* 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, 2014, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Pravin B Shelar <pravin.shelar@sun.com>
*/
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_MDS
#include <obd.h>
#include "mdd_internal.h"
const char orph_index_name[] = "PENDING";
-const char *dotdot = "..";
+static const char dotdot[] = "..";
enum {
ORPH_OP_UNLINK,
static struct dt_key* orph_key_fill(const struct lu_env *env,
const struct lu_fid *lf, __u32 op)
{
- char *key = mdd_env_info(env)->mti_orph_key;
+ char *key = mdd_env_info(env)->mti_key;
int rc;
LASSERT(key);
static struct dt_key* orph_key_fill_18(const struct lu_env *env,
const struct lu_fid *lf)
{
- char *key = mdd_env_info(env)->mti_orph_key;
+ char *key = mdd_env_info(env)->mti_key;
int rc;
LASSERT(key);
return ERR_PTR(rc);
}
-static int orphan_key_to_fid(char *key, struct lu_fid *lf)
-{
- int rc = 0;
- unsigned int op;
-
- rc = sscanf(key, ORPHAN_FILE_NAME_FORMAT,
- (long long unsigned int *)&lf->f_seq, &lf->f_oid,
- &lf->f_ver, &op);
- if (rc == 4)
- return 0;
-
- /* build igif */
- rc = sscanf(key, ORPHAN_FILE_NAME_FORMAT_18,
- (long long unsigned int *)&lf->f_seq, &lf->f_oid);
- if (rc == 2) {
- lf->f_ver = 0;
- return 0;
- }
-
- CERROR("can not parse orphan file name %s\n", key);
- return -EINVAL;
-}
-
static inline void mdd_orphan_write_lock(const struct lu_env *env,
- struct mdd_device *mdd)
+ struct mdd_device *mdd)
{
-
- struct dt_object *dor = mdd->mdd_orphans;
- dor->do_ops->do_write_lock(env, dor, MOR_TGT_ORPHAN);
+ struct dt_object *dor = mdd->mdd_orphans;
+ dt_write_lock(env, dor, MOR_TGT_ORPHAN);
}
static inline void mdd_orphan_write_unlock(const struct lu_env *env,
- struct mdd_device *mdd)
+ struct mdd_device *mdd)
{
-
- struct dt_object *dor = mdd->mdd_orphans;
- dor->do_ops->do_write_unlock(env, dor);
+ struct dt_object *dor = mdd->mdd_orphans;
+ dt_write_unlock(env, dor);
}
static inline int mdd_orphan_insert_obj(const struct lu_env *env,
- struct mdd_device *mdd,
- struct mdd_object *obj,
- __u32 op,
- struct thandle *th)
+ struct mdd_device *mdd,
+ struct mdd_object *obj,
+ __u32 op, struct thandle *th)
{
- struct dt_object *dor = mdd->mdd_orphans;
- const struct lu_fid *lf = mdo2fid(obj);
- struct dt_key *key = orph_key_fill(env, lf, op);
- ENTRY;
+ struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
+ struct dt_object *dor = mdd->mdd_orphans;
+ const struct lu_fid *lf = mdo2fid(obj);
+ struct dt_key *key = orph_key_fill(env, lf, op);
+
+ rec->rec_fid = lf;
+ rec->rec_type = mdd_object_type(obj);
- return dor->do_index_ops->dio_insert(env, dor,
- (struct dt_rec *)lf,
- key, th,
- BYPASS_CAPA, 1);
+ return dt_insert(env, dor, (const struct dt_rec *)rec, key, th, 1);
}
static inline int mdd_orphan_delete_obj(const struct lu_env *env,
struct dt_key *key,
struct thandle *th)
{
- struct dt_object *dor = mdd->mdd_orphans;
+ struct dt_object *dor = mdd->mdd_orphans;
- return dor->do_index_ops->dio_delete(env, dor,
- key, th,
- BYPASS_CAPA);
+ return dt_delete(env, dor, key, th);
}
-static inline void mdd_orphan_ref_add(const struct lu_env *env,
- struct mdd_device *mdd,
- struct thandle *th)
+static inline int mdd_orphan_ref_add(const struct lu_env *env,
+ struct mdd_device *mdd,
+ struct thandle *th)
{
- struct dt_object *dor = mdd->mdd_orphans;
- dor->do_ops->do_ref_add(env, dor, th);
+ struct dt_object *dor = mdd->mdd_orphans;
+ return dt_ref_add(env, dor, th);
}
-static inline void mdd_orphan_ref_del(const struct lu_env *env,
- struct mdd_device *mdd,
- struct thandle *th)
+static inline int mdd_orphan_ref_del(const struct lu_env *env,
+ struct mdd_device *mdd,
+ struct thandle *th)
{
- struct dt_object *dor = mdd->mdd_orphans;
- dor->do_ops->do_ref_del(env, dor, th);
+ struct dt_object *dor = mdd->mdd_orphans;
+ return dt_ref_del(env, dor, th);
}
int orph_declare_index_insert(const struct lu_env *env,
- struct mdd_object *obj,
- struct thandle *th)
+ struct mdd_object *obj,
+ umode_t mode, struct thandle *th)
{
- struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
- int rc;
+ struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
+ struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
+ struct dt_key *key;
+ int rc;
- rc = dt_declare_insert(env, mdd->mdd_orphans, NULL, NULL, th);
- if (rc)
- return rc;
+ key = orph_key_fill(env, mdo2fid(obj), ORPH_OP_UNLINK);
- rc = mdo_declare_ref_add(env, obj, th);
- if (rc)
- return rc;
+ rec->rec_fid = mdo2fid(obj);
+ rec->rec_type = mode;
+ rc = dt_declare_insert(env, mdd->mdd_orphans,
+ (const struct dt_rec *)rec, key, th);
+ if (rc != 0)
+ return rc;
- if (!S_ISDIR(mdd_object_type(obj)))
- return 0;
+ rc = mdo_declare_ref_add(env, obj, th);
+ if (rc)
+ return rc;
- rc = mdo_declare_ref_add(env, obj, th);
- if (rc)
- return rc;
+ if (!S_ISDIR(mode))
+ return 0;
- rc = dt_declare_ref_add(env, mdd->mdd_orphans, th);
- if (rc)
- return rc;
+ rc = mdo_declare_ref_add(env, obj, th);
+ if (rc)
+ return rc;
- rc = mdo_declare_index_delete(env, obj, dotdot, th);
- if (rc)
- return rc;
+ rc = dt_declare_ref_add(env, mdd->mdd_orphans, th);
+ if (rc)
+ return rc;
- rc = mdo_declare_index_insert(env, obj, NULL, dotdot, th);
+ rc = mdo_declare_index_delete(env, obj, dotdot, th);
+ if (rc)
+ return rc;
- return rc;
+ rc = mdo_declare_index_insert(env, obj,
+ lu_object_fid(&mdd->mdd_orphans->do_lu),
+ S_IFDIR, dotdot, th);
+
+ return rc;
}
static int orph_index_insert(const struct lu_env *env,
- struct mdd_object *obj,
- __u32 op,
- struct thandle *th)
+ struct mdd_object *obj,
+ __u32 op, struct thandle *th)
{
- struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
- struct dt_object *dor = mdd->mdd_orphans;
- const struct lu_fid *lf_dor = lu_object_fid(&dor->do_lu);
- struct dt_object *next = mdd_object_child(obj);
- int rc;
+ struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
+ struct dt_object *dor = mdd->mdd_orphans;
+ const struct lu_fid *lf_dor = lu_object_fid(&dor->do_lu);
+ struct dt_object *next = mdd_object_child(obj);
+ struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
+ int rc;
ENTRY;
LASSERT(mdd_write_locked(env, obj) != 0);
LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
- LASSERT(obj->mod_count > 0);
mdd_orphan_write_lock(env, mdd);
mdo_ref_add(env, obj, th);
if (!S_ISDIR(mdd_object_type(obj)))
- goto out;
+ GOTO(out, rc = 0);
mdo_ref_add(env, obj, th);
mdd_orphan_ref_add(env, mdd, th);
/* try best to fixup directory, dont return errors
* from here */
if (!dt_try_as_dir(env, next))
- goto out;
- next->do_index_ops->dio_delete(env, next,
- (const struct dt_key *)dotdot,
- th, BYPASS_CAPA);
+ GOTO(out, rc = 0);
+
+ dt_delete(env, next, (const struct dt_key *)dotdot, th);
- next->do_index_ops->dio_insert(env, next,
- (struct dt_rec *)lf_dor,
- (const struct dt_key *)dotdot,
- th, BYPASS_CAPA, 1);
+ rec->rec_fid = lf_dor;
+ rec->rec_type = S_IFDIR;
+ dt_insert(env, next, (const struct dt_rec *)rec,
+ (const struct dt_key *)dotdot, th, 1);
out:
if (rc == 0)
RETURN(rc);
}
-/**
- * Destroy OSD object on MDD and associated OST objects.
- *
- * \param obj orphan object
- * \param mdd used for sending llog msg to osts
- *
- * \retval 0 success
- * \retval -ve error
- */
-static int orphan_object_kill(const struct lu_env *env,
- struct mdd_object *obj,
- struct mdd_device *mdd,
- struct thandle *th)
-{
- struct lu_attr *la = &mdd_env_info(env)->mti_la;
- int rc = 0;
- ENTRY;
-
- /* No need to lock this object as its recovery phase, and
- * no other thread can access it. But we need to lock it
- * as its precondition for osd api we using. */
-
- mdo_ref_del(env, obj, th);
- if (S_ISDIR(mdd_object_type(obj))) {
- mdo_ref_del(env, obj, th);
- mdd_orphan_ref_del(env, mdd, th);
- } else {
- /* regular file , cleanup linked ost objects */
- rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
- if (rc == 0)
- rc = mdd_lov_destroy(env, mdd, obj, la);
- }
- mdo_destroy(env, obj, th);
- RETURN(rc);
-}
-
int orph_declare_index_delete(const struct lu_env *env,
struct mdd_object *obj,
+
struct thandle *th)
{
struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
+ struct dt_key *key;
int rc;
- rc = dt_declare_delete(env, mdd->mdd_orphans, NULL, th);
+ key = orph_key_fill(env, mdo2fid(obj), ORPH_OP_UNLINK);
+
+ rc = dt_declare_delete(env, mdd->mdd_orphans, key, th);
if (rc)
return rc;
{
struct thandle *th = NULL;
struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
- struct md_attr *ma = &mdd_env_info(env)->mti_ma;
int rc = 0;
ENTRY;
- /* init ma */
- ma->ma_lmm_size = mdd_lov_mdsize(env, mdd);
- ma->ma_lmm = mdd_max_lmm_get(env, mdd);
- ma->ma_cookie_size = mdd_lov_cookiesize(env, mdd);
- ma->ma_cookie = mdd_max_cookie_get(env, mdd);
- ma->ma_need = MA_INODE | MA_LOV | MA_COOKIE;
- ma->ma_valid = 0;
-
- th = mdd_trans_create(env, mdd);
- if (IS_ERR(th)) {
- CERROR("Cannot get thandle\n");
- RETURN(-ENOMEM);
- }
+ th = mdd_trans_create(env, mdd);
+ if (IS_ERR(th)) {
+ CERROR("Cannot get thandle\n");
+ RETURN(PTR_ERR(th));
+ }
+
rc = orph_declare_index_delete(env, obj, th);
if (rc)
GOTO(stop, rc);
- rc = mdd_declare_object_kill(env, obj, ma, th);
+ rc = mdo_declare_destroy(env, obj, th);
if (rc)
GOTO(stop, rc);
if (likely(obj->mod_count == 0)) {
mdd_orphan_write_lock(env, mdd);
rc = mdd_orphan_delete_obj(env, mdd, key, th);
- if (rc == 0)
- orphan_object_kill(env, obj, mdd, th);
- else
+ if (rc == 0) {
+ mdo_ref_del(env, obj, th);
+ if (S_ISDIR(mdd_object_type(obj))) {
+ mdo_ref_del(env, obj, th);
+ mdd_orphan_ref_del(env, mdd, th);
+ }
+ rc = mdo_destroy(env, obj, th);
+ } else
CERROR("could not delete object: rc = %d\n",rc);
mdd_orphan_write_unlock(env, mdd);
}
if (rc) /* so replay-single.sh test_37 works */
CERROR("%s: error unlinking orphan "DFID" from "
"PENDING: rc = %d\n",
- mdd->mdd_obd_dev->obd_name, PFID(lf), rc);
+ mdd2obd_dev(mdd)->obd_name, PFID(lf), rc);
} else {
mdd_write_lock(env, mdo, MOR_TGT_CHILD);
if (likely(mdo->mod_count > 0)) {
* have to be referenced (opened) by some client during recovery, or they
* will be deleted here (for clients that did not complete recovery).
*
- * \param mdd MDD device finishing recovery
+ * \param thread info about orphan cleanup thread
*
* \retval 0 success
* \retval -ve error
*/
static int orph_index_iterate(const struct lu_env *env,
- struct mdd_device *mdd)
+ struct mdd_generic_thread *thread)
{
- struct dt_object *dor = mdd->mdd_orphans;
- char *mti_key = mdd_env_info(env)->mti_orph_key;
- const struct dt_it_ops *iops;
- struct dt_it *it;
- char *key;
- struct lu_fid fid;
- int result = 0;
+ struct mdd_device *mdd = (struct mdd_device *)thread->mgt_data;
+ struct dt_object *dor = mdd->mdd_orphans;
+ struct lu_dirent *ent = &mdd_env_info(env)->mti_ent;
+ const struct dt_it_ops *iops;
+ struct dt_it *it;
+ struct lu_fid fid;
int key_sz = 0;
int rc;
__u64 cookie;
ENTRY;
- /* In recovery phase, do not need for any lock here */
iops = &dor->do_index_ops->dio_it;
- it = iops->init(env, dor, LUDA_64BITHASH, BYPASS_CAPA);
+ it = iops->init(env, dor, LUDA_64BITHASH);
if (IS_ERR(it)) {
rc = PTR_ERR(it);
CERROR("%s: cannot clean PENDING: rc = %d\n",
- mdd->mdd_obd_dev->obd_name, rc);
+ mdd2obd_dev(mdd)->obd_name, rc);
GOTO(out, rc);
}
GOTO(out_put, rc);
if (rc == 0) {
CERROR("%s: error loading iterator to clean PENDING\n",
- mdd->mdd_obd_dev->obd_name);
+ mdd2obd_dev(mdd)->obd_name);
/* Index contains no zero key? */
GOTO(out_put, rc = -EIO);
}
- do {
- key = (void *)iops->key(env, it);
- if (IS_ERR(key)) {
- CERROR("%s: key failed when clean PENDING: rc = %ld\n",
- mdd->mdd_obd_dev->obd_name, PTR_ERR(key));
- goto next;
- }
- key_sz = iops->key_size(env, it);
-
- /* filter out "." and ".." entries from PENDING dir. */
- if (key_sz < 8)
- goto next;
-
- memcpy(mti_key, key, key_sz);
- mti_key[key_sz] = 0;
-
- if (orphan_key_to_fid(mti_key, &fid))
- goto next;
- if (!fid_is_sane(&fid)) {
- CERROR("%s: bad FID "DFID" cleaning PENDING\n",
- mdd->mdd_obd_dev->obd_name, PFID(&fid));
- goto next;
- }
-
- /* kill orphan object */
- cookie = iops->store(env, it);
- iops->put(env, it);
- rc = orph_key_test_and_del(env, mdd, &fid,
- (struct dt_key *)mti_key);
-
- /* after index delete reset iterator */
- if (rc == 0)
- result = iops->get(env, it, (const void *)"");
- else
- result = iops->load(env, it, cookie);
+ do {
+ if (thread->mgt_abort)
+ break;
+
+ key_sz = iops->key_size(env, it);
+ /* filter out "." and ".." entries from PENDING dir. */
+ if (key_sz < 8)
+ goto next;
+
+ rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
+ if (rc != 0) {
+ CERROR("%s: fail to get FID for orphan it: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, rc);
+ goto next;
+ }
+
+ fid_le_to_cpu(&fid, &ent->lde_fid);
+ if (!fid_is_sane(&fid)) {
+ CERROR("%s: bad FID "DFID" cleaning PENDING\n",
+ mdd2obd_dev(mdd)->obd_name, PFID(&fid));
+ goto next;
+ }
+
+ /* kill orphan object */
+ cookie = iops->store(env, it);
+ iops->put(env, it);
+ rc = orph_key_test_and_del(env, mdd, &fid,
+ (struct dt_key *)ent->lde_name);
+
+ /* after index delete reset iterator */
+ if (rc == 0)
+ rc = iops->get(env, it, (const void *)"");
+ else
+ rc = iops->load(env, it, cookie);
next:
- result = iops->next(env, it);
- } while (result == 0);
+ rc = iops->next(env, it);
+ } while (rc == 0);
- GOTO(out_put, rc = 0);
+ GOTO(out_put, rc = 0);
out_put:
- iops->put(env, it);
- iops->fini(env, it);
+ iops->put(env, it);
+ iops->fini(env, it);
out:
- return rc;
+ return rc;
}
/**
*/
int orph_index_init(const struct lu_env *env, struct mdd_device *mdd)
{
- struct lu_fid fid;
- struct dt_object *d;
- int rc = 0;
- ENTRY;
-
- d = dt_store_open(env, mdd->mdd_child, "", orph_index_name, &fid);
- if (!IS_ERR(d)) {
- mdd->mdd_orphans = d;
- if (!dt_try_as_dir(env, d)) {
- rc = -ENOTDIR;
- CERROR("\"%s\" is not an index! : rc = %d\n",
- orph_index_name, rc);
- }
- } else {
- CERROR("cannot find \"%s\" obj %d\n",
- orph_index_name, (int)PTR_ERR(d));
- rc = PTR_ERR(d);
- }
-
- RETURN(rc);
+ struct lu_fid fid;
+ struct dt_object *d;
+ int rc = 0;
+
+ ENTRY;
+
+ /* create PENDING dir */
+ fid_zero(&fid);
+ rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
+ orph_index_name, S_IFDIR | S_IRUGO |
+ S_IWUSR | S_IXUGO, &fid);
+ if (rc < 0)
+ RETURN(rc);
+
+ d = dt_locate(env, mdd->mdd_child, &fid);
+ if (IS_ERR(d))
+ RETURN(PTR_ERR(d));
+ LASSERT(lu_object_exists(&d->do_lu));
+ if (!dt_try_as_dir(env, d)) {
+ CERROR("%s: \"%s\" is not an index: rc = %d\n",
+ mdd2obd_dev(mdd)->obd_name, orph_index_name, rc);
+ lu_object_put(env, &d->do_lu);
+ RETURN(-ENOTDIR);
+ }
+ mdd->mdd_orphans = d;
+ RETURN(0);
}
void orph_index_fini(const struct lu_env *env, struct mdd_device *mdd)
EXIT;
}
+static int __mdd_orphan_cleanup(void *args)
+{
+ struct mdd_generic_thread *thread = (struct mdd_generic_thread *)args;
+ struct lu_env *env = NULL;
+ int rc;
+ ENTRY;
+
+ complete(&thread->mgt_started);
+
+ OBD_ALLOC_PTR(env);
+ if (env == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ rc = lu_env_init(env, LCT_MD_THREAD);
+ if (rc)
+ GOTO(out, rc);
+
+ rc = orph_index_iterate(env, thread);
+
+ lu_env_fini(env);
+ GOTO(out, rc);
+out:
+ if (env)
+ OBD_FREE_PTR(env);
+ complete(&thread->mgt_finished);
+ return rc;
+}
+
/**
* Iterate orphan index to cleanup orphan objects after recovery is done.
* \param d mdd device in recovery.
*/
-int __mdd_orphan_cleanup(const struct lu_env *env, struct mdd_device *d)
+int mdd_orphan_cleanup(const struct lu_env *env, struct mdd_device *d)
{
- return orph_index_iterate(env, d);
+ int rc = -ENOMEM;
+ char *name = NULL;
+
+ OBD_ALLOC(name, MTI_NAME_MAXLEN);
+ if (name == NULL)
+ goto out;
+
+ snprintf(name, MTI_NAME_MAXLEN, "orph_cleanup_%s",
+ mdd2obd_dev(d)->obd_name);
+
+ rc = mdd_generic_thread_start(&d->mdd_orph_cleanup_thread,
+ __mdd_orphan_cleanup, (void *)d, name);
+out:
+ if (rc)
+ CERROR("%s: start orphan cleanup thread failed:%d\n",
+ mdd2obd_dev(d)->obd_name, rc);
+ if (name)
+ OBD_FREE(name, MTI_NAME_MAXLEN);
+ return rc;
}
/**