#define DEBUG_SUBSYSTEM S_MDS
-#include <linux/module.h>
-#include <ldiskfs/ldiskfs_jbd2.h>
-#include <obd.h>
#include <obd_class.h>
-#include <lustre_ver.h>
-#include <obd_support.h>
#include <lprocfs_status.h>
-
#include <lustre_fid.h>
-#include <ldiskfs/ldiskfs.h>
#include <lustre_mds.h>
-#include <lustre/lustre_idl.h>
#include <lustre_disk.h> /* for changelogs */
#include <lustre_param.h>
#include <lustre_fid.h>
static const char mdd_root_dir_name[] = "ROOT";
static const char mdd_obf_dir_name[] = "fid";
+/* Slab for MDD object allocation */
+cfs_mem_cache_t *mdd_object_kmem;
+
+static struct lu_kmem_descr mdd_caches[] = {
+ {
+ .ckd_cache = &mdd_object_kmem,
+ .ckd_name = "mdd_obj",
+ .ckd_size = sizeof(struct mdd_object)
+ },
+ {
+ .ckd_cache = NULL
+ }
+};
+
static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
const char *name, struct lu_device *next)
{
struct lu_device *d)
{
struct mdd_device *mdd = lu2mdd_dev(d);
- struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
int rc;
rc = mdd_procfs_fini(mdd);
CERROR("proc fini error %d \n", rc);
return ERR_PTR(rc);
}
- return next;
+ return NULL;
}
static void mdd_changelog_fini(const struct lu_env *env,
EXIT;
}
-static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
- void *data)
+static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
{
struct mdd_device *mdd = (struct mdd_device *)data;
struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
RETURN(LLOG_PROC_BREAK);
}
-static int changelog_user_init_cb(struct llog_handle *llh,
- struct llog_rec_hdr *hdr, void *data)
+static int changelog_user_init_cb(const struct lu_env *env,
+ struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
{
struct mdd_device *mdd = (struct mdd_device *)data;
struct llog_changelog_user_rec *rec =
return -EINVAL;
}
- rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
+ rc = llog_cat_reverse_process(NULL, ctxt->loc_handle,
+ changelog_init_cb, mdd);
llog_ctxt_put(ctxt);
if (rc < 0) {
return -EINVAL;
}
- rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
- mdd);
+ rc = llog_cat_reverse_process(NULL, ctxt->loc_handle,
+ changelog_user_init_cb, mdd);
llog_ctxt_put(ctxt);
if (rc < 0) {
return -ENXIO;
/* nested journal transaction */
- rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
+ rc = llog_add(NULL, ctxt, &rec->cr_hdr, NULL, NULL, 0);
llog_ctxt_put(ctxt);
return rc;
}
+/** Add a changelog_ext entry \a rec to the changelog llog
+ * \param mdd
+ * \param rec
+ * \param handle - currently ignored since llogs start their own transaction;
+ * this will hopefully be fixed in llog rewrite
+ * \retval 0 ok
+ */
+int mdd_changelog_ext_llog_write(struct mdd_device *mdd,
+ struct llog_changelog_ext_rec *rec,
+ struct thandle *handle)
+{
+ struct obd_device *obd = mdd2obd_dev(mdd);
+ struct llog_ctxt *ctxt;
+ int rc;
+
+ rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
+ /* llog_lvfs_write_rec sets the llog tail len */
+ rec->cr_hdr.lrh_type = CHANGELOG_REC;
+ rec->cr.cr_time = cl_time();
+ cfs_spin_lock(&mdd->mdd_cl.mc_lock);
+ /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
+ * but as long as the MDD transactions are ordered correctly for e.g.
+ * rename conflicts, I don't think this should matter. */
+ rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
+ cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
+ ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
+ if (ctxt == NULL)
+ return -ENXIO;
+
+ /* nested journal transaction */
+ rc = llog_add(NULL, ctxt, &rec->cr_hdr, NULL, NULL, 0);
+ llog_ctxt_put(ctxt);
+
+ return rc;
+}
+
/** Remove entries with indicies up to and including \a endrec from the
* changelog
* \param mdd
* \param endrec
* \retval 0 ok
*/
-int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
+int mdd_changelog_llog_cancel(const struct lu_env *env,
+ struct mdd_device *mdd, long long endrec)
{
struct obd_device *obd = mdd2obd_dev(mdd);
struct llog_ctxt *ctxt;
mdd->mdd_cl.mc_starttime = cfs_time_current_64();
/* XXX: transaction is started by llog itself */
- rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
+ rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
out:
llog_ctxt_put(ctxt);
return rc;
return 0;
}
-static int dot_lustre_mdd_attr_get(const struct lu_env *env,
- struct md_object *obj, struct md_attr *ma)
-{
- struct mdd_object *mdd_obj = md2mdd_obj(obj);
-
- return mdd_attr_get_internal_locked(env, mdd_obj, ma);
-}
-
-static int dot_lustre_mdd_attr_set(const struct lu_env *env,
- struct md_object *obj,
- const struct md_attr *ma)
-{
- return -EPERM;
-}
-
static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
struct md_object *obj, struct lu_buf *buf,
const char *name)
static struct md_object_operations mdd_dot_lustre_obj_ops = {
.moo_permission = dot_lustre_mdd_permission,
- .moo_attr_get = dot_lustre_mdd_attr_get,
- .moo_attr_set = dot_lustre_mdd_attr_set,
+ .moo_attr_get = mdd_attr_get,
+ .moo_attr_set = mdd_attr_set,
.moo_xattr_get = dot_lustre_mdd_xattr_get,
.moo_xattr_list = dot_lustre_mdd_xattr_list,
.moo_xattr_set = dot_lustre_mdd_xattr_set,
/* "fid" is a virtual object and hence does not have any "real"
* attributes. So we reuse attributes of .lustre for "fid" dir */
ma->ma_need |= MA_INODE;
- rc = dot_lustre_mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj,
- ma);
+ rc = mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
if (rc)
return rc;
ma->ma_valid |= MA_INODE;
return -EPERM;
}
+static int obf_xattr_list(const struct lu_env *env,
+ struct md_object *obj, struct lu_buf *buf)
+{
+ return 0;
+}
+
static int obf_xattr_get(const struct lu_env *env,
struct md_object *obj, struct lu_buf *buf,
const char *name)
return 0;
}
+static int obf_xattr_set(const struct lu_env *env,
+ struct md_object *obj,
+ const struct lu_buf *buf, const char *name,
+ int fl)
+{
+ return -EPERM;
+}
+
+static int obf_xattr_del(const struct lu_env *env,
+ struct md_object *obj,
+ const char *name)
+{
+ return -EPERM;
+}
+
static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
int flags)
{
}
static struct md_object_operations mdd_obf_obj_ops = {
- .moo_attr_get = obf_attr_get,
- .moo_attr_set = obf_attr_set,
- .moo_xattr_get = obf_xattr_get,
- .moo_open = obf_mdd_open,
- .moo_close = obf_mdd_close,
- .moo_readpage = obf_mdd_readpage,
- .moo_path = obf_path
+ .moo_attr_get = obf_attr_get,
+ .moo_attr_set = obf_attr_set,
+ .moo_xattr_list = obf_xattr_list,
+ .moo_xattr_get = obf_xattr_get,
+ .moo_xattr_set = obf_xattr_set,
+ .moo_xattr_del = obf_xattr_del,
+ .moo_open = obf_mdd_open,
+ .moo_close = obf_mdd_close,
+ .moo_readpage = obf_mdd_readpage,
+ .moo_path = obf_path
};
/**
sscanf(name, SFID, RFID(f));
if (!fid_is_sane(f)) {
- CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
- (__u64)1, 2, 0);
+ CWARN("%s: bad FID format [%s], should be "DFID"\n",
+ mdd->mdd_obd_dev->obd_name, lname->ln_name,
+ (__u64)FID_SEQ_NORMAL, 1, 0);
GOTO(out, rc = -EINVAL);
}
+ if (!fid_is_norm(f)) {
+ CWARN("%s: "DFID" is invalid, sequence should be "
+ ">= "LPX64"\n", mdd->mdd_obd_dev->obd_name, PFID(f),
+ (__u64)FID_SEQ_NORMAL);
+ GOTO(out, rc = -EINVAL);
+ }
+
/* Check if object with this fid exists */
child = mdd_object_find(env, mdd, f);
if (child == NULL)
mdd_changelog_init(env, m);
break;
case LCFG_CLEANUP:
+ lu_dev_del_linkage(d->ld_site, d);
mdd_device_shutdown(env, m, cfg);
default:
rc = next->ld_ops->ldo_process_config(env, next, cfg);
struct lu_device *lu)
{
struct mdd_device *m = lu2mdd_dev(lu);
- struct lu_device *next = &m->mdd_child->dd_lu_dev;
ENTRY;
LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
md_device_fini(&m->mdd_md_dev);
OBD_FREE_PTR(m);
- RETURN(next);
+ RETURN(NULL);
}
static struct obd_ops mdd_obd_device_ops = {
rec->cur_endrec = mdd->mdd_cl.mc_index;
cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
- rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
+ rc = llog_add(NULL, ctxt, &rec->cur_hdr, NULL, NULL, 0);
CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
out:
__u32 mcud_usercount;
int mcud_found:1;
struct mdd_device *mcud_mdd;
- const struct lu_env *mcud_env;
};
#define MCUD_UNREGISTER -1LL
* 1. Find the smallest record everyone is willing to purge
* 2. Update the last purgeable record for this user
*/
-static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
- struct llog_rec_hdr *hdr, void *data)
+static int mdd_changelog_user_purge_cb(const struct lu_env *env,
+ struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
{
- struct llog_changelog_user_rec *rec;
- struct mdd_changelog_user_data *mcud =
- (struct mdd_changelog_user_data *)data;
- int rc;
+ struct llog_changelog_user_rec *rec;
+ struct mdd_changelog_user_data *mcud = data;
+ int rc;
+
ENTRY;
LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
/* XXX This is a workaround for the deadlock of changelog
* adding vs. changelog cancelling. LU-81. */
- th = mdd_trans_create(mcud->mcud_env, mdd);
+ th = mdd_trans_create(env, mdd);
if (IS_ERR(th)) {
CERROR("Cannot get thandle\n");
RETURN(-ENOMEM);
}
- rc = mdd_declare_llog_cancel(mcud->mcud_env, mdd, th);
+ rc = mdd_declare_llog_cancel(env, mdd, th);
if (rc)
GOTO(stop, rc);
- rc = mdd_trans_start(mcud->mcud_env, mdd, th);
+ rc = mdd_trans_start(env, mdd, th);
if (rc)
GOTO(stop, rc);
- rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
- 1, &cookie);
+ rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
+ 1, &cookie);
if (rc == 0)
mcud->mcud_usercount--;
stop:
- mdd_trans_stop(mcud->mcud_env, mdd, 0, th);
+ mdd_trans_stop(env, mdd, 0, th);
RETURN(rc);
}
/* hdr+1 is loc of data */
hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
- rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
- hdr->lrh_index);
+ rc = llog_write_rec(env, llh, hdr, NULL, 0, (void *)(hdr + 1),
+ hdr->lrh_index);
RETURN(rc);
}
data.mcud_usercount = 0;
data.mcud_endrec = endrec;
data.mcud_mdd = mdd;
- data.mcud_env = env;
cfs_spin_lock(&mdd->mdd_cl.mc_lock);
endrec = mdd->mdd_cl.mc_index;
cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
return -ENXIO;
LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
- rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
- (void *)&data, 0, 0);
+ rc = llog_cat_process(env, ctxt->loc_handle,
+ mdd_changelog_user_purge_cb, (void *)&data,
+ 0, 0);
if ((rc >= 0) && (data.mcud_minrec > 0)) {
CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
", referenced by "CHANGELOG_USER_PREFIX"%d\n",
data.mcud_minrec, data.mcud_minid);
- rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
+ rc = mdd_changelog_llog_cancel(env, mdd, data.mcud_minrec);
} else {
CWARN("Could not determine changelog records to purge; rc=%d\n",
rc);
*mntopts = mdd->mdd_dt_conf.ddp_mntopts;
RETURN(0);
}
+ case OBD_IOC_START_LFSCK: {
+ struct lfsck_start *start = karg;
+ struct md_lfsck *lfsck = &mdd->mdd_lfsck;
+
+ /* Return the kernel service version. */
+ /* XXX: version can be used for compatibility in the future. */
+ start->ls_version = lfsck->ml_version;
+ rc = mdd_lfsck_start(env, lfsck, start);
+ RETURN(rc);
+ }
+ case OBD_IOC_STOP_LFSCK: {
+ rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck);
+ RETURN(rc);
+ }
}
/* Below ioctls use obd_ioctl_data */
.llod_feat = &dt_directory_features,
};
-static struct lu_local_obj_desc llod_lfsck_bookmark_key = {
+static struct lu_local_obj_desc llod_lfsck_bookmark = {
.llod_name = lfsck_bookmark_name,
.llod_oid = LFSCK_BOOKMARK_OID,
.llod_is_index = 0,
static int __init mdd_mod_init(void)
{
struct lprocfs_static_vars lvars;
+ int rc;
+
lprocfs_mdd_init_vars(&lvars);
+ rc = lu_kmem_init(mdd_caches);
+ if (rc)
+ return rc;
+
llo_local_obj_register(&llod_capa_key);
llo_local_obj_register(&llod_mdd_orphan);
llo_local_obj_register(&llod_mdd_root);
- llo_local_obj_register(&llod_lfsck_bookmark_key);
+ llo_local_obj_register(&llod_lfsck_bookmark);
- return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
- LUSTRE_MDD_NAME, &mdd_device_type);
+ rc = class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
+ LUSTRE_MDD_NAME, &mdd_device_type);
+ if (rc)
+ lu_kmem_fini(mdd_caches);
+ return rc;
}
static void __exit mdd_mod_exit(void)
llo_local_obj_unregister(&llod_capa_key);
llo_local_obj_unregister(&llod_mdd_orphan);
llo_local_obj_unregister(&llod_mdd_root);
- llo_local_obj_unregister(&llod_lfsck_bookmark_key);
+ llo_local_obj_unregister(&llod_lfsck_bookmark);
class_unregister_type(LUSTRE_MDD_NAME);
+ lu_kmem_fini(mdd_caches);
}
MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
git://git.whamcloud.com / fs / lustre-release.git / blobdiff