-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
const struct lu_name *lname, struct lu_fid* fid,
int mask);
+static int mdd_declare_links_add(const struct lu_env *env,
+ struct mdd_object *mdd_obj,
+ struct thandle *handle);
static int mdd_links_add(const struct lu_env *env,
struct mdd_object *mdd_obj,
const struct lu_fid *pfid,
const struct lu_name *lname,
- struct thandle *handle);
+ struct thandle *handle, int first);
static int mdd_links_rename(const struct lu_env *env,
struct mdd_object *mdd_obj,
const struct lu_fid *oldpfid,
RETURN(-ENOTDIR);
iops = &obj->do_index_ops->dio_it;
- it = iops->init(env, obj, BYPASS_CAPA);
- if (it != NULL) {
+ it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
+ if (!IS_ERR(it)) {
result = iops->get(env, it, (const void *)"");
if (result > 0) {
int i;
iops->put(env, it);
iops->fini(env, it);
} else
- result = -ENOMEM;
+ result = PTR_ERR(it);
RETURN(result);
}
RETURN(rc);
}
-/**
- * If subdir count is up to ddp_max_nlink, then enable MNLINK_OBJ flag and
- * assign i_nlink to 1 which means the i_nlink for subdir count is incredible
- * (maybe too large to be represented). It is a trick to break through the
- * "i_nlink" limitation for subdir count.
- */
-void __mdd_ref_add(const struct lu_env *env, struct mdd_object *obj,
- struct thandle *handle)
-{
- struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
- struct mdd_device *m = mdd_obj2mdd_dev(obj);
-
- if (!mdd_is_mnlink(obj)) {
- if (S_ISDIR(mdd_object_type(obj))) {
- if (mdd_la_get(env, obj, tmp_la, BYPASS_CAPA))
- return;
-
- if (tmp_la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink) {
- obj->mod_flags |= MNLINK_OBJ;
- tmp_la->la_nlink = 1;
- tmp_la->la_valid = LA_NLINK;
- mdd_attr_set_internal(env, obj, tmp_la, handle,
- 0);
- return;
- }
- }
- mdo_ref_add(env, obj, handle);
- }
-}
-
-void __mdd_ref_del(const struct lu_env *env, struct mdd_object *obj,
- struct thandle *handle, int is_dot)
-{
- if (!mdd_is_mnlink(obj) || is_dot)
- mdo_ref_del(env, obj, handle);
-}
-
static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
const char *name, struct thandle *handle,
struct lustre_capa *capa)
rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
if (rc == 0 && is_dir) {
mdd_write_lock(env, pobj, MOR_TGT_PARENT);
- __mdd_ref_add(env, pobj, handle);
+ mdo_ref_add(env, pobj, handle);
mdd_write_unlock(env, pobj);
}
RETURN(rc);
rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
if (rc == 0 && is_dir) {
- int is_dot = 0;
-
- if (name != NULL && name[0] == '.' && name[1] == 0)
- is_dot = 1;
mdd_write_lock(env, pobj, MOR_TGT_PARENT);
- __mdd_ref_del(env, pobj, handle, is_dot);
+ mdo_ref_del(env, pobj, handle);
mdd_write_unlock(env, pobj);
}
RETURN(rc);
}
+int mdd_declare_llog_record(const struct lu_env *env, struct mdd_device *mdd,
+ int reclen, struct thandle *handle)
+{
+ int rc;
+
+ /* XXX: this is a temporary solution to declare llog changes
+ * will be fixed in 2.3 with new llog implementation */
+
+ LASSERT(mdd->mdd_capa);
+
+ /* XXX: Since we use the 'mdd_capa' as fake llog object here, we
+ * have to set the parameter 'size' as INT_MAX or 0 to inform
+ * OSD that this record write is for a llog write or catalog
+ * header update, and osd declare function will reserve less
+ * credits for optimization purpose.
+ *
+ * Reserve 6 blocks for a llog write, since the llog file is
+ * usually small, reserve 2 blocks for catalog header update,
+ * because we know for sure that catalog header is already
+ * allocated.
+ *
+ * This hack should be removed in 2.3.
+ */
+
+ /* record itself */
+ rc = dt_declare_record_write(env, mdd->mdd_capa,
+ DECLARE_LLOG_WRITE, 0, handle);
+ if (rc)
+ return rc;
+
+ /* header will be updated as well */
+ rc = dt_declare_record_write(env, mdd->mdd_capa,
+ DECLARE_LLOG_WRITE, 0, handle);
+ if (rc)
+ return rc;
+
+ /* also we should be able to create new plain log */
+ rc = dt_declare_create(env, mdd->mdd_capa, NULL, NULL, NULL, handle);
+ if (rc)
+ return rc;
+
+ /* new record referencing new plain llog */
+ rc = dt_declare_record_write(env, mdd->mdd_capa,
+ DECLARE_LLOG_WRITE, 0, handle);
+ if (rc)
+ return rc;
+
+ /* catalog's header will be updated as well */
+ rc = dt_declare_record_write(env, mdd->mdd_capa,
+ DECLARE_LLOG_REWRITE, 0, handle);
+
+ return rc;
+}
+
+int mdd_declare_changelog_store(const struct lu_env *env,
+ struct mdd_device *mdd,
+ const struct lu_name *fname,
+ struct thandle *handle)
+{
+ int reclen;
+
+ /* Not recording */
+ if (!(mdd->mdd_cl.mc_flags & CLM_ON))
+ return 0;
+
+ /* we'll be writing payload + llog header */
+ reclen = sizeof(struct llog_changelog_rec);
+ if (fname)
+ reclen += llog_data_len(fname->ln_namelen);
+
+ return mdd_declare_llog_record(env, mdd, reclen, handle);
+}
/** Store a namespace change changelog record
* If this fails, we must fail the whole transaction; we don't
static int mdd_changelog_ns_store(const struct lu_env *env,
struct mdd_device *mdd,
enum changelog_rec_type type,
+ int flags,
struct mdd_object *target,
struct mdd_object *parent,
const struct lu_fid *tf,
int rc;
ENTRY;
+ /* Not recording */
if (!(mdd->mdd_cl.mc_flags & CLM_ON))
RETURN(0);
+ if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
+ RETURN(0);
LASSERT(parent != NULL);
LASSERT(tname != NULL);
RETURN(-ENOMEM);
rec = (struct llog_changelog_rec *)buf->lb_buf;
- rec->cr.cr_flags = CLF_VERSION;
+ rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
rec->cr.cr_type = (__u32)type;
tfid = tf ? tf : mdo2fid(target);
rec->cr.cr_tfid = *tfid;
return 0;
}
+static int mdd_declare_link(const struct lu_env *env,
+ struct mdd_device *mdd,
+ struct mdd_object *p,
+ struct mdd_object *c,
+ const struct lu_name *name,
+ struct thandle *handle)
+{
+ int rc;
+
+ rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
+ if (rc)
+ return rc;
+
+ rc = mdo_declare_ref_add(env, c, handle);
+ if (rc)
+ return rc;
+
+ rc = mdo_declare_attr_set(env, p, NULL, handle);
+ if (rc)
+ return rc;
+
+ rc = mdo_declare_attr_set(env, c, NULL, handle);
+ if (rc)
+ return rc;
+
+ rc = mdd_declare_links_add(env, c, handle);
+ if (rc)
+ return rc;
+
+ rc = mdd_declare_changelog_store(env, mdd, name, handle);
+
+ return rc;
+}
+
static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
struct md_object *src_obj, const struct lu_name *lname,
struct md_attr *ma)
struct thandle *handle;
#ifdef HAVE_QUOTA_SUPPORT
struct obd_device *obd = mdd->mdd_obd_dev;
+ struct obd_export *exp = md_quota(env)->mq_exp;
struct mds_obd *mds = &obd->u.mds;
unsigned int qids[MAXQUOTAS] = { 0, 0 };
int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
quota_opc = FSFILT_OP_LINK;
mdd_quota_wrapper(la_tmp, qids);
/* get block quota for parent */
- lquota_chkquota(mds_quota_interface_ref, obd,
+ lquota_chkquota(mds_quota_interface_ref, obd, exp,
qids, rec_pending, 1, NULL,
LQUOTA_FLAGS_BLK, data, 1);
}
}
#endif
- mdd_txn_param_build(env, mdd, MDD_TXN_LINK_OP);
- handle = mdd_trans_start(env, mdd);
+ handle = mdd_trans_create(env, mdd);
if (IS_ERR(handle))
GOTO(out_pending, rc = PTR_ERR(handle));
+ rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = mdd_trans_start(env, mdd, handle);
+ if (rc)
+ GOTO(stop, rc);
+
dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
if (dlh == NULL)
GOTO(out_trans, rc = -ENOMEM);
if (rc)
GOTO(out_unlock, rc);
- __mdd_ref_add(env, mdd_sobj, handle);
+ mdo_ref_add(env, mdd_sobj, handle);
LASSERT(ma->ma_attr.la_valid & LA_CTIME);
la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
la->la_valid = LA_CTIME;
rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
- if (rc == 0)
- mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj), lname, handle);
+ if (rc == 0) {
+ mdd_links_add(env, mdd_sobj,
+ mdo2fid(mdd_tobj), lname, handle, 0);
+ }
EXIT;
out_unlock:
mdd_pdo_write_unlock(env, mdd_tobj, dlh);
out_trans:
if (rc == 0)
- rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, mdd_sobj,
+ rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
mdd_tobj, NULL, lname, handle);
+stop:
mdd_trans_stop(env, mdd, rc, handle);
out_pending:
#ifdef HAVE_QUOTA_SUPPORT
return rc;
}
+int mdd_declare_finish_unlink(const struct lu_env *env,
+ struct mdd_object *obj,
+ struct md_attr *ma,
+ struct thandle *handle)
+{
+ int rc;
+
+ rc = orph_declare_index_insert(env, obj, handle);
+ if (rc)
+ return rc;
+
+ return mdd_declare_object_kill(env, obj, ma, handle);
+}
+
/* caller should take a lock before calling */
int mdd_finish_unlink(const struct lu_env *env,
struct mdd_object *obj, struct md_attr *ma,
{
int rc;
int reset = 1;
+ int is_dir = S_ISDIR(ma->ma_attr.la_mode);
ENTRY;
LASSERT(mdd_write_locked(env, obj) != 0);
- rc = mdd_iattr_get(env, obj, ma);
- if (rc == 0 && ma->ma_attr.la_nlink == 0) {
+ /* read HSM flags, needed to set changelogs flags */
+ ma->ma_need = MA_HSM | MA_INODE;
+ rc = mdd_attr_get_internal(env, obj, ma);
+ if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) {
obj->mod_flags |= DEAD_OBJ;
/* add new orphan and the object
* will be deleted during mdd_close() */
PFID(mdd_object_fid(obj)),
obj->mod_count);
} else {
- rc = mdd_object_kill(env, obj, ma);
+ rc = mdd_object_kill(env, obj, ma, th);
if (rc == 0)
reset = 0;
}
+ /* get the i_nlink */
+ ma->ma_need = MA_INODE;
+ rc = mdd_attr_get_internal(env, obj, ma);
}
if (reset)
ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
RETURN(rc);
}
+static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
+ struct mdd_object *p, struct mdd_object *c,
+ const struct lu_name *name, struct md_attr *ma,
+ struct thandle *handle)
+{
+ int rc;
+
+ rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
+ if (rc)
+ return rc;
+
+ rc = mdo_declare_ref_del(env, p, handle);
+ if (rc)
+ return rc;
+
+ rc = mdo_declare_ref_del(env, c, handle);
+ if (rc)
+ return rc;
+
+ rc = mdo_declare_ref_del(env, c, handle);
+ if (rc)
+ return rc;
+
+ rc = mdo_declare_attr_set(env, p, NULL, handle);
+ if (rc)
+ return rc;
+
+ rc = mdo_declare_attr_set(env, c, NULL, handle);
+ if (rc)
+ return rc;
+
+ rc = mdd_declare_finish_unlink(env, c, ma, handle);
+ if (rc)
+ return rc;
+
+ rc = mdd_declare_links_add(env, c, handle);
+ if (rc)
+ return rc;
+
+ rc = mdd_declare_changelog_store(env, mdd, name, handle);
+
+ return rc;
+}
+
static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
struct md_object *cobj, const struct lu_name *lname,
struct md_attr *ma)
int rc;
ENTRY;
- LASSERTF(mdd_object_exists(mdd_cobj) > 0, "FID is "DFID"\n",
- PFID(mdd_object_fid(mdd_cobj)));
-
- rc = mdd_log_txn_param_build(env, cobj, ma, MDD_TXN_UNLINK_OP);
- if (rc)
- RETURN(rc);
+ if (mdd_object_exists(mdd_cobj) <= 0)
+ RETURN(-ENOENT);
- handle = mdd_trans_start(env, mdd);
+ handle = mdd_trans_create(env, mdd);
if (IS_ERR(handle))
RETURN(PTR_ERR(handle));
+ rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
+ lname, ma, handle);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = mdd_trans_start(env, mdd, handle);
+ if (rc)
+ GOTO(stop, rc);
+
dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
if (dlh == NULL)
GOTO(out_trans, rc = -ENOMEM);
if (rc)
GOTO(cleanup, rc);
- __mdd_ref_del(env, mdd_cobj, handle, 0);
+ mdo_ref_del(env, mdd_cobj, handle);
if (is_dir)
/* unlink dot */
- __mdd_ref_del(env, mdd_cobj, handle, 1);
+ mdo_ref_del(env, mdd_cobj, handle);
LASSERT(ma->ma_attr.la_valid & LA_CTIME);
la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
if (rc)
GOTO(cleanup, rc);
- la->la_valid = LA_CTIME;
- rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
- if (rc)
- GOTO(cleanup, rc);
+ if (ma->ma_attr.la_nlink > 0 || mdd_cobj->mod_count > 0) {
+ /* update ctime of an unlinked file only if it is still
+ * opened or a link still exists */
+ la->la_valid = LA_CTIME;
+ rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
+ if (rc)
+ GOTO(cleanup, rc);
+ }
rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
#ifdef HAVE_QUOTA_SUPPORT
mdd_write_unlock(env, mdd_cobj);
mdd_pdo_write_unlock(env, mdd_pobj, dlh);
out_trans:
- if (rc == 0)
+ if (rc == 0) {
+ int cl_flags;
+
+ cl_flags = (ma->ma_attr.la_nlink == 0) ? CLF_UNLINK_LAST : 0;
+ if ((ma->ma_valid & MA_HSM) &&
+ (ma->ma_hsm.mh_flags & HS_EXISTS))
+ cl_flags |= CLF_UNLINK_HSM_EXISTS;
+
rc = mdd_changelog_ns_store(env, mdd,
- is_dir ? CL_RMDIR : CL_UNLINK,
- mdd_cobj, mdd_pobj, NULL, lname,
- handle);
+ is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
+ mdd_cobj, mdd_pobj, NULL, lname, handle);
+ }
+stop:
mdd_trans_stop(env, mdd, rc, handle);
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,3,50,0)
+ if (rc == 0 && ma->ma_valid & MA_COOKIE && ma->ma_valid & MA_LOV &&
+ ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_UNLINK_DESTROY)
+ /* Since echo client is incapable of destorying ost object,
+ * it will destory the object here. */
+ rc = mdd_lovobj_unlink(env, mdd, mdd_cobj, la,
+ ma->ma_lmm, ma->ma_lmm_size,
+ ma->ma_cookie, 1);
+#else
+#warning "please remove this after 2.4 (LOD/OSP)."
+#endif
+
#ifdef HAVE_QUOTA_SUPPORT
if (quota_opc)
/* Trigger dqrel on the owner of child and parent. If failed,
#ifdef HAVE_QUOTA_SUPPORT
struct md_ucred *uc = md_ucred(env);
struct obd_device *obd = mdd->mdd_obd_dev;
+ struct obd_export *exp = md_quota(env)->mq_exp;
struct mds_obd *mds = &obd->u.mds;
unsigned int qids[MAXQUOTAS] = { 0, 0 };
int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
int rc;
ENTRY;
+ /* XXX: this code won't be used ever:
+ * DNE uses slightly different approach */
+ LBUG();
+
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota) {
if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) {
mdd_quota_wrapper(la_tmp, qids);
/* get block quota for parent */
lquota_chkquota(mds_quota_interface_ref, obd,
- qids, rec_pending, 1, NULL,
+ exp, qids, rec_pending, 1, NULL,
LQUOTA_FLAGS_BLK, data, 1);
}
} else {
}
}
#endif
- mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_INSERT_OP);
- handle = mdd_trans_start(env, mdo2mdd(pobj));
+ handle = mdd_trans_create(env, mdo2mdd(pobj));
if (IS_ERR(handle))
GOTO(out_pending, rc = PTR_ERR(handle));
+ rc = mdd_trans_start(env, mdo2mdd(pobj), handle);
+
dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
if (dlh == NULL)
GOTO(out_trans, rc = -ENOMEM);
int rc;
ENTRY;
+ /* XXX: this code won't be used ever:
+ * DNE uses slightly different approach */
+ LBUG();
+
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota) {
struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
}
}
#endif
- mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_DELETE_OP);
- handle = mdd_trans_start(env, mdd);
+ handle = mdd_trans_create(env, mdd);
if (IS_ERR(handle))
GOTO(out_pending, rc = PTR_ERR(handle));
+ rc = mdd_trans_start(env, mdd, handle);
+
dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
if (dlh == NULL)
GOTO(out_trans, rc = -ENOMEM);
struct thandle *handle;
#ifdef HAVE_QUOTA_SUPPORT
struct obd_device *obd = mdd->mdd_obd_dev;
+ struct obd_export *exp = md_quota(env)->mq_exp;
struct mds_obd *mds = &obd->u.mds;
unsigned int qcids[MAXQUOTAS] = { 0, 0 };
unsigned int qpids[MAXQUOTAS] = { 0, 0 };
int rc;
ENTRY;
+ /* XXX: this code won't be used ever:
+ * DNE uses slightly different approach */
+ LBUG();
+
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota && !tobj) {
struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
quota_popc = FSFILT_OP_LINK;
mdd_quota_wrapper(la_tmp, qpids);
/* get block quota for target parent */
- lquota_chkquota(mds_quota_interface_ref, obd,
+ lquota_chkquota(mds_quota_interface_ref, obd, exp,
qpids, rec_pending, 1, NULL,
LQUOTA_FLAGS_BLK, data, 1);
}
}
#endif
- mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_TGT_OP);
- handle = mdd_trans_start(env, mdd);
+ handle = mdd_trans_create(env, mdd);
if (IS_ERR(handle))
GOTO(out_pending, rc = PTR_ERR(handle));
+ rc = mdd_trans_start(env, mdd, handle);
+
dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT);
if (dlh == NULL)
GOTO(out_trans, rc = -ENOMEM);
* it must be local one.
*/
if (tobj && mdd_object_exists(mdd_tobj)) {
- __mdd_ref_del(env, mdd_tobj, handle, 0);
+ mdo_ref_del(env, mdd_tobj, handle);
/* Remove dot reference. */
if (S_ISDIR(ma->ma_attr.la_mode))
- __mdd_ref_del(env, mdd_tobj, handle, 1);
+ mdo_ref_del(env, mdd_tobj, handle);
la->la_valid = LA_CTIME;
rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
if (rc == 0)
/* Bare EXT record with no RENAME in front of it signifies
a partial slave op */
- rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
+ rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj,
mdd_tpobj, NULL, lname, handle);
mdd_trans_stop(env, mdd, rc, handle);
}
+static int mdd_declare_create_data(const struct lu_env *env,
+ struct mdd_device *mdd,
+ struct mdd_object *obj,
+ int lmm_size,
+ struct thandle *handle)
+{
+ struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
+ int rc;
+
+ buf->lb_buf = NULL;
+ buf->lb_len = lmm_size;
+ rc = mdo_declare_xattr_set(env, obj, buf, XATTR_NAME_LOV,
+ 0, handle);
+ if (rc)
+ return rc;
+
+ rc = mdd_declare_lov_objid_update(env, mdd, handle);
+
+ return rc;
+}
+
static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
struct md_object *cobj, const struct md_op_spec *spec,
struct md_attr *ma)
struct mdd_device *mdd = mdo2mdd(cobj);
struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
struct mdd_object *son = md2mdd_obj(cobj);
- struct lu_attr *attr = &ma->ma_attr;
struct lov_mds_md *lmm = NULL;
int lmm_size = 0;
struct thandle *handle;
if (!md_should_create(spec->sp_cr_flags))
RETURN(0);
lmm_size = ma->ma_lmm_size;
- rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
- spec, attr);
+
+ rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size, spec, ma);
if (rc)
RETURN(rc);
- mdd_txn_param_build(env, mdd, MDD_TXN_CREATE_DATA_OP);
- handle = mdd_trans_start(env, mdd);
+ handle = mdd_trans_create(env, mdd);
if (IS_ERR(handle))
GOTO(out_free, rc = PTR_ERR(handle));
+ rc = mdd_declare_create_data(env, mdd, son, lmm_size, handle);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = mdd_trans_start(env, mdd, handle);
+ if (rc)
+ GOTO(stop, rc);
+
/*
* XXX: Setting the lov ea is not locked but setting the attr is locked?
* Should this be fixed?
if (rc == 0)
mdd_lov_objid_update(mdd, lmm);
+stop:
mdd_trans_stop(env, mdd, rc, handle);
out_free:
/* Finish mdd_lov_create() stuff. */
+ /* if no_create == 0 (not replay), we free lmm allocated by
+ * mdd_lov_create() */
mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
RETURN(rc);
}
else if (unlikely(rc < 0)) {
CERROR("Object "DFID" locates on remote server\n",
PFID(mdo2fid(mdd_obj)));
- LBUG();
+ RETURN(-EINVAL);
}
/* The common filename length check. */
RETURN(rc);
}
+int mdd_declare_object_initialize(const struct lu_env *env,
+ struct mdd_object *child,
+ struct md_attr *ma,
+ struct thandle *handle)
+{
+ int rc;
+
+ rc = mdo_declare_attr_set(env, child, &ma->ma_attr, handle);
+ if (rc == 0 && S_ISDIR(ma->ma_attr.la_mode)) {
+ rc = mdo_declare_index_insert(env, child, mdo2fid(child),
+ dot, handle);
+ if (rc == 0)
+ rc = mdo_declare_ref_add(env, child, handle);
+ }
+ if (rc == 0)
+ mdd_declare_links_add(env, child, handle);
+
+ return rc;
+}
+
int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
const struct lu_name *lname, struct mdd_object *child,
struct md_attr *ma, struct thandle *handle,
if (S_ISDIR(ma->ma_attr.la_mode)) {
/* Add "." and ".." for newly created dir */
- __mdd_ref_add(env, child, handle);
+ mdo_ref_add(env, child, handle);
rc = __mdd_index_insert_only(env, child, mdo2fid(child),
dot, handle, BYPASS_CAPA);
if (rc == 0)
dotdot, handle,
BYPASS_CAPA);
if (rc != 0)
- __mdd_ref_del(env, child, handle, 1);
+ mdo_ref_del(env, child, handle);
}
if (rc == 0)
- mdd_links_add(env, child, pfid, lname, handle);
+ mdd_links_add(env, child, pfid, lname, handle, 1);
RETURN(rc);
}
RETURN(rc);
}
+static int mdd_declare_create(const struct lu_env *env,
+ struct mdd_device *mdd,
+ struct mdd_object *p,
+ struct mdd_object *c,
+ const struct lu_name *name,
+ struct md_attr *ma,
+ int lmm_size,
+ struct thandle *handle,
+ const struct md_op_spec *spec)
+{
+ struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
+ int rc = 0;
+
+ rc = mdd_declare_object_create_internal(env, p, c, ma, handle, spec);
+ if (rc)
+ GOTO(out, rc);
+
+ /* if dir, then can inherit default ACl */
+ buf->lb_buf = NULL;
+ buf->lb_len = lmm_size;
+ if (S_ISDIR(ma->ma_attr.la_mode)) {
+ rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_ACL_DEFAULT,
+ 0, handle);
+ if (rc == 0)
+ rc = mdo_declare_ref_add(env, p, handle);
+ }
+ if (rc)
+ GOTO(out, rc);
+
+ rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_ACL_ACCESS,
+ 0, handle);
+ if (rc)
+ GOTO(out, rc);
+
+ rc = mdd_declare_object_initialize(env, c, ma, handle);
+ if (rc)
+ GOTO(out, rc);
+
+ rc = mdo_declare_index_insert(env, p, mdo2fid(c),
+ name->ln_name, handle);
+ if (rc)
+ GOTO(out, rc);
+
+ rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
+ 0, handle);
+ if (rc)
+ GOTO(out, rc);
+
+ if (S_ISLNK(ma->ma_attr.la_mode)) {
+ rc = dt_declare_record_write(env, mdd_object_child(c),
+ strlen(spec->u.sp_symname), 0,
+ handle);
+ if (rc)
+ GOTO(out, rc);
+ }
+ rc = mdo_declare_attr_set(env, p, &ma->ma_attr, handle);
+ if (rc)
+ return rc;
+
+ rc = mdd_declare_changelog_store(env, mdd, name, handle);
+ if (rc)
+ return rc;
+
+ rc = mdd_declare_lov_objid_update(env, mdd, handle);
+
+out:
+ return rc;
+}
+
+
/*
* Create object and insert it into namespace.
*/
int got_def_acl = 0;
#ifdef HAVE_QUOTA_SUPPORT
struct obd_device *obd = mdd->mdd_obd_dev;
+ struct obd_export *exp = md_quota(env)->mq_exp;
struct mds_obd *mds = &obd->u.mds;
unsigned int qcids[MAXQUOTAS] = { 0, 0 };
unsigned int qpids[MAXQUOTAS] = { 0, 0 };
mdd_quota_wrapper(&ma->ma_attr, qcids);
mdd_quota_wrapper(la_tmp, qpids);
/* get file quota for child */
- lquota_chkquota(mds_quota_interface_ref, obd, qcids,
- inode_pending, 1, NULL, 0, NULL, 0);
+ lquota_chkquota(mds_quota_interface_ref, obd, exp,
+ qcids, inode_pending, 1, NULL, 0, NULL,
+ 0);
switch (ma->ma_attr.la_mode & S_IFMT) {
case S_IFLNK:
case S_IFDIR:
/* get block quota for child and parent */
if (block_count)
lquota_chkquota(mds_quota_interface_ref, obd,
- qcids, block_pending,
+ exp, qcids, block_pending,
block_count, NULL,
LQUOTA_FLAGS_BLK, NULL, 0);
if (!same)
lquota_chkquota(mds_quota_interface_ref, obd,
- qpids, parent_pending, 1, NULL,
- LQUOTA_FLAGS_BLK, NULL, 0);
+ exp, qpids, parent_pending, 1,
+ NULL, LQUOTA_FLAGS_BLK, NULL,
+ 0);
}
}
#endif
+ if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
+ GOTO(out_pending, rc = -EINPROGRESS);
+
/*
* No RPC inside the transaction, so OST objects should be created at
* first.
if (S_ISREG(attr->la_mode)) {
lmm_size = ma->ma_lmm_size;
rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
- spec, attr);
+ spec, ma);
if (rc)
GOTO(out_pending, rc);
}
got_def_acl = 1;
}
- mdd_txn_param_build(env, mdd, MDD_TXN_MKDIR_OP);
- handle = mdd_trans_start(env, mdd);
+ handle = mdd_trans_create(env, mdd);
if (IS_ERR(handle))
GOTO(out_free, rc = PTR_ERR(handle));
+ rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, ma,
+ lmm_size, handle, spec);
+ if (rc)
+ GOTO(out_stop, rc);
+
+ rc = mdd_trans_start(env, mdd, handle);
+ if (rc)
+ GOTO(out_stop, rc);
+
dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
if (dlh == NULL)
GOTO(out_trans, rc = -ENOMEM);
}
if (lmm && lmm_size > 0) {
/* Set Lov here, do not get lmm again later */
+ if (lmm_size > ma->ma_lmm_size) {
+ /* Reply buffer is smaller, need bigger one */
+ mdd_max_lmm_buffer(env, lmm_size);
+ if (unlikely(info->mti_max_lmm == NULL))
+ GOTO(cleanup, rc = -ENOMEM);
+ ma->ma_lmm = info->mti_max_lmm;
+ ma->ma_big_lmm_used = 1;
+ }
memcpy(ma->ma_lmm, lmm, lmm_size);
ma->ma_lmm_size = lmm_size;
ma->ma_valid |= MA_LOV;
if (rc2 == 0) {
mdd_write_lock(env, son, MOR_TGT_CHILD);
- __mdd_ref_del(env, son, handle, 0);
+ mdo_ref_del(env, son, handle);
if (initialized && S_ISDIR(attr->la_mode))
- __mdd_ref_del(env, son, handle, 1);
+ mdo_ref_del(env, son, handle);
mdd_write_unlock(env, son);
}
}
S_ISDIR(attr->la_mode) ? CL_MKDIR :
S_ISREG(attr->la_mode) ? CL_CREATE :
S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
- son, mdd_pobj, NULL, lname, handle);
+ 0, son, mdd_pobj, NULL, lname, handle);
+out_stop:
mdd_trans_stop(env, mdd, rc, handle);
out_free:
- /* finis lov_create stuff, free all temporary data */
+ /* finish lov_create stuff, free all temporary data */
mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
out_pending:
#ifdef HAVE_QUOTA_SUPPORT
quota_opc);
}
#endif
+ /* The child object shouldn't be cached anymore */
+ if (rc)
+ cfs_set_bit(LU_OBJECT_HEARD_BANSHEE,
+ &child->mo_lu.lo_header->loh_flags);
return rc;
}
RETURN(rc);
}
+static int mdd_declare_rename(const struct lu_env *env,
+ struct mdd_device *mdd,
+ struct mdd_object *mdd_spobj,
+ struct mdd_object *mdd_tpobj,
+ struct mdd_object *mdd_sobj,
+ struct mdd_object *mdd_tobj,
+ const struct lu_name *sname,
+ const struct lu_name *tname,
+ struct md_attr *ma,
+ struct thandle *handle)
+{
+ int rc;
+
+ LASSERT(mdd_spobj);
+ LASSERT(mdd_tpobj);
+ LASSERT(mdd_sobj);
+
+ /* name from source dir */
+ rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
+ if (rc)
+ return rc;
+
+ /* .. from source child */
+ if (S_ISDIR(mdd_object_type(mdd_sobj))) {
+ /* source child can be directory,
+ * counted by source dir's nlink */
+ rc = mdo_declare_ref_del(env, mdd_spobj, handle);
+ if (rc)
+ return rc;
+
+ rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle);
+ if (rc)
+ return rc;
+
+ rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj),
+ dotdot, handle);
+ if (rc)
+ return rc;
+
+ /* new target child can be directory,
+ * counted by target dir's nlink */
+ rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
+ if (rc)
+ return rc;
+
+ }
+
+ rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle);
+ if (rc)
+ return rc;
+
+ rc = mdo_declare_attr_set(env, mdd_sobj, NULL, handle);
+ if (rc)
+ return rc;
+ mdd_declare_links_add(env, mdd_sobj, handle);
+ if (rc)
+ return rc;
+
+ rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle);
+ if (rc)
+ return rc;
+
+ /* new name */
+ rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
+ tname->ln_name, handle);
+ if (rc)
+ return rc;
+
+ /* name from target dir (old name), we declare it unconditionally
+ * as mdd_rename() calls delete unconditionally as well. so just
+ * to balance declarations vs calls to change ... */
+ rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
+ if (rc)
+ return rc;
+
+ if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
+ /* delete target child in target parent directory */
+ rc = mdo_declare_ref_del(env, mdd_tobj, handle);
+ if (rc)
+ return rc;
+
+ if (S_ISDIR(mdd_object_type(mdd_tobj))) {
+ /* target child can be directory,
+ * delete "." reference in target child directory */
+ rc = mdo_declare_ref_del(env, mdd_tobj, handle);
+ if (rc)
+ return rc;
+
+ /* delete ".." reference in target parent directory */
+ rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
+ if (rc)
+ return rc;
+ }
+
+ rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle);
+ if (rc)
+ return rc;
+
+ mdd_declare_links_add(env, mdd_tobj, handle);
+ if (rc)
+ return rc;
+
+ rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
+ if (rc)
+ return rc;
+ }
+
+ rc = mdd_declare_changelog_store(env, mdd, tname, handle);
+ if (rc)
+ return rc;
+
+ rc = mdd_declare_changelog_store(env, mdd, sname, handle);
+ if (rc)
+ return rc;
+
+ return rc;
+}
+
/* src object can be remote that is why we use only fid and type of object */
static int mdd_rename(const struct lu_env *env,
struct md_object *src_pobj, struct md_object *tgt_pobj,
#ifdef HAVE_QUOTA_SUPPORT
struct obd_device *obd = mdd->mdd_obd_dev;
+ struct obd_export *exp = md_quota(env)->mq_exp;
struct mds_obd *mds = &obd->u.mds;
unsigned int qspids[MAXQUOTAS] = { 0, 0 };
unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
mdd_quota_wrapper(la_tmp, qtpids);
/* get block quota for target parent */
lquota_chkquota(mds_quota_interface_ref,
- obd, qtpids,
+ obd, exp, qtpids,
rec_pending, 1, NULL,
LQUOTA_FLAGS_BLK,
data, 1);
}
}
#endif
- mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_OP);
- handle = mdd_trans_start(env, mdd);
+ mdd_sobj = mdd_object_find(env, mdd, lf);
+
+ handle = mdd_trans_create(env, mdd);
if (IS_ERR(handle))
GOTO(out_pending, rc = PTR_ERR(handle));
+ rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
+ mdd_tobj, lsname, ltname, ma, handle);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = mdd_trans_start(env, mdd, handle);
+ if (rc)
+ GOTO(stop, rc);
+
/* FIXME: Should consider tobj and sobj too in rename_lock. */
rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
if (rc < 0)
if (sdlh == NULL || tdlh == NULL)
GOTO(cleanup, rc = -ENOMEM);
- mdd_sobj = mdd_object_find(env, mdd, lf);
rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj,
mdd_sobj, mdd_tobj, ma);
if (rc)
rc = -EINVAL;
goto cleanup;
}
- __mdd_ref_del(env, mdd_tobj, handle, 0);
+ mdo_ref_del(env, mdd_tobj, handle);
/* Remove dot reference. */
if (is_dir)
- __mdd_ref_del(env, mdd_tobj, handle, 1);
+ mdo_ref_del(env, mdd_tobj, handle);
la->la_valid = LA_CTIME;
rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
if (rc == -ENOENT)
/* Old files might not have EA entry */
mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
- lsname, handle);
+ lsname, handle, 0);
mdd_write_unlock(env, mdd_sobj);
/* We don't fail the transaction if the link ea can't be
updated -- fid2path will use alternate lookup method. */
mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
cleanup_unlocked:
if (rc == 0)
- rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, mdd_tobj,
+ rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, 0, mdd_tobj,
mdd_spobj, lf, lsname, handle);
- if (rc == 0)
- rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
- mdd_tpobj, lf, ltname, handle);
+ if (rc == 0) {
+ struct lu_fid zero_fid;
+ fid_zero(&zero_fid);
+ /* If the rename target exist, The CL_EXT record should save
+ * the target fid as tfid, otherwise, use zero fid. LU-543 */
+ rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj,
+ mdd_tpobj,
+ mdd_tobj ? NULL : &zero_fid,
+ ltname, handle);
+ }
+stop:
mdd_trans_stop(env, mdd, rc, handle);
if (mdd_sobj)
mdd_object_put(env, mdd_sobj);
rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
if (rc == -ERANGE) {
/* Buf was too small, figure out what we need. */
- buf->lb_buf = NULL;
- buf->lb_len = 0;
+ mdd_buf_put(buf);
rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
if (rc < 0)
return ERR_PTR(rc);
static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
const struct lu_fid *pfid)
{
- int reclen;
+ struct lu_fid tmpfid;
+ int reclen;
- fid_cpu_to_be(&lee->lee_parent_fid, pfid);
- strncpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
+ fid_cpu_to_be(&tmpfid, pfid);
+ memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
+ memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
lee->lee_reclen[0] = (reclen >> 8) & 0xff;
struct lu_name *lname, struct lu_fid *pfid)
{
*reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
- fid_be_to_cpu(pfid, &lee->lee_parent_fid);
+ memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
+ fid_be_to_cpu(pfid, pfid);
lname->ln_name = lee->lee_name;
lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
}
return 0;
}
+static int mdd_declare_links_add(const struct lu_env *env,
+ struct mdd_object *mdd_obj,
+ struct thandle *handle)
+{
+ int rc;
+
+ /* XXX: max size? */
+ rc = mdo_declare_xattr_set(env, mdd_obj,
+ mdd_buf_get_const(env, NULL, 4096),
+ XATTR_NAME_LINK, 0, handle);
+
+ return rc;
+}
+
/* For pathologic linkers, we don't want to spend lots of time scanning the
* link ea. Limit ourseleves to something reasonable; links not in the EA
* can be looked up via (slower) parent lookup.
struct mdd_object *mdd_obj,
const struct lu_fid *pfid,
const struct lu_name *lname,
- struct thandle *handle)
+ struct thandle *handle, int first)
{
struct lu_buf *buf;
struct link_ea_header *leh;
if (!mdd_linkea_enable)
RETURN(0);
- buf = mdd_links_get(env, mdd_obj);
+ buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj);
if (IS_ERR(buf)) {
rc = PTR_ERR(buf);
if (rc != -ENODATA) {
rc = __mdd_xattr_set(env, mdd_obj,
mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
XATTR_NAME_LINK, 0, handle);
- if (rc)
- CERROR("link_ea add failed %d "DFID"\n", rc,
- PFID(mdd_object_fid(mdd_obj)));
+ if (rc) {
+ if (rc == -ENOSPC)
+ CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
+ PFID(mdd_object_fid(mdd_obj)));
+ else
+ CERROR("link_ea add failed %d "DFID"\n", rc,
+ PFID(mdd_object_fid(mdd_obj)));
+ }
- if (buf->lb_vmalloc)
+ if (buf->lb_len > OBD_ALLOC_BIG)
/* if we vmalloced a large buffer drop it */
mdd_buf_put(buf);
buf = mdd_links_get(env, mdd_obj);
if (IS_ERR(buf)) {
rc = PTR_ERR(buf);
- CERROR("link_ea read failed %d "DFID"\n",
- rc, PFID(mdd_object_fid(mdd_obj)));
+ if (rc == -ENODATA)
+ CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
+ rc, PFID(mdd_object_fid(mdd_obj)));
+ else
+ CERROR("link_ea read failed %d "DFID"\n",
+ rc, PFID(mdd_object_fid(mdd_obj)));
RETURN(rc);
}
leh = buf->lb_buf;
oldlname->ln_namelen, oldlname->ln_name, rc,
PFID(mdd_object_fid(mdd_obj)));
- if (buf->lb_vmalloc)
+ if (buf->lb_len > OBD_ALLOC_BIG)
/* if we vmalloced a large buffer drop it */
mdd_buf_put(buf);
.mdo_rename = mdd_rename,
.mdo_link = mdd_link,
.mdo_unlink = mdd_unlink,
+ .mdo_lum_lmm_cmp = mdd_lum_lmm_cmp,
.mdo_name_insert = mdd_name_insert,
.mdo_name_remove = mdd_name_remove,
.mdo_rename_tgt = mdd_rename_tgt,
- .mdo_create_data = mdd_create_data
+ .mdo_create_data = mdd_create_data,
};
git://git.whamcloud.com / fs / lustre-release.git / blobdiff