOTO(put_parent, rc);

	} else {
		mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
				  rr->rr_namelen);
		rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
				     MDT_LOCAL_LOCK);
		if (rc)
			GOTO(put_parent, rc);

		rc = mdt_version_get_check_save(info, mp, 0);
		if (rc)
			GOTO(unlock_parent, rc);
	}

	/* step 2: find & lock the child */
	/* lookup child object along with version checking */
	fid_zero(child_fid);
	rc = mdt_lookup_version_check(info, mp, lname, child_fid, 1);
	if (rc != 0)
		GOTO(unlock_parent, rc);

        mdt_reint_init_ma(info, ma);

	/* We will lock the child regardless it is local or remote. No harm. */
	mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
	if (IS_ERR(mc))
		GOTO(unlock_parent, rc = PTR_ERR(mc));

        child_lh = &info->mti_lh[MDT_LH_CHILD];
        mdt_lock_reg_init(child_lh, LCK_EX);
	if (mdt_object_remote(mc)) {
		struct mdt_body	 *repbody;

		if (!fid_is_zero(rr->rr_fid2)) {
			CDEBUG(D_INFO, "%s: name %s can not find "DFID"\n",
			       mdt2obd_dev(info->mti_mdt)->obd_name,
			       (char *)rr->rr_name, PFID(mdt_object_fid(mc)));
			GOTO(unlock_parent, rc = -ENOENT);
		}
		CDEBUG(D_INFO, "%s: name %s: "DFID" is another MDT\n",
		       mdt2obd_dev(info->mti_mdt)->obd_name,
		       (char *)rr->rr_name, PFID(mdt_object_fid(mc)));

		if (info->mti_spec.sp_rm_entry) {
			struct lu_ucred *uc  = mdt_ucred(info);

			if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
				CERROR("%s: unlink remote entry is only "
				       "permitted for administrator: rc = %d\n",
					mdt2obd_dev(info->mti_mdt)->obd_name,
					-EPERM);
				GOTO(unlock_parent, rc = -EPERM);
			}

			ma->ma_need = MA_INODE;
			ma->ma_valid = 0;
			mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
			rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
					NULL, lname, ma);
			mdt_object_put(info->mti_env, mc);
			GOTO(unlock_parent, rc);
		}
		/* Revoke the LOOKUP lock of the remote object granted by
		 * this MDT. Since the unlink will happen on another MDT,
		 * it will release the LOOKUP lock right away. Then What
		 * would happen if another client try to grab the LOOKUP
		 * lock at the same time with unlink XXX */
		mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
				MDT_CROSS_LOCK);
		repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
		LASSERT(repbody != NULL);
		repbody->fid1 = *mdt_object_fid(mc);
		repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
		mdt_object_unlock_put(info, mc, child_lh, rc);
		GOTO(unlock_parent, rc = -EREMOTE);
	} else if (info->mti_spec.sp_rm_entry) {
		CERROR("%s: lfs rmdir should not be used on local dir %s\n",
		       mdt2obd_dev(info->mti_mdt)->obd_name,
		       (char *)rr->rr_name);
		mdt_object_put(info->mti_env, mc);
		GOTO(unlock_parent, rc = -EPERM);
	}

        rc = mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_FULL,
                             MDT_CROSS_LOCK);
	if (rc != 0) {
		mdt_object_put(info->mti_env, mc);
		GOTO(unlock_parent, rc);
	}

        mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
                       OBD_FAIL_MDS_REINT_UNLINK_WRITE);
        /* save version when object is locked */
        mdt_version_get_save(info, mc, 1);
        /*
         * Now we can only make sure we need MA_INODE, in mdd layer, will check
         * whether need MA_LOV and MA_COOKIE.
         */
        ma->ma_need = MA_INODE;
        ma->ma_valid = 0;
        mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
        rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
                        mdt_object_child(mc), lname, ma);
	if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
		rc = mdt_attr_get_complex(info, mc, ma);
        if (rc == 0)
                mdt_handle_last_unlink(info, mc, ma);

        if (ma->ma_valid & MA_INODE) {
                switch (ma->ma_attr.la_mode & S_IFMT) {
                case S_IFDIR:
			mdt_counter_incr(req, LPROC_MDT_RMDIR);
                        break;
                case S_IFREG:
                case S_IFLNK:
                case S_IFCHR:
                case S_IFBLK:
                case S_IFIFO:
                case S_IFSOCK:
			mdt_counter_incr(req, LPROC_MDT_UNLINK);
                        break;
                default:
                        LASSERTF(0, "bad file type %o unlinking\n",
                                 ma->ma_attr.la_mode);
                }
        }

        EXIT;

        mdt_object_unlock_put(info, mc, child_lh, rc);
unlock_parent:
	mdt_object_unlock(info, mp, parent_lh, rc);
put_parent:
	mdt_object_put(info->mti_env, mp);
out:
        return rc;
}
Exemplo n.º 3
0
/*
 * VBR: save parent version in reply and child version getting by its name.
 * Version of child is getting and checking during its lookup. If
 */
static int mdt_reint_unlink(struct mdt_thread_info *info,
                            struct mdt_lock_handle *lhc)
{
        struct mdt_reint_record *rr = &info->mti_rr;
        struct ptlrpc_request   *req = mdt_info_req(info);
        struct md_attr          *ma = &info->mti_attr;
        struct lu_fid           *child_fid = &info->mti_tmp_fid1;
        struct mdt_object       *mp;
        struct mdt_object       *mc;
        struct mdt_lock_handle  *parent_lh;
        struct mdt_lock_handle  *child_lh;
        struct lu_name          *lname;
        int                      rc;
        ENTRY;

        DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
                  rr->rr_name);

        if (info->mti_dlm_req)
                ldlm_request_cancel(req, info->mti_dlm_req, 0);

        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
                RETURN(err_serious(-ENOENT));

        /*
         * step 1: lock the parent. Note, this may be child in case of
         * remote operation denoted by ->mti_cross_ref flag.
         */
        parent_lh = &info->mti_lh[MDT_LH_PARENT];
        if (info->mti_cross_ref) {
                /*
                 * Init reg lock for cross ref case when we need to do only
                 * ref del locally.
                 */
                mdt_lock_reg_init(parent_lh, LCK_PW);
        } else {
                mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
                                  rr->rr_namelen);
        }
        mp = mdt_object_find_lock(info, rr->rr_fid1, parent_lh,
                                  MDS_INODELOCK_UPDATE, MDT_OBJ_MUST_EXIST);
        if (IS_ERR(mp)) {
                rc = PTR_ERR(mp);
                /* errors are possible here in cross-ref cases, see below */
                if (info->mti_cross_ref)
                        rc = 0;
                GOTO(out, rc);
        }

        rc = mdt_version_get_check_save(info, mp, 0);
        if (rc)
                GOTO(out_unlock_parent, rc);

        mdt_reint_init_ma(info, ma);
        if (!ma->ma_lmm || !ma->ma_cookie)
                GOTO(out_unlock_parent, rc = -EINVAL);

        if (info->mti_cross_ref) {
                /*
                 * Remote partial operation. It is possible that replay may
                 * happen on parent MDT and this operation will be repeated.
                 * Therefore the object absense is allowed case and nothing
                 * should be done here.
                 */
                if (mdt_object_exists(mp) > 0) {
                        mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
                        rc = mo_ref_del(info->mti_env,
                                        mdt_object_child(mp), ma);
                        if (rc == 0)
                                mdt_handle_last_unlink(info, mp, ma);
                } else
                        rc = 0;
                GOTO(out_unlock_parent, rc);
        }

        /* step 2: find & lock the child */
        lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
        /* lookup child object along with version checking */
        rc = mdt_lookup_version_check(info, mp, lname, child_fid, 1);
        if (rc != 0)
                 GOTO(out_unlock_parent, rc);

        /* We will lock the child regardless it is local or remote. No harm. */
        mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid,
                             MDT_OBJ_MUST_EXIST);
        if (IS_ERR(mc))
                GOTO(out_unlock_parent, rc = PTR_ERR(mc));
        child_lh = &info->mti_lh[MDT_LH_CHILD];
        mdt_lock_reg_init(child_lh, LCK_EX);
        rc = mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_FULL,
                             MDT_CROSS_LOCK);
        if (rc != 0) {
                mdt_object_put(info->mti_env, mc);
                GOTO(out_unlock_parent, rc);
        }

        mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
                       OBD_FAIL_MDS_REINT_UNLINK_WRITE);
        /* save version when object is locked */
        mdt_version_get_save(info, mc, 1);
        /*
         * Now we can only make sure we need MA_INODE, in mdd layer, will check
         * whether need MA_LOV and MA_COOKIE.
         */
        ma->ma_need = MA_INODE;
        ma->ma_valid = 0;
        mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
        rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
                        mdt_object_child(mc), lname, ma);
        if (rc == 0)
                mdt_handle_last_unlink(info, mc, ma);

        if (ma->ma_valid & MA_INODE) {
                switch (ma->ma_attr.la_mode & S_IFMT) {
                case S_IFDIR:
                        mdt_counter_incr(req->rq_export, LPROC_MDT_RMDIR);
                        break;
                case S_IFREG:
                case S_IFLNK:
                case S_IFCHR:
                case S_IFBLK:
                case S_IFIFO:
                case S_IFSOCK:
                        mdt_counter_incr(req->rq_export, LPROC_MDT_UNLINK);
                        break;
                default:
                        LASSERTF(0, "bad file type %o unlinking\n",
                                 ma->ma_attr.la_mode);
                }
        }

        EXIT;

        mdt_object_unlock_put(info, mc, child_lh, rc);
out_unlock_parent:
        mdt_object_unlock_put(info, mp, parent_lh, rc);
out:
        return rc;
}