-/* -*- 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/
#define DEBUG_SUBSYSTEM S_MDS
#include <linux/module.h>
+#ifdef HAVE_EXT4_LDISKFS
+#include <ldiskfs/ldiskfs_jbd2.h>
+#else
#include <linux/jbd.h>
+#endif
#include <obd.h>
#include <obd_class.h>
#include <lustre_ver.h>
#include <obd_support.h>
#include <lprocfs_status.h>
-
+#ifdef HAVE_EXT4_LDISKFS
+#include <ldiskfs/ldiskfs.h>
+#else
#include <linux/ldiskfs_fs.h>
+#endif
#include <lustre_mds.h>
#include <lustre/lustre_idl.h>
#include <lustre_fid.h>
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, int first);
+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);
+
static int
__mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj,
const struct lu_name *lname, struct lu_fid* fid, int mask)
return rc;
}
-static int mdd_lookup(const struct lu_env *env,
- struct md_object *pobj, const struct lu_name *lname,
- struct lu_fid* fid, struct md_op_spec *spec)
+int mdd_lookup(const struct lu_env *env,
+ struct md_object *pobj, const struct lu_name *lname,
+ struct lu_fid* fid, struct md_op_spec *spec)
{
int rc;
ENTRY;
RETURN(rc);
}
-
static int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj,
struct lu_fid *fid)
{
}
/*
- * For root fid use special function, whcih does not compare version component
- * of fid. Vresion component is different for root fids on all MDTs.
+ * For root fid use special function, which does not compare version component
+ * of fid. Version component is different for root fids on all MDTs.
*/
-static int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
+int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
{
return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
*
* returns < 0: if error
*/
-static int mdd_is_subdir(const struct lu_env *env,
- struct md_object *mo, const struct lu_fid *fid,
- struct lu_fid *sfid)
+int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
+ const struct lu_fid *fid, struct lu_fid *sfid)
{
struct mdd_device *mdd = mdo2mdd(mo);
int rc;
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);
}
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);
int rc = 0;
ENTRY;
+ if (!mdd_object_exists(src_obj))
+ RETURN(-ENOENT);
+
+ if (mdd_is_dead_obj(src_obj))
+ RETURN(-ESTALE);
+
/* Local ops, no lookup before link, check filename length here. */
if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
RETURN(-ENAMETOOLONG);
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)
+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)
{
- struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
- struct mdd_device *m = mdd_obj2mdd_dev(obj);
+ struct dt_object *next = mdd_object_child(pobj);
+ int rc;
+ ENTRY;
- 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);
- }
-}
+ if (dt_try_as_dir(env, next)) {
+ rc = next->do_index_ops->dio_delete(env, next,
+ (struct dt_key *)name,
+ handle, capa);
+ } else
+ rc = -ENOTDIR;
-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);
+ RETURN(rc);
}
-/* insert named index, add reference if isdir */
-static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
- const struct lu_fid *lf, const char *name, int is_dir,
- struct thandle *handle, struct lustre_capa *capa)
+static int __mdd_index_insert_only(const struct lu_env *env,
+ struct mdd_object *pobj,
+ const struct lu_fid *lf, const char *name,
+ struct thandle *handle,
+ struct lustre_capa *capa)
{
struct dt_object *next = mdd_object_child(pobj);
int rc;
struct md_ucred *uc = md_ucred(env);
rc = next->do_index_ops->dio_insert(env, next,
- __mdd_fid_rec(env, lf),
+ (struct dt_rec*)lf,
(const struct dt_key *)name,
handle, capa, uc->mu_cap &
CFS_CAP_SYS_RESOURCE_MASK);
} else {
rc = -ENOTDIR;
}
+ RETURN(rc);
+}
- if (rc == 0) {
- if (is_dir) {
- mdd_write_lock(env, pobj, MOR_TGT_PARENT);
- __mdd_ref_add(env, pobj, handle);
- mdd_write_unlock(env, pobj);
- }
+/* insert named index, add reference if isdir */
+static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
+ const struct lu_fid *lf, const char *name, int is_dir,
+ struct thandle *handle, struct lustre_capa *capa)
+{
+ int rc;
+ ENTRY;
+
+ rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
+ if (rc == 0 && is_dir) {
+ mdd_write_lock(env, pobj, MOR_TGT_PARENT);
+ mdo_ref_add(env, pobj, handle);
+ mdd_write_unlock(env, pobj);
}
RETURN(rc);
}
const char *name, int is_dir, struct thandle *handle,
struct lustre_capa *capa)
{
- struct dt_object *next = mdd_object_child(pobj);
int rc;
ENTRY;
- if (dt_try_as_dir(env, next)) {
- rc = next->do_index_ops->dio_delete(env, next,
- (struct dt_key *)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);
- mdd_write_unlock(env, pobj);
- }
- } else
- rc = -ENOTDIR;
+ rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
+ if (rc == 0 && is_dir) {
+ mdd_write_lock(env, pobj, MOR_TGT_PARENT);
+ mdo_ref_del(env, pobj, handle);
+ mdd_write_unlock(env, pobj);
+ }
RETURN(rc);
}
-static int
-__mdd_index_insert_only(const struct lu_env *env, struct mdd_object *pobj,
- const struct lu_fid *lf, const char *name,
- struct thandle *handle, struct lustre_capa *capa)
+int mdd_declare_llog_record(const struct lu_env *env, struct mdd_device *mdd,
+ int reclen, struct thandle *handle)
{
- struct dt_object *next = mdd_object_child(pobj);
- int rc;
+ 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
+ * want the change to commit without the log entry.
+ * \param target - mdd_object of change
+ * \param parent - parent dir/object
+ * \param tf - target lu_fid, overrides fid of \a target if this is non-null
+ * \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,
+ int flags,
+ struct mdd_object *target,
+ struct mdd_object *parent,
+ const struct lu_fid *tf,
+ const struct lu_name *tname,
+ struct thandle *handle)
+{
+ const struct lu_fid *tfid;
+ const struct lu_fid *tpfid = mdo2fid(parent);
+ struct llog_changelog_rec *rec;
+ struct lu_buf *buf;
+ int reclen;
+ int rc;
ENTRY;
- if (dt_try_as_dir(env, next)) {
- struct md_ucred *uc = md_ucred(env);
+ /* Not recording */
+ if (!(mdd->mdd_cl.mc_flags & CLM_ON))
+ RETURN(0);
+ if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
+ RETURN(0);
- rc = next->do_index_ops->dio_insert(env, next,
- __mdd_fid_rec(env, lf),
- (const struct dt_key *)name,
- handle, capa, uc->mu_cap &
- CFS_CAP_SYS_RESOURCE_MASK);
- } else {
- rc = -ENOTDIR;
+ LASSERT(parent != NULL);
+ LASSERT(tname != NULL);
+ LASSERT(handle != NULL);
+
+ /* target */
+ reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
+ buf = mdd_buf_alloc(env, reclen);
+ if (buf->lb_buf == NULL)
+ RETURN(-ENOMEM);
+ rec = (struct llog_changelog_rec *)buf->lb_buf;
+
+ 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;
+ rec->cr.cr_pfid = *tpfid;
+ rec->cr.cr_namelen = tname->ln_namelen;
+ memcpy(rec->cr.cr_name, tname->ln_name, rec->cr.cr_namelen);
+ if (likely(target))
+ target->mod_cltime = cfs_time_current_64();
+
+ rc = mdd_changelog_llog_write(mdd, rec, handle);
+ if (rc < 0) {
+ CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
+ rc, type, tname->ln_name, PFID(tfid), PFID(tpfid));
+ return -EFAULT;
}
- RETURN(rc);
+
+ 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 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 = 0;
+ int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
#endif
int rc;
ENTRY;
rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA);
if (!rc) {
+ void *data = NULL;
+ mdd_data_get(env, mdd_tobj, &data);
quota_opc = FSFILT_OP_LINK;
mdd_quota_wrapper(la_tmp, qids);
/* get block quota for parent */
- lquota_chkquota(mds_quota_interface_ref, obd,
- qids[USRQUOTA], qids[GRPQUOTA], 1,
- &rec_pending, NULL, LQUOTA_FLAGS_BLK);
+ 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, 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, NULL, lname, handle);
+stop:
mdd_trans_stop(env, mdd, rc, handle);
out_pending:
#ifdef HAVE_QUOTA_SUPPORT
if (quota_opc) {
- if (rec_pending)
- lquota_pending_commit(mds_quota_interface_ref, obd,
- qids[USRQUOTA], qids[GRPQUOTA],
- 1, 1);
+ lquota_pending_commit(mds_quota_interface_ref, obd,
+ qids, rec_pending, 1);
/* Trigger dqacq for the parent owner. If failed,
* the next call for lquota_chkquota will process it. */
lquota_adjust(mds_quota_interface_ref, obd, 0, qids, 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 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;
- rc = mdd_iattr_get(env, obj, ma);
- if (rc == 0 && ma->ma_attr.la_nlink == 0) {
+ LASSERT(mdd_write_locked(env, obj) != 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() */
if (obj->mod_count) {
rc = __mdd_orphan_add(env, obj, th);
if (rc == 0)
- obj->mod_flags |= ORPHAN_OBJ;
- }
-
- obj->mod_flags |= DEAD_OBJ;
- if (!(obj->mod_flags & ORPHAN_OBJ)) {
- rc = mdd_object_kill(env, obj, ma);
+ CDEBUG(D_HA, "Object "DFID" is inserted into "
+ "orphan list, open count = %d\n",
+ PFID(mdd_object_fid(obj)),
+ obj->mod_count);
+ else
+ CERROR("Object "DFID" fail to be an orphan, "
+ "open count = %d, maybe cause failed "
+ "open replay\n",
+ PFID(mdd_object_fid(obj)),
+ obj->mod_count);
+ } else {
+ 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)
unsigned int qpids[MAXQUOTAS] = { 0, 0 };
int quota_opc = 0;
#endif
- int rc, is_dir;
+ int is_dir = S_ISDIR(ma->ma_attr.la_mode);
+ 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);
mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
- is_dir = S_ISDIR(ma->ma_attr.la_mode);
rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma);
if (rc)
GOTO(cleanup, rc);
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
}
}
#endif
+ if (!is_dir)
+ /* old files may not have link ea; ignore errors */
+ mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
+ lname, NULL, NULL, handle);
- if (rc == 0)
- obd_set_info_async(mdd2obd_dev(mdd)->u.mds.mds_osc_exp,
- sizeof(KEY_UNLINKED), KEY_UNLINKED, 0,
- NULL, NULL);
EXIT;
cleanup:
mdd_write_unlock(env, mdd_cobj);
mdd_pdo_write_unlock(env, mdd_pobj, dlh);
out_trans:
+ 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, 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 = 0;
+ int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
cfs_cap_t save = uc->mu_cap;
#endif
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)) {
rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
if (!rc) {
+ void *data = NULL;
+ mdd_data_get(env, mdd_obj, &data);
quota_opc = FSFILT_OP_LINK;
mdd_quota_wrapper(la_tmp, qids);
/* get block quota for parent */
lquota_chkquota(mds_quota_interface_ref, obd,
- qids[USRQUOTA], qids[GRPQUOTA],
- 1, &rec_pending, NULL,
- LQUOTA_FLAGS_BLK);
+ exp, qids, rec_pending, 1, NULL,
+ LQUOTA_FLAGS_BLK, data, 1);
}
} else {
uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK;
}
}
#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);
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota) {
if (quota_opc) {
- if (rec_pending)
- lquota_pending_commit(mds_quota_interface_ref,
- obd, qids[USRQUOTA],
- qids[GRPQUOTA], 1, 1);
+ lquota_pending_commit(mds_quota_interface_ref,
+ obd, qids, rec_pending, 1);
/* Trigger dqacq for the parent owner. If failed,
* the next call for lquota_chkquota will process it*/
lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
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);
* processed in cmr_rename_tgt before mdd_rename_tgt and enable
* MDS_PERM_BYPASS.
* So check may_delete, but not check nlink of tgt_pobj. */
- LASSERT(tobj);
+
rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
RETURN(rc);
}
+/* Partial rename op on slave MDD */
static int mdd_rename_tgt(const struct lu_env *env,
struct md_object *pobj, struct md_object *tobj,
const struct lu_fid *lf, const struct lu_name *lname,
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 quota_opc = 0, rec_pending = 0;
+ int quota_copc = 0, quota_popc = 0;
+ int rec_pending[MAXQUOTAS] = { 0, 0 };
#endif
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;
rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA);
if (!rc) {
- quota_opc = FSFILT_OP_LINK;
+ void *data = NULL;
+ mdd_data_get(env, mdd_tpobj, &data);
+ quota_popc = FSFILT_OP_LINK;
mdd_quota_wrapper(la_tmp, qpids);
/* get block quota for target parent */
- lquota_chkquota(mds_quota_interface_ref, obd,
- qpids[USRQUOTA], qpids[GRPQUOTA], 1,
- &rec_pending, NULL, LQUOTA_FLAGS_BLK);
+ 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);
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota && ma->ma_valid & MA_INODE &&
ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
- quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
+ quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
mdd_quota_wrapper(&ma->ma_attr, qcids);
}
#endif
mdd_write_unlock(env, mdd_tobj);
mdd_pdo_write_unlock(env, mdd_tpobj, dlh);
out_trans:
+ 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, 0, mdd_tobj,
+ mdd_tpobj, NULL, lname, handle);
+
mdd_trans_stop(env, mdd, rc, handle);
out_pending:
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota) {
- if (rec_pending)
+ if (quota_popc)
lquota_pending_commit(mds_quota_interface_ref, obd,
- qpids[USRQUOTA],
- qpids[GRPQUOTA],
- 1, 1);
- if (quota_opc)
- /* Trigger dqrel/dqacq on the target owner of child and
- * parent. If failed, the next call for lquota_chkquota
+ qpids, rec_pending, 1);
+
+ if (quota_copc)
+ /* Trigger dqrel on the target owner of child.
+ * If failed, the next call for lquota_chkquota
* will process it. */
- lquota_adjust(mds_quota_interface_ref, obd, qcids,
- qpids, rc, quota_opc);
+ lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids,
+ rc, quota_copc);
}
#endif
return rc;
}
+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);
}
+/* Get fid from name and parent */
static int
__mdd_lookup(const struct lu_env *env, struct md_object *pobj,
const struct lu_name *lname, struct lu_fid* fid, int mask)
struct mdd_object *mdd_obj = md2mdd_obj(pobj);
struct mdd_device *m = mdo2mdd(pobj);
struct dt_object *dir = mdd_object_child(mdd_obj);
- struct lu_fid_pack *pack = &mdd_env_info(env)->mti_pack;
int rc;
ENTRY;
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. */
if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
dt_try_as_dir(env, dir))) {
+
rc = dir->do_index_ops->dio_lookup(env, dir,
- (struct dt_rec *)pack, key,
+ (struct dt_rec *)fid, key,
mdd_object_capa(env, mdd_obj));
if (rc > 0)
- rc = fid_unpack(pack, fid);
+ rc = 0;
else if (rc == 0)
rc = -ENOENT;
} else
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,
- struct mdd_object *child, struct md_attr *ma,
- struct thandle *handle, const struct md_op_spec *spec)
+ const struct lu_name *lname, struct mdd_object *child,
+ struct md_attr *ma, struct thandle *handle,
+ const struct md_op_spec *spec)
{
int rc;
ENTRY;
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) {
+ if (rc == 0)
rc = __mdd_index_insert_only(env, child, pfid,
dotdot, handle,
BYPASS_CAPA);
- if (rc != 0) {
- int rc2;
-
- rc2 = __mdd_index_delete(env, child, dot, 1,
- handle, BYPASS_CAPA);
- if (rc2 != 0)
- CERROR("Failure to cleanup after dotdot"
- " creation: %d (%d)\n", rc2, rc);
- }
- }
+ if (rc != 0)
+ mdo_ref_del(env, child, handle);
}
+ if (rc == 0)
+ 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 };
int quota_opc = 0, block_count = 0;
- int inode_pending = 0, block_pending = 0, parent_pending = 0;
+ int inode_pending[MAXQUOTAS] = { 0, 0 };
+ int block_pending[MAXQUOTAS] = { 0, 0 };
+ int parent_pending[MAXQUOTAS] = { 0, 0 };
#endif
ENTRY;
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[USRQUOTA], qcids[GRPQUOTA], 1,
- &inode_pending, 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[USRQUOTA], qcids[GRPQUOTA],
- block_count,
- &block_pending, NULL,
- LQUOTA_FLAGS_BLK);
+ exp, qcids, block_pending,
+ block_count, NULL,
+ LQUOTA_FLAGS_BLK, NULL, 0);
if (!same)
lquota_chkquota(mds_quota_interface_ref, obd,
- qpids[USRQUOTA], qpids[GRPQUOTA], 1,
- &parent_pending, NULL,
- LQUOTA_FLAGS_BLK);
+ 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);
}
#endif
- rc = mdd_object_initialize(env, mdo2fid(mdd_pobj),
+ rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
son, ma, handle, spec);
mdd_write_unlock(env, son);
if (rc)
}
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);
}
}
mdd_pdo_write_unlock(env, mdd_pobj, dlh);
out_trans:
+ if (rc == 0)
+ rc = mdd_changelog_ns_store(env, mdd,
+ S_ISDIR(attr->la_mode) ? CL_MKDIR :
+ S_ISREG(attr->la_mode) ? CL_CREATE :
+ S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
+ 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
if (quota_opc) {
- if (inode_pending)
- lquota_pending_commit(mds_quota_interface_ref, obd,
- qcids[USRQUOTA], qcids[GRPQUOTA],
- 1, 0);
- if (block_pending)
- lquota_pending_commit(mds_quota_interface_ref, obd,
- qcids[USRQUOTA], qcids[GRPQUOTA],
- block_count, 1);
- if (parent_pending)
- lquota_pending_commit(mds_quota_interface_ref, obd,
- qpids[USRQUOTA], qpids[GRPQUOTA],
- 1, 1);
+ lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
+ inode_pending, 0);
+ lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
+ block_pending, 1);
+ lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
+ parent_pending, 1);
/* Trigger dqacq on the owner of child and parent. If failed,
* the next call for lquota_chkquota will process it. */
lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
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;
}
* the other case has been processed in cml_rename
* before mdd_rename and enable MDS_PERM_BYPASS. */
LASSERT(sobj);
+
rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0);
if (rc)
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,
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 mdd_object *mdd_spobj = md2mdd_obj(src_pobj);
+ 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;
+ 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);
int is_dir;
- int rc;
+ int rc, rc2;
#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 };
unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
- int quota_opc = 0, rec_pending = 0;
+ int quota_copc = 0, quota_popc = 0;
+ int rec_pending[MAXQUOTAS] = { 0, 0 };
#endif
ENTRY;
rc = mdd_la_get(env, mdd_tpobj, la_tmp,
BYPASS_CAPA);
if (!rc) {
- quota_opc = FSFILT_OP_LINK;
+ void *data = NULL;
+ mdd_data_get(env, mdd_tpobj, &data);
+ quota_popc = FSFILT_OP_LINK;
mdd_quota_wrapper(la_tmp, qtpids);
/* get block quota for target parent */
lquota_chkquota(mds_quota_interface_ref,
- obd, qtpids[USRQUOTA],
- qtpids[GRPQUOTA], 1,
- &rec_pending, NULL,
- LQUOTA_FLAGS_BLK);
+ 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)
GOTO(cleanup, rc);
+ /* Remove source name from source directory */
rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
mdd_object_capa(env, mdd_spobj));
if (rc)
GOTO(cleanup, rc);
/* "mv dir1 dir2" needs "dir1/.." link update */
- if (is_dir) {
- rc = __mdd_index_delete(env, mdd_sobj, dotdot, is_dir, handle,
- mdd_object_capa(env, mdd_spobj));
+ if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
+ rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
+ mdd_object_capa(env, mdd_sobj));
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_spobj2, rc);
- rc = __mdd_index_insert(env, mdd_sobj, tpobj_fid, dotdot,
- is_dir, handle,
- mdd_object_capa(env, mdd_tpobj));
+ rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
+ handle, mdd_object_capa(env, mdd_sobj));
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_spobj, rc);
}
- /*
+ /* Remove target name from target directory
* Here tobj can be remote one, so we do index_delete unconditionally
* and -ENOENT is allowed.
*/
rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
mdd_object_capa(env, mdd_tpobj));
- if (rc != 0 && rc != -ENOENT)
- GOTO(cleanup, rc);
+ if (rc != 0) {
+ if (mdd_tobj) {
+ /* tname might been renamed to something else */
+ GOTO(fixup_spobj, rc);
+ }
+ if (rc != -ENOENT)
+ GOTO(fixup_spobj, rc);
+ }
+ /* Insert new fid with target name into target dir */
rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
mdd_object_capa(env, mdd_tpobj));
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_tpobj, rc);
LASSERT(ma->ma_attr.la_valid & LA_CTIME);
la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
handle, 0);
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_tpobj, rc);
}
- /*
+ /* Remove old target object
* For tobj is remote case cmm layer has processed
* and set tobj to NULL then. So when tobj is NOT NULL,
* it must be local one.
*/
if (tobj && mdd_object_exists(mdd_tobj)) {
mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
- __mdd_ref_del(env, mdd_tobj, handle, 0);
+ if (mdd_is_dead_obj(mdd_tobj)) {
+ mdd_write_unlock(env, mdd_tobj);
+ /* shld not be dead, something is wrong */
+ CERROR("tobj is dead, something is wrong\n");
+ rc = -EINVAL;
+ goto cleanup;
+ }
+ 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)
- GOTO(cleanup, rc);
+ GOTO(fixup_tpobj, rc);
rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
mdd_write_unlock(env, mdd_tobj);
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_tpobj, rc);
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota && ma->ma_valid & MA_INODE &&
ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
- quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
+ quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
mdd_quota_wrapper(&ma->ma_attr, qtcids);
}
#endif
la->la_valid = LA_CTIME | LA_MTIME;
rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_tpobj, rc);
if (mdd_spobj != mdd_tpobj) {
la->la_valid = LA_CTIME | LA_MTIME;
handle, 0);
}
+ 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);
+ if (rc == -ENOENT)
+ /* Old files might not have EA entry */
+ mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
+ 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. */
+ rc = 0;
+ }
+
EXIT;
+
+fixup_tpobj:
+ if (rc) {
+ rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
+ BYPASS_CAPA);
+ if (rc2)
+ CWARN("tp obj fix error %d\n",rc2);
+
+ if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
+ !mdd_is_dead_obj(mdd_tobj)) {
+ rc2 = __mdd_index_insert(env, mdd_tpobj,
+ mdo2fid(mdd_tobj), tname,
+ is_dir, handle,
+ BYPASS_CAPA);
+
+ if (rc2)
+ CWARN("tp obj fix error %d\n",rc2);
+ }
+ }
+
+fixup_spobj:
+ if (rc && is_dir && mdd_sobj) {
+ rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
+ BYPASS_CAPA);
+
+ if (rc2)
+ CWARN("sp obj dotdot delete error %d\n",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);
+ }
+
+fixup_spobj2:
+ if (rc) {
+ rc2 = __mdd_index_insert(env, mdd_spobj,
+ lf, sname, is_dir, handle, BYPASS_CAPA);
+ if (rc2)
+ CWARN("sp obj fix error %d\n",rc2);
+ }
cleanup:
if (likely(tdlh) && sdlh != tdlh)
mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
if (likely(sdlh))
mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
cleanup_unlocked:
+ if (rc == 0)
+ rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, 0, mdd_tobj,
+ mdd_spobj, lf, lsname, 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);
out_pending:
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota) {
- if (rec_pending)
+ if (quota_popc)
lquota_pending_commit(mds_quota_interface_ref, obd,
- qtpids[USRQUOTA],
- qtpids[GRPQUOTA],
- 1, 1);
- /* Trigger dqrel on the source owner of parent.
- * If failed, the next call for lquota_chkquota will
- * process it. */
- lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
- FSFILT_OP_UNLINK_PARTIAL_PARENT);
- if (quota_opc)
- /* Trigger dqrel/dqacq on the target owner of child and
- * parent. If failed, the next call for lquota_chkquota
+ qtpids, rec_pending, 1);
+
+ if (quota_copc) {
+ /* Trigger dqrel on the source owner of parent.
+ * If failed, the next call for lquota_chkquota will
+ * process it. */
+ lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
+ FSFILT_OP_UNLINK_PARTIAL_PARENT);
+
+ /* Trigger dqrel on the target owner of child.
+ * If failed, the next call for lquota_chkquota
* will process it. */
lquota_adjust(mds_quota_interface_ref, obd, qtcids,
- qtpids, rc, quota_opc);
+ qtpids, rc, quota_copc);
+ }
}
#endif
return rc;
}
+/** enable/disable storing of hardlink info */
+int mdd_linkea_enable = 1;
+CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
+ "record hardlink info in EAs");
+
+/** 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 lu_buf *buf;
+ struct lustre_capa *capa;
+ struct link_ea_header *leh;
+ int rc;
+
+ /* First try a small buf */
+ buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
+ if (buf->lb_buf == NULL)
+ return ERR_PTR(-ENOMEM);
+
+ capa = mdd_object_capa(env, mdd_obj);
+ rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
+ if (rc == -ERANGE) {
+ /* Buf was too small, figure out what we need. */
+ mdd_buf_put(buf);
+ rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
+ if (rc < 0)
+ return ERR_PTR(rc);
+ buf = mdd_buf_alloc(env, rc);
+ if (buf->lb_buf == NULL)
+ return ERR_PTR(-ENOMEM);
+ rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
+ }
+ if (rc < 0)
+ return ERR_PTR(rc);
+
+ leh = 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 ERR_PTR(-EINVAL);
+ if (leh->leh_reccount == 0)
+ return ERR_PTR(-ENODATA);
+
+ return buf;
+}
+
+/** 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);
+}
+
+/** Add a record to the end of link ea buf */
+static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
+ const struct lu_fid *pfid,
+ const struct lu_name *lname)
+{
+ struct link_ea_header *leh;
+ struct link_ea_entry *lee;
+ int reclen;
+
+ if (lname == NULL || pfid == NULL)
+ return -EINVAL;
+
+ /* Make sure our buf is big enough for the new one */
+ leh = buf->lb_buf;
+ reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
+ if (leh->leh_len + reclen > buf->lb_len) {
+ if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
+ return -ENOMEM;
+ }
+
+ leh = buf->lb_buf;
+ lee = buf->lb_buf + leh->leh_len;
+ reclen = mdd_lee_pack(lee, lname, pfid);
+ leh->leh_len += reclen;
+ leh->leh_reccount++;
+ 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.
+ */
+#define LINKEA_MAX_COUNT 128
+
+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, int first)
+{
+ struct lu_buf *buf;
+ struct link_ea_header *leh;
+ int rc;
+ ENTRY;
+
+ if (!mdd_linkea_enable)
+ RETURN(0);
+
+ buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj);
+ if (IS_ERR(buf)) {
+ rc = PTR_ERR(buf);
+ if (rc != -ENODATA) {
+ CERROR("link_ea read failed %d "DFID"\n", rc,
+ PFID(mdd_object_fid(mdd_obj)));
+ RETURN (rc);
+ }
+ /* empty EA; start one */
+ buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
+ if (buf->lb_buf == NULL)
+ RETURN(-ENOMEM);
+ leh = buf->lb_buf;
+ leh->leh_magic = LINK_EA_MAGIC;
+ leh->leh_len = sizeof(struct link_ea_header);
+ leh->leh_reccount = 0;
+ }
+
+ leh = buf->lb_buf;
+ if (leh->leh_reccount > LINKEA_MAX_COUNT)
+ RETURN(-EOVERFLOW);
+
+ rc = __mdd_links_add(env, buf, pfid, lname);
+ if (rc)
+ RETURN(rc);
+
+ leh = buf->lb_buf;
+ 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) {
+ 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_len > OBD_ALLOC_BIG)
+ /* if we vmalloced a large buffer drop it */
+ mdd_buf_put(buf);
+
+ RETURN (rc);
+}
+
+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)
+{
+ struct lu_buf *buf;
+ struct link_ea_header *leh;
+ struct link_ea_entry *lee;
+ struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
+ struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
+ int reclen = 0;
+ int count;
+ int rc, rc2 = 0;
+ ENTRY;
+
+ if (!mdd_linkea_enable)
+ RETURN(0);
+
+ if (mdd_obj->mod_flags & DEAD_OBJ)
+ /* No more links, don't bother */
+ RETURN(0);
+
+ buf = mdd_links_get(env, mdd_obj);
+ if (IS_ERR(buf)) {
+ rc = PTR_ERR(buf);
+ 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;
+ lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
+
+ /* Find the old record */
+ for(count = 0; count < leh->leh_reccount; count++) {
+ mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
+ if (tmpname->ln_namelen == oldlname->ln_namelen &&
+ lu_fid_eq(tmpfid, oldpfid) &&
+ (strncmp(tmpname->ln_name, oldlname->ln_name,
+ tmpname->ln_namelen) == 0))
+ break;
+ lee = (struct link_ea_entry *)((char *)lee + reclen);
+ }
+ if ((count + 1) > leh->leh_reccount) {
+ CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
+ oldlname->ln_namelen, oldlname->ln_name);
+ GOTO(out, rc = -ENOENT);
+ }
+
+ /* Remove the old record */
+ leh->leh_reccount--;
+ leh->leh_len -= reclen;
+ memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
+ (char *)lee);
+
+ /* If renaming, add the new record */
+ if (newpfid != NULL) {
+ /* if the add fails, we still delete the out-of-date old link */
+ rc2 = __mdd_links_add(env, buf, newpfid, newlname);
+ leh = buf->lb_buf;
+ }
+
+ rc = __mdd_xattr_set(env, mdd_obj,
+ mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
+ XATTR_NAME_LINK, 0, handle);
+
+out:
+ if (rc == 0)
+ rc = rc2;
+ if (rc)
+ CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
+ oldlname->ln_namelen, oldlname->ln_name, rc,
+ PFID(mdd_object_fid(mdd_obj)));
+
+ if (buf->lb_len > OBD_ALLOC_BIG)
+ /* if we vmalloced a large buffer drop it */
+ mdd_buf_put(buf);
+
+ RETURN (rc);
+}
+
const struct md_dir_operations mdd_dir_ops = {
.mdo_is_subdir = mdd_is_subdir,
.mdo_lookup = mdd_lookup,
.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