* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
*
* Lustre Metadata Server (mdd) routines
*
- * Author: Wang Di <wangdi@clusterfs.com>
+ * Author: Wang Di <wangdi@intel.com>
*/
#define DEBUG_SUBSYSTEM S_MDS
sizeof(dotdot) - 1
};
-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 inline 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, int first);
-static inline int mdd_links_del(const struct lu_env *env,
- struct mdd_object *mdd_obj,
- const struct lu_fid *pfid,
- const struct lu_name *lname,
- struct thandle *handle);
-static int mdd_links_rename(const struct lu_env *env,
- struct mdd_object *mdd_obj,
- const struct lu_fid *oldpfid,
- const struct lu_name *oldlname,
- const struct lu_fid *newpfid,
- const struct lu_name *newlname,
- struct thandle *handle,
- int first, int check);
+static int __mdd_lookup(const struct lu_env *, struct md_object *,
+ const struct lu_name *, struct lu_fid*, int);
static int
__mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj,
RETURN(rc);
}
-static int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj,
- struct lu_fid *fid)
+int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj,
+ struct lu_fid *fid)
{
return __mdd_lookup_locked(env, &obj->mod_obj, &lname_dotdot, fid, 0);
}
GOTO(out, rc = 1);
if (parent)
mdd_object_put(env, parent);
- parent = mdd_object_find(env, mdd, pfid);
-
- /* cross-ref parent */
- if (parent == NULL) {
- if (pf != NULL)
- *pf = *pfid;
- GOTO(out, rc = -EREMOTE);
- } else if (IS_ERR(parent))
- GOTO(out, rc = PTR_ERR(parent));
- p1 = parent;
+
+ parent = mdd_object_find(env, mdd, pfid);
+ if (IS_ERR(parent)) {
+ GOTO(out, rc = PTR_ERR(parent));
+ } else if (mdd_object_remote(parent)) {
+ /*FIXME: Because of the restriction of rename in Phase I.
+ * If the parent is remote, we just assumed lf is not the
+ * parent of P1 for now */
+ GOTO(out, rc = 0);
+ }
+ p1 = parent;
}
EXIT;
out:
int rc = 0;
ENTRY;
- if (cobj && mdd_object_exists(cobj))
+ if (cobj && mdd_object_exists(cobj))
RETURN(-EEXIST);
if (mdd_is_dead_obj(pobj))
if (tmp_la->la_uid == uc->uc_fsuid)
return 0;
- return !mdd_capable(uc, CFS_CAP_FOWNER);
+ return !md_capable(uc, CFS_CAP_FOWNER);
+}
+
+static int mdd_may_delete_entry(const struct lu_env *env,
+ struct mdd_object *pobj, int check_perm)
+{
+ ENTRY;
+
+ LASSERT(pobj != NULL);
+ if (!mdd_object_exists(pobj))
+ RETURN(-ENOENT);
+
+ if (mdd_is_dead_obj(pobj))
+ RETURN(-ENOENT);
+
+ if (check_perm) {
+ int rc;
+ rc = mdd_permission_internal_locked(env, pobj, NULL,
+ MAY_WRITE | MAY_EXEC,
+ MOR_TGT_PARENT);
+ if (rc)
+ RETURN(rc);
+ }
+
+ if (mdd_is_append(pobj))
+ RETURN(-EPERM);
+
+ RETURN(0);
}
/*
int rc = 0;
ENTRY;
- LASSERT(cobj);
+ if (pobj) {
+ rc = mdd_may_delete_entry(env, pobj, check_perm);
+ if (rc != 0)
+ RETURN(rc);
+ }
+
+ if (cobj == NULL)
+ RETURN(0);
+
if (!mdd_object_exists(cobj))
RETURN(-ENOENT);
if (mdd_is_dead_obj(cobj))
RETURN(-ESTALE);
- if (pobj) {
- if (!mdd_object_exists(pobj))
- RETURN(-ENOENT);
-
- if (mdd_is_dead_obj(pobj))
- RETURN(-ENOENT);
-
- if (check_perm) {
- rc = mdd_permission_internal_locked(env, pobj, NULL,
- MAY_WRITE | MAY_EXEC,
- MOR_TGT_PARENT);
- if (rc)
- RETURN(rc);
- }
-
- if (mdd_is_append(pobj))
- RETURN(-EPERM);
- }
if (mdd_is_sticky(env, pobj, cobj))
RETURN(-EPERM);
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,
reclen = llog_data_len(sizeof(*rec) +
(fname != NULL ? fname->ln_namelen : 0));
- buf = mdd_buf_alloc(env, reclen);
+ buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
if (buf->lb_buf == NULL)
return -ENOMEM;
reclen = llog_data_len(sizeof(*rec) +
(tname != NULL ? tname->ln_namelen : 0) +
(sname != NULL ? 1 + sname->ln_namelen : 0));
- buf = mdd_buf_alloc(env, reclen);
+ buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
if (buf->lb_buf == NULL)
return -ENOMEM;
* \param tname - target name string
* \param handle - transacion handle
*/
-static int mdd_changelog_ns_store(const struct lu_env *env,
- struct mdd_device *mdd,
- enum changelog_rec_type type,
- unsigned flags,
- struct mdd_object *target,
- struct mdd_object *parent,
- const struct lu_name *tname,
- struct thandle *handle)
+int mdd_changelog_ns_store(const struct lu_env *env, struct mdd_device *mdd,
+ enum changelog_rec_type type, unsigned flags,
+ struct mdd_object *target, struct mdd_object *parent,
+ const struct lu_name *tname, struct thandle *handle)
{
struct llog_changelog_rec *rec;
struct lu_buf *buf;
LASSERT(handle != NULL);
reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
- buf = mdd_buf_alloc(env, reclen);
+ buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
if (buf->lb_buf == NULL)
RETURN(-ENOMEM);
rec = buf->lb_buf;
reclen = llog_data_len(sizeof(*rec) +
sname != NULL ? 1 + sname->ln_namelen : 0);
- buf = mdd_buf_alloc(env, reclen);
+ buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
if (buf->lb_buf == NULL)
RETURN(-ENOMEM);
rec = buf->lb_buf;
rec->cr.cr_namelen = tname->ln_namelen;
memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
if (sname) {
- LASSERT(sfid != NULL);
rec->cr.cr_name[tname->ln_namelen] = '\0';
memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name,
sname->ln_namelen);
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)
+static int __mdd_links_add(const struct lu_env *env,
+ struct mdd_object *mdd_obj,
+ struct linkea_data *ldata,
+ const struct lu_name *lname,
+ const struct lu_fid *pfid,
+ int first, int check)
{
- int rc;
-
- rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
- if (rc)
- return rc;
+ int rc;
- rc = mdo_declare_ref_add(env, c, handle);
- if (rc)
- return rc;
+ if (ldata->ld_leh == NULL) {
+ rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
+ if (rc) {
+ if (rc != -ENODATA)
+ return rc;
+ rc = linkea_data_new(ldata,
+ &mdd_env_info(env)->mti_link_buf);
+ if (rc)
+ return rc;
+ }
+ }
- rc = mdo_declare_attr_set(env, p, NULL, handle);
- if (rc)
- return rc;
+ if (check) {
+ rc = linkea_links_find(ldata, lname, pfid);
+ if (rc && rc != -ENOENT)
+ return rc;
+ if (rc == 0)
+ return -EEXIST;
+ }
- rc = mdo_declare_attr_set(env, c, NULL, handle);
- if (rc)
- return rc;
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
+ struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
- rc = mdd_declare_links_add(env, c, handle);
- if (rc)
- return rc;
+ *tfid = *pfid;
+ tfid->f_ver = ~0;
+ linkea_add_buf(ldata, lname, tfid);
+ }
- rc = mdd_declare_changelog_store(env, mdd, name, handle);
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
+ linkea_add_buf(ldata, lname, pfid);
- return rc;
+ return linkea_add_buf(ldata, lname, pfid);
}
-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)
+static int __mdd_links_del(const struct lu_env *env,
+ struct mdd_object *mdd_obj,
+ struct linkea_data *ldata,
+ const struct lu_name *lname,
+ const struct lu_fid *pfid)
{
- const char *name = lname->ln_name;
- struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
- struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
- struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
- struct mdd_device *mdd = mdo2mdd(src_obj);
- struct dynlock_handle *dlh;
- struct thandle *handle;
- int rc;
- ENTRY;
+ int rc;
- handle = mdd_trans_create(env, mdd);
- if (IS_ERR(handle))
- GOTO(out_pending, rc = PTR_ERR(handle));
+ if (ldata->ld_leh == NULL) {
+ rc = mdd_links_read(env, mdd_obj, ldata);
+ if (rc)
+ return rc;
+ }
- rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle);
- if (rc)
- GOTO(stop, rc);
+ rc = linkea_links_find(ldata, lname, pfid);
+ if (rc)
+ return rc;
- rc = mdd_trans_start(env, mdd, handle);
- if (rc)
- GOTO(stop, rc);
+ linkea_del_buf(ldata, lname);
+ return 0;
+}
- dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
- if (dlh == NULL)
- GOTO(out_trans, rc = -ENOMEM);
- mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
+static int mdd_linkea_prepare(const struct lu_env *env,
+ struct mdd_object *mdd_obj,
+ const struct lu_fid *oldpfid,
+ const struct lu_name *oldlname,
+ const struct lu_fid *newpfid,
+ const struct lu_name *newlname,
+ int first, int check,
+ struct linkea_data *ldata)
+{
+ int rc = 0;
+ int rc2 = 0;
+ ENTRY;
- rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
- if (rc)
- GOTO(out_unlock, rc);
+ if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
+ return 0;
- rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
- name, handle,
- mdd_object_capa(env, mdd_tobj));
- if (rc)
- GOTO(out_unlock, rc);
+ LASSERT(oldpfid != NULL || newpfid != NULL);
- rc = mdo_ref_add(env, mdd_sobj, handle);
- if (rc != 0) {
- __mdd_index_delete_only(env, mdd_tobj, name, handle,
- mdd_object_capa(env, mdd_tobj));
- GOTO(out_unlock, rc);
+ if (mdd_obj->mod_flags & DEAD_OBJ) {
+ /* Prevent linkea to be updated which is NOT necessary. */
+ ldata->ld_reclen = 0;
+ /* No more links, don't bother */
+ RETURN(0);
}
- LASSERT(ma->ma_attr.la_valid & LA_CTIME);
- la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
+ if (oldpfid != NULL) {
+ rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
+ if (rc) {
+ if ((check == 0) ||
+ (rc != -ENODATA && rc != -ENOENT))
+ RETURN(rc);
+ /* No changes done. */
+ rc = 0;
+ }
+ }
- la->la_valid = LA_CTIME | LA_MTIME;
- rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
- if (rc)
- GOTO(out_unlock, rc);
+ /* If renaming, add the new record */
+ if (newpfid != NULL) {
+ /* even if the add fails, we still delete the out-of-date
+ * old link */
+ rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
+ first, check);
+ if (rc2 == -EEXIST)
+ rc2 = 0;
+ }
- 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, 0);
- }
+ rc = rc != 0 ? rc : rc2;
- EXIT;
-out_unlock:
- mdd_write_unlock(env, mdd_sobj);
- mdd_pdo_write_unlock(env, mdd_tobj, dlh);
-out_trans:
- if (rc == 0)
- rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
- mdd_tobj, lname, handle);
-stop:
- mdd_trans_stop(env, mdd, rc, handle);
-out_pending:
- return rc;
+ RETURN(rc);
}
-int mdd_declare_finish_unlink(const struct lu_env *env,
- struct mdd_object *obj,
- struct md_attr *ma,
- struct thandle *handle)
+int mdd_links_rename(const struct lu_env *env,
+ struct mdd_object *mdd_obj,
+ const struct lu_fid *oldpfid,
+ const struct lu_name *oldlname,
+ const struct lu_fid *newpfid,
+ const struct lu_name *newlname,
+ struct thandle *handle,
+ struct linkea_data *ldata,
+ int first, int check)
{
- int rc;
+ int rc2 = 0;
+ int rc = 0;
+ ENTRY;
- rc = orph_declare_index_insert(env, obj, handle);
- if (rc)
- return rc;
+ if (ldata == NULL) {
+ ldata = &mdd_env_info(env)->mti_link_data;
+ memset(ldata, 0, sizeof(*ldata));
+ rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
+ newpfid, newlname, first, check,
+ ldata);
+ if (rc != 0)
+ GOTO(out, rc);
+ }
- return mdo_declare_destroy(env, obj, handle);
+ if (ldata->ld_reclen != 0)
+ rc = mdd_links_write(env, mdd_obj, ldata, handle);
+ EXIT;
+out:
+ if (rc == 0)
+ rc = rc2;
+ if (rc) {
+ int error = 1;
+ if (rc == -EOVERFLOW || rc == -ENOENT || rc == -ENOSPC)
+ error = 0;
+ if (oldpfid == NULL)
+ CDEBUG(error ? D_ERROR : D_OTHER,
+ "link_ea add '%.*s' failed %d "DFID"\n",
+ newlname->ln_namelen, newlname->ln_name,
+ rc, PFID(mdd_object_fid(mdd_obj)));
+ else if (newpfid == NULL)
+ CDEBUG(error ? D_ERROR : D_OTHER,
+ "link_ea del '%.*s' failed %d "DFID"\n",
+ oldlname->ln_namelen, oldlname->ln_name,
+ rc, PFID(mdd_object_fid(mdd_obj)));
+ else
+ CDEBUG(error ? D_ERROR : D_OTHER,
+ "link_ea rename '%.*s'->'%.*s' failed %d "
+ DFID"\n",
+ oldlname->ln_namelen, oldlname->ln_name,
+ newlname->ln_namelen, newlname->ln_name,
+ rc, PFID(mdd_object_fid(mdd_obj)));
+ }
+
+ if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
+ /* if we vmalloced a large buffer drop it */
+ lu_buf_free(ldata->ld_buf);
+
+ return rc;
}
-/* caller should take a lock before calling */
-int mdd_finish_unlink(const struct lu_env *env,
- struct mdd_object *obj, struct md_attr *ma,
- struct thandle *th)
+static inline 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 linkea_data *data, int first)
{
- int rc = 0;
- int is_dir = S_ISDIR(ma->ma_attr.la_mode);
- ENTRY;
+ return mdd_links_rename(env, mdd_obj, NULL, NULL,
+ pfid, lname, handle, data, first, 0);
+}
- LASSERT(mdd_write_locked(env, obj) != 0);
+static inline int mdd_links_del(const struct lu_env *env,
+ struct mdd_object *mdd_obj,
+ const struct lu_fid *pfid,
+ const struct lu_name *lname,
+ struct thandle *handle)
+{
+ return mdd_links_rename(env, mdd_obj, pfid, lname,
+ NULL, NULL, handle, NULL, 0, 0);
+}
- 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() */
- if (obj->mod_count) {
- rc = __mdd_orphan_add(env, obj, th);
- if (rc == 0)
- CDEBUG(D_HA, "Object "DFID" is inserted into "
+/** Read the link EA into a temp buffer.
+ * Uses the mdd_thread_info::mti_big_buf since it is generally large.
+ * A pointer to the buffer is stored in \a ldata::ld_buf.
+ *
+ * \retval 0 or error
+ */
+int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
+ struct linkea_data *ldata)
+{
+ int rc;
+
+ /* First try a small buf */
+ LASSERT(env != NULL);
+ ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
+ CFS_PAGE_SIZE);
+ if (ldata->ld_buf->lb_buf == NULL)
+ return -ENOMEM;
+
+ if (!mdd_object_exists(mdd_obj))
+ return -ENODATA;
+
+ rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK,
+ BYPASS_CAPA);
+ if (rc == -ERANGE) {
+ /* Buf was too small, figure out what we need. */
+ lu_buf_free(ldata->ld_buf);
+ rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
+ XATTR_NAME_LINK, BYPASS_CAPA);
+ if (rc < 0)
+ return rc;
+ ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
+ if (ldata->ld_buf->lb_buf == NULL)
+ return -ENOMEM;
+ rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
+ XATTR_NAME_LINK, BYPASS_CAPA);
+ }
+ if (rc < 0)
+ return rc;
+
+ linkea_init(ldata);
+ return 0;
+}
+
+/** Read the link EA into a temp buffer.
+ * Uses the name_buf since it is generally large.
+ * \retval IS_ERR err
+ * \retval ptr to \a lu_buf (always \a mti_big_buf)
+ */
+struct lu_buf *mdd_links_get(const struct lu_env *env,
+ struct mdd_object *mdd_obj)
+{
+ struct linkea_data ldata = { 0 };
+ int rc;
+
+ rc = mdd_links_read(env, mdd_obj, &ldata);
+ return rc ? ERR_PTR(rc) : ldata.ld_buf;
+}
+
+int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
+ struct linkea_data *ldata, struct thandle *handle)
+{
+ const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
+ ldata->ld_leh->leh_len);
+ return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle,
+ mdd_object_capa(env, mdd_obj));
+}
+
+int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
+ struct thandle *handle, struct linkea_data *ldata)
+{
+ int rc;
+ int ea_len;
+ void *linkea;
+
+ if (ldata != NULL && ldata->ld_lee != NULL) {
+ ea_len = ldata->ld_leh->leh_len;
+ linkea = ldata->ld_buf->lb_buf;
+ } else {
+ ea_len = DEFAULT_LINKEA_SIZE;
+ linkea = NULL;
+ }
+
+ /* XXX: max size? */
+ rc = mdo_declare_xattr_set(env, mdd_obj,
+ mdd_buf_get_const(env, linkea, ea_len),
+ XATTR_NAME_LINK, 0, handle);
+ return rc;
+}
+
+static inline int mdd_declare_links_del(const struct lu_env *env,
+ struct mdd_object *c,
+ struct thandle *handle)
+{
+ int rc = 0;
+
+ /* For directory, the linkEA will be removed together
+ * with the object. */
+ if (!S_ISDIR(mdd_object_type(c)))
+ rc = mdd_declare_links_add(env, c, handle, NULL);
+
+ return rc;
+}
+
+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,
+ struct lu_attr *la,
+ struct linkea_data *data)
+{
+ 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;
+
+ la->la_valid = LA_CTIME | LA_MTIME;
+ rc = mdo_declare_attr_set(env, p, la, handle);
+ if (rc != 0)
+ return rc;
+
+ la->la_valid = LA_CTIME;
+ rc = mdo_declare_attr_set(env, c, la, handle);
+ if (rc)
+ return rc;
+
+ rc = mdd_declare_links_add(env, c, handle, data);
+ 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)
+{
+ const char *name = lname->ln_name;
+ struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
+ struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
+ struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
+ struct mdd_device *mdd = mdo2mdd(src_obj);
+ struct dynlock_handle *dlh;
+ struct thandle *handle;
+ struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
+ int rc;
+ ENTRY;
+
+ handle = mdd_trans_create(env, mdd);
+ if (IS_ERR(handle))
+ GOTO(out_pending, rc = PTR_ERR(handle));
+
+ memset(ldata, 0, sizeof(*ldata));
+
+ LASSERT(ma->ma_attr.la_valid & LA_CTIME);
+ la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
+
+ rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
+ la, ldata);
+ 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);
+ mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
+
+ rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
+ if (rc)
+ GOTO(out_unlock, rc);
+
+ rc = mdo_ref_add(env, mdd_sobj, handle);
+ if (rc)
+ GOTO(out_unlock, rc);
+
+
+ rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
+ name, handle,
+ mdd_object_capa(env, mdd_tobj));
+ if (rc != 0) {
+ mdo_ref_del(env, mdd_sobj, handle);
+ GOTO(out_unlock, rc);
+ }
+
+ la->la_valid = LA_CTIME | LA_MTIME;
+ rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
+ if (rc)
+ GOTO(out_unlock, rc);
+
+ la->la_valid = LA_CTIME;
+ rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
+ if (rc == 0) {
+ rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
+ mdo2fid(mdd_tobj), lname, 0, 0,
+ ldata);
+ if (rc == 0)
+ mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
+ lname, handle, ldata, 0);
+ /* The failure of links_add should not cause the link
+ * failure, reset rc here */
+ rc = 0;
+ }
+ EXIT;
+out_unlock:
+ mdd_write_unlock(env, mdd_sobj);
+ mdd_pdo_write_unlock(env, mdd_tobj, dlh);
+out_trans:
+ if (rc == 0)
+ rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
+ mdd_tobj, lname, handle);
+stop:
+ mdd_trans_stop(env, mdd, rc, handle);
+
+ if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
+ /* if we vmalloced a large buffer drop it */
+ lu_buf_free(ldata->ld_buf);
+out_pending:
+ 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, mdd_object_type(obj), handle);
+ if (rc)
+ return rc;
+
+ return mdo_declare_destroy(env, obj, 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,
+ struct thandle *th)
+{
+ int rc = 0;
+ int is_dir = S_ISDIR(ma->ma_attr.la_mode);
+ ENTRY;
+
+ LASSERT(mdd_write_locked(env, obj) != 0);
+
+ if (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() */
+ if (obj->mod_count) {
+ rc = __mdd_orphan_add(env, obj, th);
+ if (rc == 0)
+ CDEBUG(D_HA, "Object "DFID" is inserted into "
"orphan list, open count = %d\n",
PFID(mdd_object_fid(obj)),
obj->mod_count);
}
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)
+ struct mdd_object *p, struct mdd_object *c,
+ const struct lu_name *name, struct md_attr *ma,
+ struct thandle *handle, int no_name)
{
+ struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
int rc;
- rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
- if (rc)
- return rc;
+ if (likely(no_name == 0)) {
+ 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;
+ LASSERT(ma->ma_attr.la_valid & LA_CTIME);
+ la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
+ la->la_valid = LA_CTIME | LA_MTIME;
+ rc = mdo_declare_attr_set(env, p, la, handle);
+ if (rc)
+ return rc;
- rc = mdo_declare_ref_del(env, c, handle);
- if (rc)
- return rc;
+ if (c != NULL) {
+ 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_ref_del(env, c, handle);
+ if (rc)
+ return rc;
- rc = mdo_declare_attr_set(env, c, NULL, handle);
- if (rc)
- return rc;
+ la->la_valid = LA_CTIME;
+ rc = mdo_declare_attr_set(env, c, la, handle);
+ if (rc)
+ return rc;
- rc = mdd_declare_finish_unlink(env, c, ma, 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_links_del(env, c, handle);
+ if (rc != 0)
+ return rc;
- rc = mdd_declare_changelog_store(env, mdd, name, handle);
+ /* FIXME: need changelog for remove entry */
+ rc = mdd_declare_changelog_store(env, mdd, name, handle);
+ }
- return rc;
+ return rc;
}
+/**
+ * Delete name entry and the object.
+ * Note: no_name == 1 means it only destory the object, i.e. name_entry
+ * does not exist for this object, and it could only happen during resending
+ * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
+ * is also needed in this case(needed by changelog), so we have to add another
+ * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
+ * the ENOENT failure should be able to be fixed by redo mechanism.
+ */
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)
+ struct md_object *cobj, const struct lu_name *lname,
+ struct md_attr *ma, int no_name)
{
const char *name = lname->ln_name;
struct lu_attr *cattr = &mdd_env_info(env)->mti_cattr;
struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
- struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
+ struct mdd_object *mdd_cobj = NULL;
struct mdd_device *mdd = mdo2mdd(pobj);
struct dynlock_handle *dlh;
struct thandle *handle;
- int rc, is_dir;
+ int rc, is_dir = 0;
ENTRY;
- if (mdd_object_exists(mdd_cobj) <= 0)
- RETURN(-ENOENT);
+ /* cobj == NULL means only delete name entry */
+ if (likely(cobj != NULL)) {
+ mdd_cobj = md2mdd_obj(cobj);
+ if (mdd_object_exists(mdd_cobj) == 0)
+ RETURN(-ENOENT);
+ /* currently it is assume, it could only delete
+ * name entry of remote directory */
+ is_dir = 1;
+ }
- handle = mdd_trans_create(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_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
+ lname, ma, handle, no_name);
+ if (rc)
+ GOTO(stop, rc);
- rc = mdd_trans_start(env, mdd, 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)
+ dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
+ if (dlh == NULL)
GOTO(stop, rc = -ENOMEM);
- mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
- /* fetch cattr */
- rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj));
- if (rc)
- GOTO(cleanup, rc);
+ if (likely(mdd_cobj != NULL)) {
+ mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
- is_dir = S_ISDIR(cattr->la_mode);
+ /* fetch cattr */
+ rc = mdd_la_get(env, mdd_cobj, cattr,
+ mdd_object_capa(env, mdd_cobj));
+ if (rc)
+ GOTO(cleanup, rc);
- rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr);
- if (rc)
- GOTO(cleanup, rc);
+ is_dir = S_ISDIR(cattr->la_mode);
- rc = mdo_ref_del(env, mdd_cobj, handle);
- if (rc != 0) {
- __mdd_index_insert_only(env, mdd_pobj, mdo2fid(mdd_cobj),
- name, handle,
- mdd_object_capa(env, mdd_pobj));
- GOTO(cleanup, rc);
}
- rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
- mdd_object_capa(env, mdd_pobj));
+ rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr);
if (rc)
GOTO(cleanup, rc);
- if (is_dir)
- /* unlink dot */
- mdo_ref_del(env, mdd_cobj, handle);
+ if (likely(no_name == 0)) {
+ rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
+ mdd_object_capa(env, mdd_pobj));
+ if (rc)
+ GOTO(cleanup, rc);
+ }
- /* fetch updated nlink */
- rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj));
- if (rc)
- GOTO(cleanup, rc);
+ if (likely(mdd_cobj != NULL)) {
+ rc = mdo_ref_del(env, mdd_cobj, handle);
+ if (rc != 0) {
+ __mdd_index_insert_only(env, mdd_pobj,
+ mdo2fid(mdd_cobj),
+ name, handle,
+ mdd_object_capa(env, mdd_pobj));
+ GOTO(cleanup, rc);
+ }
- LASSERT(ma->ma_attr.la_valid & LA_CTIME);
- la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
+ if (is_dir)
+ /* unlink dot */
+ mdo_ref_del(env, mdd_cobj, handle);
- la->la_valid = LA_CTIME | LA_MTIME;
+ /* fetch updated nlink */
+ rc = mdd_la_get(env, mdd_cobj, cattr,
+ mdd_object_capa(env, mdd_cobj));
+ if (rc)
+ GOTO(cleanup, rc);
+ }
+
+ LASSERT(ma->ma_attr.la_valid & LA_CTIME);
+ la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
+
+ la->la_valid = LA_CTIME | LA_MTIME;
rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
- if (rc)
- GOTO(cleanup, rc);
+ if (rc)
+ GOTO(cleanup, rc);
+
+ /* Enough for only unlink the entry */
+ if (unlikely(mdd_cobj == NULL)) {
+ mdd_pdo_write_unlock(env, mdd_pobj, dlh);
+ GOTO(stop, rc);
+ }
if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
/* update ctime of an unlinked file only if it is still
int rc;
ENTRY;
- /* do not let users to create stripes via .lustre/
- * mdd_obf_setup() sets IMMUTE_OBJ on this directory */
- if (pobj && mdd_pobj->mod_flags & IMMUTE_OBJ)
- RETURN(-ENOENT);
-
rc = mdd_cd_sanity_check(env, son);
if (rc)
RETURN(rc);
spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
spec->sp_cr_flags, spec->no_create);
- if (spec->no_create) {
- /* replay case */
+ if (spec->no_create || spec->sp_cr_flags & MDS_OPEN_HAS_EA) {
+ /* replay case or lfs setstripe */
buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
spec->u.sp_ea.eadatalen);
- } else if (!(spec->sp_cr_flags & MDS_OPEN_HAS_OBJS)) {
- if (spec->sp_cr_flags & MDS_OPEN_HAS_EA) {
- /* lfs setstripe */
- buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
- spec->u.sp_ea.eadatalen);
- } else {
- buf = &LU_BUF_NULL;
- }
} else {
- /* MDS_OPEN_HAS_OBJS is not used anymore ? */
- LBUG();
+ buf = &LU_BUF_NULL;
}
rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
if (rc)
GOTO(stop, rc);
+ rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
+ if (rc)
+ GOTO(stop, rc);
+
rc = mdd_trans_start(env, mdd, handle);
if (rc)
GOTO(stop, rc);
rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
0, handle, mdd_object_capa(env, son));
+
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
+
stop:
mdd_trans_stop(env, mdd, rc, handle);
out_free:
if (unlikely(mdd_is_dead_obj(mdd_obj)))
RETURN(-ESTALE);
- rc = mdd_object_exists(mdd_obj);
- if (unlikely(rc == 0))
- RETURN(-ESTALE);
- else if (unlikely(rc < 0)) {
- CERROR("Object "DFID" locates on remote server\n",
- PFID(mdo2fid(mdd_obj)));
- RETURN(-EINVAL);
- }
+ if (mdd_object_remote(mdd_obj)) {
+ CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
+ mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
+ } else if (!mdd_object_exists(mdd_obj)) {
+ RETURN(-ESTALE);
+ }
/* The common filename length check. */
if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
RETURN(rc);
}
-int mdd_declare_object_initialize(const struct lu_env *env,
- struct mdd_object *child,
- struct lu_attr *attr,
- struct thandle *handle)
+static int mdd_declare_object_initialize(const struct lu_env *env,
+ struct mdd_object *parent,
+ struct mdd_object *child,
+ struct lu_attr *attr,
+ struct thandle *handle,
+ struct linkea_data *ldata)
{
int rc;
+ ENTRY;
+ /*
+ * inode mode has been set in creation time, and it's based on umask,
+ * la_mode and acl, don't set here again! (which will go wrong
+ * because below function doesn't consider umask).
+ * I'd suggest set all object attributes in creation time, see above.
+ */
+ LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
+ attr->la_valid &= ~(LA_MODE | LA_TYPE);
rc = mdo_declare_attr_set(env, child, attr, handle);
+ attr->la_valid |= LA_MODE | LA_TYPE;
if (rc == 0 && S_ISDIR(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);
+
+ rc = mdo_declare_index_insert(env, child, mdo2fid(parent),
+ dotdot, handle);
}
- if (rc == 0)
- mdd_declare_links_add(env, child, handle);
+ if (rc == 0)
+ mdd_declare_links_add(env, child, handle, ldata);
- return rc;
+ 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 lu_attr *attr, struct thandle *handle,
- const struct md_op_spec *spec)
+static int mdd_object_initialize(const struct lu_env *env,
+ const struct lu_fid *pfid,
+ const struct lu_name *lname,
+ struct mdd_object *child,
+ struct lu_attr *attr, struct thandle *handle,
+ const struct md_op_spec *spec,
+ struct linkea_data *ldata)
{
int rc;
ENTRY;
* (2) maybe, the child attributes should be set in OSD when creation.
*/
- /*
- * inode mode has been set in creation time, and it's based on umask,
- * la_mode and acl, don't set here again! (which will go wrong
- * because below function doesn't consider umask).
- * I'd suggest set all object attributes in creation time, see above.
- */
- attr->la_valid &= ~LA_MODE;
rc = mdd_attr_set_internal(env, child, attr, handle, 0);
- if (rc != 0)
- RETURN(rc);
-
+ /* arguments are supposed to stay the same */
if (S_ISDIR(attr->la_mode)) {
/* Add "." and ".." for newly created dir */
mdo_ref_add(env, child, handle);
if (rc != 0)
mdo_ref_del(env, child, handle);
}
- if (rc == 0)
- mdd_links_add(env, child, pfid, lname, handle, 1);
- RETURN(rc);
+ if (rc == 0)
+ mdd_links_add(env, child, pfid, lname, handle, ldata, 1);
+
+ RETURN(rc);
}
/* has not lock on pobj yet */
struct lu_fid *fid = &info->mti_fid;
struct mdd_object *obj = md2mdd_obj(pobj);
struct mdd_device *m = mdo2mdd(pobj);
- int lookup = spec->sp_cr_lookup;
int rc;
ENTRY;
if (mdd_is_dead_obj(obj))
RETURN(-ENOENT);
- /*
+ /*
* In some cases this lookup is not needed - we know before if name
* exists or not because MDT performs lookup for it.
* name length check is done in lookup.
*/
- if (lookup) {
+ if (spec->sp_cr_lookup) {
/*
* Check if the name already exist, though it will be checked in
* _index_insert also, for avoiding rolling back if exists
struct lu_attr *attr,
int got_def_acl,
struct thandle *handle,
- const struct md_op_spec *spec)
+ const struct md_op_spec *spec,
+ struct linkea_data *ldata)
{
- struct mdd_thread_info *info = mdd_env_info(env);
- int rc = 0;
+ int rc;
rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec);
if (rc)
GOTO(out, rc);
}
- rc = mdo_declare_attr_set(env, c, &info->mti_pattr, handle);
+ rc = mdo_declare_attr_set(env, c, attr, handle);
if (rc)
GOTO(out, rc);
GOTO(out, rc);
}
- rc = mdd_declare_object_initialize(env, c, attr, handle);
- if (rc)
- GOTO(out, rc);
+ rc = mdd_declare_object_initialize(env, p, c, attr, handle, ldata);
+ if (rc)
+ GOTO(out, rc);
- rc = mdo_declare_index_insert(env, p, mdo2fid(c),
- name->ln_name, handle);
- if (rc)
- GOTO(out, rc);
+ if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
+ rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
+ else
+ rc = mdo_declare_index_insert(env, p, mdo2fid(c),
+ name->ln_name, handle);
+ if (rc)
+ GOTO(out, rc);
/* replay case, create LOV EA from client data */
if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
GOTO(out, rc);
}
- rc = mdo_declare_attr_set(env, p, attr, handle);
- if (rc)
- return rc;
+ if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
+ rc = mdo_declare_attr_set(env, p, attr, handle);
+ if (rc)
+ return rc;
+ }
rc = mdd_declare_changelog_store(env, mdd, name, handle);
if (rc)
return rc;
}
+static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
+ struct lu_attr *la, struct lu_buf *acl_buf,
+ int *got_def_acl, int *reset_acl)
+{
+ int rc;
+ ENTRY;
+
+ if (S_ISLNK(la->la_mode))
+ RETURN(0);
+
+ mdd_read_lock(env, pobj, MOR_TGT_PARENT);
+ rc = mdo_xattr_get(env, pobj, acl_buf,
+ XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
+ mdd_read_unlock(env, pobj);
+ if (rc > 0) {
+ /* If there are default ACL, fix mode by default ACL */
+ *got_def_acl = rc;
+ acl_buf->lb_len = rc;
+ rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
+ if (rc < 0)
+ RETURN(rc);
+ *reset_acl = rc;
+ } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
+ /* If there are no default ACL, fix mode by mask */
+ struct lu_ucred *uc = lu_ucred(env);
+
+ /* The create triggered by MDT internal events, such as
+ * LFSCK reset, will not contain valid "uc". */
+ if (unlikely(uc != NULL))
+ la->la_mode &= ~uc->uc_umask;
+ rc = 0;
+ }
+
+ RETURN(rc);
+}
/*
* Create object and insert it into namespace.
const struct lu_name *lname, struct md_object *child,
struct md_op_spec *spec, struct md_attr* ma)
{
- struct mdd_thread_info *info = mdd_env_info(env);
- struct lu_attr *la = &info->mti_la_for_fix;
- struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
- struct mdd_object *son = md2mdd_obj(child);
- struct mdd_device *mdd = mdo2mdd(pobj);
- struct lu_attr *attr = &ma->ma_attr;
- struct thandle *handle;
- struct lu_attr *pattr = &info->mti_pattr;
- struct dynlock_handle *dlh;
- const char *name = lname->ln_name;
- int rc, created = 0, initialized = 0, inserted = 0;
- int got_def_acl = 0;
- ENTRY;
+ struct mdd_thread_info *info = mdd_env_info(env);
+ struct lu_attr *la = &info->mti_la_for_fix;
+ struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
+ struct mdd_object *son = md2mdd_obj(child);
+ struct mdd_device *mdd = mdo2mdd(pobj);
+ struct lu_attr *attr = &ma->ma_attr;
+ struct thandle *handle;
+ struct lu_attr *pattr = &info->mti_pattr;
+ struct lu_buf acl_buf;
+ struct linkea_data *ldata = &info->mti_link_data;
+ struct dynlock_handle *dlh;
+ const char *name = lname->ln_name;
+ int rc, created = 0, initialized = 0, inserted = 0;
+ int got_def_acl = 0;
+ int reset_acl = 0;
+ ENTRY;
/*
* Two operations have to be performed:
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
GOTO(out_free, rc = -EINPROGRESS);
- if (!S_ISLNK(attr->la_mode)) {
- struct lu_buf *acl_buf;
-
- acl_buf = mdd_buf_get(env, info->mti_xattr_buf,
- sizeof(info->mti_xattr_buf));
- mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
- rc = mdo_xattr_get(env, mdd_pobj, acl_buf,
- XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
- mdd_read_unlock(env, mdd_pobj);
- if (rc > 0)
- got_def_acl = rc;
- else if (rc < 0 && rc != -EOPNOTSUPP && rc != -ENODATA)
- GOTO(out_free, rc);
- }
+ acl_buf.lb_buf = info->mti_xattr_buf;
+ acl_buf.lb_len = sizeof(info->mti_xattr_buf);
+ rc = mdd_acl_init(env, mdd_pobj, attr, &acl_buf, &got_def_acl,
+ &reset_acl);
+ if (rc < 0)
+ GOTO(out_free, rc);
mdd_object_make_hint(env, mdd_pobj, son, attr);
if (IS_ERR(handle))
GOTO(out_free, rc = PTR_ERR(handle));
+ memset(ldata, 0, sizeof(*ldata));
+ mdd_linkea_prepare(env, son, NULL, NULL, mdd_object_fid(mdd_pobj),
+ lname, 1, 0, ldata);
rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
- got_def_acl, handle, spec);
+ got_def_acl, handle, spec, ldata);
if (rc)
GOTO(out_stop, rc);
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);
+ dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
+ if (dlh == NULL)
+ GOTO(out_trans, rc = -ENOMEM);
- mdd_write_lock(env, son, MOR_TGT_CHILD);
- rc = mdd_object_create_internal(env, mdd_pobj, son, attr, handle, spec);
- if (rc) {
- mdd_write_unlock(env, son);
- GOTO(cleanup, rc);
- }
+ mdd_write_lock(env, son, MOR_TGT_CHILD);
+ rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec);
+ if (rc) {
+ mdd_write_unlock(env, son);
+ GOTO(cleanup, rc);
+ }
- created = 1;
+ created = 1;
#ifdef CONFIG_FS_POSIX_ACL
- if (got_def_acl) {
- struct lu_buf *acl_buf;
+ if (got_def_acl) {
+ /* set default acl */
+ if (S_ISDIR(attr->la_mode)) {
+ LASSERTF(acl_buf.lb_len == got_def_acl,
+ "invalid acl_buf: %p:%d got_def %d\n",
+ acl_buf.lb_buf, (int)acl_buf.lb_len,
+ got_def_acl);
+ rc = mdo_xattr_set(env, son, &acl_buf,
+ XATTR_NAME_ACL_DEFAULT, 0,
+ handle, BYPASS_CAPA);
+ if (rc) {
+ mdd_write_unlock(env, son);
+ GOTO(cleanup, rc);
+ }
+ }
- acl_buf = mdd_buf_get(env, info->mti_xattr_buf, got_def_acl);
- rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
- if (rc) {
- mdd_write_unlock(env, son);
- GOTO(cleanup, rc);
+ /* set its own acl */
+ if (reset_acl) {
+ LASSERTF(acl_buf.lb_buf != NULL && acl_buf.lb_len != 0,
+ "invalid acl_buf %p:%d\n", acl_buf.lb_buf,
+ (int)acl_buf.lb_len);
+ rc = mdo_xattr_set(env, son, &acl_buf,
+ XATTR_NAME_ACL_ACCESS,
+ 0, handle, BYPASS_CAPA);
+ if (rc) {
+ mdd_write_unlock(env, son);
+ GOTO(cleanup, rc);
+ }
}
- }
+ }
#endif
- rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
- son, attr, handle, spec);
+ rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
+ son, attr, handle, spec, ldata);
/*
* in case of replay we just set LOVEA provided by the client
* MDT calls this xattr_set(LOV) in a different transaction.
* probably this way we code can be made better.
*/
- if (rc == 0 &&
- (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA))) {
+ if (rc == 0 && (spec->no_create ||
+ (spec->sp_cr_flags & MDS_OPEN_HAS_EA))) {
const struct lu_buf *buf;
buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
rc = mdo_xattr_set(env, son, buf, XATTR_NAME_LOV, 0, handle,
BYPASS_CAPA);
}
- mdd_write_unlock(env, son);
- if (rc)
- /*
- * Object has no links, so it will be destroyed when last
- * reference is released. (XXX not now.)
- */
- GOTO(cleanup, rc);
- initialized = 1;
+ if (rc == 0 && spec->sp_cr_flags & MDS_OPEN_VOLATILE)
+ rc = __mdd_orphan_add(env, son, handle);
- rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
- name, S_ISDIR(attr->la_mode), handle,
- mdd_object_capa(env, mdd_pobj));
- if (rc)
- GOTO(cleanup, rc);
+ mdd_write_unlock(env, son);
- inserted = 1;
+ if (rc != 0)
+ /*
+ * Object has no links, so it will be destroyed when last
+ * reference is released. (XXX not now.)
+ */
+ GOTO(cleanup, rc);
+
+ initialized = 1;
+
+ if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE))
+ rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
+ name, S_ISDIR(attr->la_mode), handle,
+ mdd_object_capa(env, mdd_pobj));
+
+ if (rc != 0)
+ GOTO(cleanup, rc);
+
+ inserted = 1;
if (S_ISLNK(attr->la_mode)) {
struct lu_ucred *uc = lu_ucred_assert(env);
GOTO(cleanup, rc = -EFAULT);
}
+ /* volatile file creation does not update parent directory times */
+ if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
+ GOTO(cleanup, rc = 0);
+
+ /* update parent directory mtime/ctime */
*la = *attr;
- la->la_valid = LA_CTIME | LA_MTIME;
+ la->la_valid = LA_CTIME | LA_MTIME;
rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
- if (rc)
- GOTO(cleanup, rc);
+ if (rc)
+ GOTO(cleanup, rc);
EXIT;
cleanup:
int rc2;
if (inserted != 0) {
- rc2 = __mdd_index_delete(env, mdd_pobj, name,
- S_ISDIR(attr->la_mode),
- handle, BYPASS_CAPA);
+ if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
+ rc2 = __mdd_orphan_del(env, son, handle);
+ else
+ rc2 = __mdd_index_delete(env, mdd_pobj, name,
+ S_ISDIR(attr->la_mode),
+ handle, BYPASS_CAPA);
if (rc2 != 0)
goto out_stop;
}
mdd_pdo_write_unlock(env, mdd_pobj, dlh);
out_trans:
- if (rc == 0)
+ if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)))
rc = mdd_changelog_ns_store(env, mdd,
S_ISDIR(attr->la_mode) ? CL_MKDIR :
S_ISREG(attr->la_mode) ? CL_CREATE :
out_stop:
mdd_trans_stop(env, mdd, rc, handle);
out_free:
+ if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
+ /* if we vmalloced a large buffer drop it */
+ lu_buf_free(ldata->ld_buf);
+
/* The child object shouldn't be cached anymore */
if (rc)
set_bit(LU_OBJECT_HEARD_BANSHEE,
struct lu_attr *so_attr,
struct lu_attr *tg_attr)
{
- int rc = 0;
- ENTRY;
+ int rc = 0;
+ ENTRY;
- /* XXX: when get here, sobj must NOT be NULL,
- * the other case has been processed in cml_rename
- * before mdd_rename and enable MDS_PERM_BYPASS. */
- LASSERT(sobj);
+ /* XXX: when get here, sobj must NOT be NULL,
+ * the other case has been processed in cld_rename
+ * before mdd_rename and enable MDS_PERM_BYPASS. */
+ LASSERT(sobj);
rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0);
- if (rc)
- RETURN(rc);
+ if (rc)
+ RETURN(rc);
- /* XXX: when get here, "tobj == NULL" means tobj must
- * NOT exist (neither on remote MDS, such case has been
- * processed in cml_rename before mdd_rename and enable
- * MDS_PERM_BYPASS).
- * So check may_create, but not check may_unlink. */
- if (!tobj)
- rc = mdd_may_create(env, tgt_pobj, NULL,
- (src_pobj != tgt_pobj), 0);
- else
+ /* XXX: when get here, "tobj == NULL" means tobj must
+ * NOT exist (neither on remote MDS, such case has been
+ * processed in cld_rename before mdd_rename and enable
+ * MDS_PERM_BYPASS).
+ * So check may_create, but not check may_unlink. */
+ if (!tobj)
+ rc = mdd_may_create(env, tgt_pobj, NULL,
+ (src_pobj != tgt_pobj), 0);
+ else
rc = mdd_may_delete(env, tgt_pobj, tobj, tg_attr, so_attr,
- (src_pobj != tgt_pobj), 1);
+ (src_pobj != tgt_pobj), 1);
- if (!rc && !tobj && (src_pobj != tgt_pobj) &&
+ if (!rc && !tobj && (src_pobj != tgt_pobj) &&
S_ISDIR(so_attr->la_mode))
- rc = __mdd_may_link(env, tgt_pobj);
+ rc = __mdd_may_link(env, tgt_pobj);
- 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 *tname,
+ 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 *tname,
const struct lu_name *sname,
- struct md_attr *ma,
- struct thandle *handle)
+ struct md_attr *ma,
+ struct linkea_data *ldata,
+ struct thandle *handle)
{
- int rc;
+ struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
+ int rc;
+
+ LASSERT(ma->ma_attr.la_valid & LA_CTIME);
+ la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
LASSERT(mdd_spobj);
LASSERT(mdd_tpobj);
rc = mdo_declare_ref_del(env, mdd_spobj, handle);
if (rc)
return rc;
+ if (mdd_spobj != mdd_tpobj) {
+ rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
+ 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;
-
+ 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);
}
- rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle);
- if (rc)
- return rc;
+ la->la_valid = LA_CTIME | LA_MTIME;
+ rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
+ if (rc != 0)
+ 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, la, handle);
+ if (rc != 0)
+ return rc;
- rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle);
- if (rc)
- return rc;
+ la->la_valid = LA_CTIME;
+ rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
+ if (rc)
+ return rc;
+
+ rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
+ if (rc)
+ return rc;
/* new name */
rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
return rc;
}
- rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle);
+ la->la_valid = LA_CTIME;
+ rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
if (rc)
return rc;
- mdd_declare_links_add(env, mdd_tobj, handle);
+ mdd_declare_links_del(env, mdd_tobj, handle);
if (rc)
return rc;
struct md_object *tobj, const struct lu_name *ltname,
struct md_attr *ma)
{
- const char *sname = lsname->ln_name;
- const char *tname = ltname->ln_name;
- struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
+ const char *sname = lsname->ln_name;
+ const char *tname = ltname->ln_name;
+ struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
struct lu_attr *so_attr = &mdd_env_info(env)->mti_cattr;
struct lu_attr *tg_attr = &mdd_env_info(env)->mti_pattr;
- struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
- struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
- struct mdd_device *mdd = mdo2mdd(src_pobj);
- struct mdd_object *mdd_sobj = NULL; /* source object */
- struct mdd_object *mdd_tobj = NULL;
- struct dynlock_handle *sdlh, *tdlh;
- struct thandle *handle;
- const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
- const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
- bool is_dir;
+ struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
+ struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
+ struct mdd_device *mdd = mdo2mdd(src_pobj);
+ struct mdd_object *mdd_sobj = NULL; /* source object */
+ struct mdd_object *mdd_tobj = NULL;
+ struct dynlock_handle *sdlh = NULL, *tdlh = NULL;
+ struct thandle *handle;
+ struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
+ const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
+ const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
+ bool is_dir;
bool tobj_ref = 0;
bool tobj_locked = 0;
unsigned cl_flags = 0;
- int rc, rc2;
- ENTRY;
+ int rc, rc2;
+ ENTRY;
if (tobj)
mdd_tobj = md2mdd_obj(tobj);
mdd_sobj = mdd_object_find(env, mdd, lf);
+ rc = mdd_la_get(env, mdd_sobj, so_attr,
+ mdd_object_capa(env, mdd_sobj));
+ if (rc)
+ GOTO(out_pending, rc);
+
+ if (mdd_tobj) {
+ rc = mdd_la_get(env, mdd_tobj, tg_attr,
+ mdd_object_capa(env, mdd_tobj));
+ if (rc)
+ GOTO(out_pending, rc);
+ }
+
+ rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj,
+ mdd_tobj, so_attr, tg_attr);
+ if (rc)
+ GOTO(out_pending, rc);
+
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);
+ memset(ldata, 0, sizeof(*ldata));
+ mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj),
+ ltname, 1, 0, ldata);
+ rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
+ mdd_tobj, lsname, ltname, ma, ldata, handle);
+ if (rc)
+ GOTO(stop, rc);
rc = mdd_trans_start(env, mdd, handle);
if (rc)
if (sdlh == NULL || tdlh == NULL)
GOTO(cleanup, rc = -ENOMEM);
- rc = mdd_la_get(env, mdd_sobj, so_attr,
- mdd_object_capa(env, mdd_sobj));
- if (rc)
- GOTO(cleanup, rc);
-
- if (mdd_tobj) {
- rc = mdd_la_get(env, mdd_tobj, tg_attr,
- mdd_object_capa(env, mdd_tobj));
- if (rc)
- GOTO(cleanup, rc);
- }
-
- rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj,
- mdd_tobj, so_attr, tg_attr);
- if (rc)
- GOTO(cleanup, rc);
-
is_dir = S_ISDIR(so_attr->la_mode);
/* Remove source name from source directory */
handle, 0);
}
- if (rc == 0 && mdd_sobj) {
- mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
+ if (rc == 0 && mdd_sobj) {
+ mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
- mdo2fid(mdd_tpobj), ltname, handle, 0, 0);
+ mdo2fid(mdd_tpobj), ltname, handle, NULL,
+ 0, 0);
if (rc == -ENOENT)
/* Old files might not have EA entry */
mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
- lsname, handle, 0);
+ lsname, handle, NULL, 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. */
}
fixup_spobj:
- if (rc && is_dir && mdd_sobj) {
- rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
- BYPASS_CAPA);
+ if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
+ rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
+ BYPASS_CAPA);
- if (rc2)
- CWARN("sp obj dotdot delete error %d\n",rc2);
+ if (rc2)
+ CWARN("%s: sp obj dotdot delete error: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, rc2);
- rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
- dotdot, handle, BYPASS_CAPA);
- if (rc2)
- CWARN("sp obj dotdot insert error %d\n",rc2);
- }
+ rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
+ dotdot, handle, BYPASS_CAPA);
+ if (rc2)
+ CWARN("%s: sp obj dotdot insert error: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, rc2);
+ }
fixup_spobj2:
if (rc) {
stop:
mdd_trans_stop(env, mdd, rc, handle);
- if (mdd_sobj)
- mdd_object_put(env, mdd_sobj);
out_pending:
- return rc;
-}
-
-/**
- * The data that link search is done on.
- */
-struct mdd_link_data {
- /**
- * Buffer to keep link EA body.
- */
- struct lu_buf *ml_buf;
- /**
- * The matched header, entry and its lenght in the EA
- */
- struct link_ea_header *ml_leh;
- struct link_ea_entry *ml_lee;
- int ml_reclen;
-};
-
-static int mdd_links_new(const struct lu_env *env,
- struct mdd_link_data *ldata)
-{
- ldata->ml_buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
- if (ldata->ml_buf->lb_buf == NULL)
- return -ENOMEM;
- ldata->ml_leh = ldata->ml_buf->lb_buf;
- ldata->ml_leh->leh_magic = LINK_EA_MAGIC;
- ldata->ml_leh->leh_len = sizeof(struct link_ea_header);
- ldata->ml_leh->leh_reccount = 0;
- return 0;
-}
-
-/** Read the link EA into a temp buffer.
- * Uses the mdd_thread_info::mti_big_buf since it is generally large.
- * A pointer to the buffer is stored in \a ldata::ml_buf.
- *
- * \retval 0 or error
- */
-int mdd_links_read(const struct lu_env *env,
- struct mdd_object *mdd_obj,
- struct mdd_link_data *ldata)
-{
- struct lustre_capa *capa;
- struct link_ea_header *leh;
- int rc;
-
- /* First try a small buf */
- LASSERT(env != NULL);
- ldata->ml_buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
- if (ldata->ml_buf->lb_buf == NULL)
- return -ENOMEM;
-
- if (!mdd_object_exists(mdd_obj))
- return -ENODATA;
-
- capa = mdd_object_capa(env, mdd_obj);
- rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf,
- XATTR_NAME_LINK, capa);
- if (rc == -ERANGE) {
- /* Buf was too small, figure out what we need. */
- mdd_buf_put(ldata->ml_buf);
- rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf,
- XATTR_NAME_LINK, capa);
- if (rc < 0)
- return rc;
- ldata->ml_buf = mdd_buf_alloc(env, rc);
- if (ldata->ml_buf->lb_buf == NULL)
- return -ENOMEM;
- rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf,
- XATTR_NAME_LINK, capa);
- }
- if (rc < 0)
- return rc;
-
- leh = ldata->ml_buf->lb_buf;
- if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
- leh->leh_magic = LINK_EA_MAGIC;
- leh->leh_reccount = __swab32(leh->leh_reccount);
- leh->leh_len = __swab64(leh->leh_len);
- /* entries are swabbed by mdd_lee_unpack */
- }
- if (leh->leh_magic != LINK_EA_MAGIC)
- return -EINVAL;
- if (leh->leh_reccount == 0)
- return -ENODATA;
-
- ldata->ml_leh = leh;
- return 0;
-}
-
-/** Read the link EA into a temp buffer.
- * Uses the name_buf since it is generally large.
- * \retval IS_ERR err
- * \retval ptr to \a lu_buf (always \a mti_big_buf)
- */
-struct lu_buf *mdd_links_get(const struct lu_env *env,
- struct mdd_object *mdd_obj)
-{
- struct mdd_link_data ldata = { 0 };
- int rc;
-
- rc = mdd_links_read(env, mdd_obj, &ldata);
- return rc ? ERR_PTR(rc) : ldata.ml_buf;
-}
-
-static int mdd_links_write(const struct lu_env *env,
- struct mdd_object *mdd_obj,
- struct mdd_link_data *ldata,
- struct thandle *handle)
-{
- const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ml_buf->lb_buf,
- ldata->ml_leh->leh_len);
- return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle,
- mdd_object_capa(env, mdd_obj));
-}
-
-/** Pack a link_ea_entry.
- * All elements are stored as chars to avoid alignment issues.
- * Numbers are always big-endian
- * \retval record length
- */
-static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
- const struct lu_fid *pfid)
-{
- struct lu_fid tmpfid;
- int reclen;
-
- 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;
- lee->lee_reclen[1] = reclen & 0xff;
- return reclen;
-}
-
-void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
- struct lu_name *lname, struct lu_fid *pfid)
-{
- *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
- 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);
-}
-
-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.
- */
-#define LINKEA_MAX_COUNT 128
-
-/** Add a record to the end of link ea buf */
-static int mdd_links_add_buf(const struct lu_env *env,
- struct mdd_link_data *ldata,
- const struct lu_name *lname,
- const struct lu_fid *pfid)
-{
- LASSERT(ldata->ml_leh != NULL);
-
- if (lname == NULL || pfid == NULL)
- return -EINVAL;
-
- /* Make sure our buf is big enough for the new one */
- if (ldata->ml_leh->leh_reccount > LINKEA_MAX_COUNT)
- return -EOVERFLOW;
-
- ldata->ml_reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
- if (ldata->ml_leh->leh_len + ldata->ml_reclen >
- ldata->ml_buf->lb_len) {
- if (mdd_buf_grow(env, ldata->ml_leh->leh_len +
- ldata->ml_reclen) < 0)
- return -ENOMEM;
- }
-
- ldata->ml_leh = ldata->ml_buf->lb_buf;
- ldata->ml_lee = ldata->ml_buf->lb_buf + ldata->ml_leh->leh_len;
- ldata->ml_reclen = mdd_lee_pack(ldata->ml_lee, lname, pfid);
- ldata->ml_leh->leh_len += ldata->ml_reclen;
- ldata->ml_leh->leh_reccount++;
- CDEBUG(D_INODE, "New link_ea name '%.*s' is added\n",
- lname->ln_namelen, lname->ln_name);
- return 0;
-}
-
-/** Del the current record from the link ea buf */
-static void mdd_links_del_buf(const struct lu_env *env,
- struct mdd_link_data *ldata,
- const struct lu_name *lname)
-{
- LASSERT(ldata->ml_leh != NULL);
-
- ldata->ml_leh->leh_reccount--;
- ldata->ml_leh->leh_len -= ldata->ml_reclen;
- memmove(ldata->ml_lee, (char *)ldata->ml_lee + ldata->ml_reclen,
- (char *)ldata->ml_leh + ldata->ml_leh->leh_len -
- (char *)ldata->ml_lee);
- CDEBUG(D_INODE, "Old link_ea name '%.*s' is removed\n",
- lname->ln_namelen, lname->ln_name);
-
-}
-
-/**
- * Check if such a link exists in linkEA.
- *
- * \param mdd_obj object being handled
- * \param pfid parent fid the link to be found for
- * \param lname name in the parent's directory entry pointing to this object
- * \param ldata link data the search to be done on
- *
- * \retval 0 success
- * \retval -ENOENT link does not exist
- * \retval -ve on error
- */
-static int mdd_links_find(const struct lu_env *env,
- struct mdd_object *mdd_obj,
- struct mdd_link_data *ldata,
- const struct lu_name *lname,
- const struct lu_fid *pfid)
-{
- struct lu_name *tmpname = &mdd_env_info(env)->mti_name2;
- struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
- int count;
-
- LASSERT(ldata->ml_leh != NULL);
-
- /* link #0 */
- ldata->ml_lee = (struct link_ea_entry *)(ldata->ml_leh + 1);
-
- for (count = 0; count < ldata->ml_leh->leh_reccount; count++) {
- mdd_lee_unpack(ldata->ml_lee, &ldata->ml_reclen,
- tmpname, tmpfid);
- if (tmpname->ln_namelen == lname->ln_namelen &&
- lu_fid_eq(tmpfid, pfid) &&
- (strncmp(tmpname->ln_name, lname->ln_name,
- tmpname->ln_namelen) == 0))
- break;
- ldata->ml_lee = (struct link_ea_entry *)((char *)ldata->ml_lee +
- ldata->ml_reclen);
- }
-
- if (count == ldata->ml_leh->leh_reccount) {
- CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
- lname->ln_namelen, lname->ln_name);
- return -ENOENT;
- }
- return 0;
-}
-
-static int __mdd_links_add(const struct lu_env *env,
- struct mdd_object *mdd_obj,
- struct mdd_link_data *ldata,
- const struct lu_name *lname,
- const struct lu_fid *pfid,
- int first, int check)
-{
- int rc;
-
- if (ldata->ml_leh == NULL) {
- rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
- if (rc) {
- if (rc != -ENODATA)
- return rc;
- rc = mdd_links_new(env, ldata);
- if (rc)
- return rc;
- }
- }
-
- if (check) {
- rc = mdd_links_find(env, mdd_obj, ldata, lname, pfid);
- if (rc && rc != -ENOENT)
- return rc;
- if (rc == 0)
- return -EEXIST;
- }
-
- return mdd_links_add_buf(env, ldata, lname, pfid);
-}
-
-static int __mdd_links_del(const struct lu_env *env,
- struct mdd_object *mdd_obj,
- struct mdd_link_data *ldata,
- const struct lu_name *lname,
- const struct lu_fid *pfid)
-{
- int rc;
-
- if (ldata->ml_leh == NULL) {
- rc = mdd_links_read(env, mdd_obj, ldata);
- if (rc)
- return rc;
- }
-
- rc = mdd_links_find(env, mdd_obj, ldata, lname, pfid);
- if (rc)
- return rc;
-
- mdd_links_del_buf(env, ldata, lname);
- return 0;
-}
-
-static int mdd_links_rename(const struct lu_env *env,
- struct mdd_object *mdd_obj,
- const struct lu_fid *oldpfid,
- const struct lu_name *oldlname,
- const struct lu_fid *newpfid,
- const struct lu_name *newlname,
- struct thandle *handle,
- int first, int check)
-{
- struct mdd_link_data ldata = { 0 };
- int updated = 0;
- int rc2 = 0;
- int rc = 0;
- ENTRY;
-
- LASSERT(oldpfid != NULL || newpfid != NULL);
-
- if (mdd_obj->mod_flags & DEAD_OBJ)
- /* No more links, don't bother */
- RETURN(0);
-
- if (oldpfid != NULL) {
- rc = __mdd_links_del(env, mdd_obj, &ldata,
- oldlname, oldpfid);
- if (rc) {
- if ((check == 0) ||
- (rc != -ENODATA && rc != -ENOENT))
- GOTO(out, rc);
- /* No changes done. */
- rc = 0;
- } else {
- updated = 1;
- }
- }
-
- /* If renaming, add the new record */
- if (newpfid != NULL) {
- /* even if the add fails, we still delete the out-of-date
- * old link */
- rc2 = __mdd_links_add(env, mdd_obj, &ldata,
- newlname, newpfid, first, check);
- if (rc2 == -EEXIST)
- rc2 = 0;
- else if (rc2 == 0)
- updated = 1;
- }
-
- if (updated)
- rc = mdd_links_write(env, mdd_obj, &ldata, handle);
- EXIT;
-out:
- if (rc == 0)
- rc = rc2;
- if (rc) {
- if (oldpfid == NULL)
- CERROR("link_ea add '%.*s' failed %d "DFID"\n",
- newlname->ln_namelen, newlname->ln_name,
- rc, PFID(mdd_object_fid(mdd_obj)));
- else if (newpfid == NULL)
- CERROR("link_ea del '%.*s' failed %d "DFID"\n",
- oldlname->ln_namelen, oldlname->ln_name,
- rc, PFID(mdd_object_fid(mdd_obj)));
- else
- CERROR("link_ea rename '%.*s'->'%.*s' failed %d "
- DFID"\n",
- oldlname->ln_namelen, oldlname->ln_name,
- newlname->ln_namelen, newlname->ln_name,
- rc, PFID(mdd_object_fid(mdd_obj)));
- }
-
- if (ldata.ml_buf && ldata.ml_buf->lb_len > OBD_ALLOC_BIG)
- /* if we vmalloced a large buffer drop it */
- mdd_buf_put(ldata.ml_buf);
-
+ mdd_object_put(env, mdd_sobj);
return rc;
}
-static inline 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, int first)
-{
- return mdd_links_rename(env, mdd_obj, NULL, NULL,
- pfid, lname, handle, first, 0);
-}
-
-static inline int mdd_links_del(const struct lu_env *env,
- struct mdd_object *mdd_obj,
- const struct lu_fid *pfid,
- const struct lu_name *lname,
- struct thandle *handle)
-{
- return mdd_links_rename(env, mdd_obj, pfid, lname,
- NULL, NULL, handle, 0, 0);
-}
-
const struct md_dir_operations mdd_dir_ops = {
.mdo_is_subdir = mdd_is_subdir,
.mdo_lookup = mdd_lookup,