* GPL HEADER END
*/
/*
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2013, Intel Corporation.
*/
/*
* lustre/obdclass/local_storage.c
#include "local_storage.h"
/* all initialized local storages on this node are linked on this */
-static CFS_LIST_HEAD(ls_list_head);
+static struct list_head ls_list_head = LIST_HEAD_INIT(ls_list_head);
static DEFINE_MUTEX(ls_list_mutex);
static int ls_object_init(const struct lu_env *env, struct lu_object *o,
{
struct ls_device *ls, *ret = NULL;
- cfs_list_for_each_entry(ls, &ls_list_head, ls_linkage) {
+ list_for_each_entry(ls, &ls_list_head, ls_linkage) {
if (ls->ls_osd == dev) {
- cfs_atomic_inc(&ls->ls_refcount);
+ atomic_inc(&ls->ls_refcount);
ret = ls;
break;
}
if (ls == NULL)
GOTO(out_ls, ls = ERR_PTR(-ENOMEM));
- cfs_atomic_set(&ls->ls_refcount, 1);
- CFS_INIT_LIST_HEAD(&ls->ls_los_list);
+ atomic_set(&ls->ls_refcount, 1);
+ INIT_LIST_HEAD(&ls->ls_los_list);
mutex_init(&ls->ls_los_mutex);
ls->ls_osd = dev;
ls->ls_top_dev.dd_lu_dev.ld_site = dev->dd_lu_dev.ld_site;
/* finally add ls to the list */
- cfs_list_add(&ls->ls_linkage, &ls_list_head);
+ list_add(&ls->ls_linkage, &ls_list_head);
out_ls:
mutex_unlock(&ls_list_mutex);
RETURN(ls);
void ls_device_put(const struct lu_env *env, struct ls_device *ls)
{
LASSERT(env);
- if (!cfs_atomic_dec_and_test(&ls->ls_refcount))
+ if (!atomic_dec_and_test(&ls->ls_refcount))
return;
mutex_lock(&ls_list_mutex);
- if (cfs_atomic_read(&ls->ls_refcount) == 0) {
- LASSERT(cfs_list_empty(&ls->ls_los_list));
- cfs_list_del(&ls->ls_linkage);
+ if (atomic_read(&ls->ls_refcount) == 0) {
+ LASSERT(list_empty(&ls->ls_los_list));
+ list_del(&ls->ls_linkage);
lu_site_purge(env, ls->ls_top_dev.dd_lu_dev.ld_site, ~0);
lu_device_fini(&ls->ls_top_dev.dd_lu_dev);
OBD_FREE_PTR(ls);
mutex_lock(&los->los_id_lock);
fid->f_seq = los->los_seq;
- fid->f_oid = los->los_last_oid++;
+ fid->f_oid = ++los->los_last_oid;
fid->f_ver = 0;
mutex_unlock(&los->los_id_lock);
/* update fid generation file */
if (los != NULL) {
LASSERT(dt_object_exists(los->los_obj));
+ dti->dti_lb.lb_buf = NULL;
+ dti->dti_lb.lb_len = sizeof(struct los_ondisk);
rc = dt_declare_record_write(env, los->los_obj,
- sizeof(struct los_ondisk), 0, th);
+ &dti->dti_lb, 0, th);
if (rc)
RETURN(rc);
}
struct dt_object_format *dof, struct thandle *th)
{
struct dt_thread_info *dti = dt_info(env);
- struct los_ondisk losd;
+ obd_id lastid;
int rc;
ENTRY;
/* update local oid number on disk so that
* we know the last one used after reboot */
- losd.lso_magic = cpu_to_le32(LOS_MAGIC);
- losd.lso_next_oid = cpu_to_le32(los->los_last_oid);
+ lastid = cpu_to_le64(los->los_last_oid);
dti->dti_off = 0;
- dti->dti_lb.lb_buf = &losd;
- dti->dti_lb.lb_len = sizeof(losd);
+ dti->dti_lb.lb_buf = &lastid;
+ dti->dti_lb.lb_len = sizeof(lastid);
rc = dt_record_write(env, los->los_obj, &dti->dti_lb, &dti->dti_off,
th);
mutex_unlock(&los->los_id_lock);
const char *name, struct lu_attr *attr,
struct dt_object_format *dof)
{
- struct dt_thread_info *dti = dt_info(env);
+ struct dt_thread_info *dti = dt_info(env);
+ struct lu_object_conf *conf = &dti->dti_conf;
+ struct dt_insert_rec *rec = &dti->dti_dt_rec;
struct dt_object *dto;
struct thandle *th;
int rc;
- dto = ls_locate(env, ls, fid);
+ /* We know that the target object does not exist, to be created,
+ * then give some hints - LOC_F_NEW to help low layer to handle
+ * that efficiently and properly. */
+ memset(conf, 0, sizeof(*conf));
+ conf->loc_flags = LOC_F_NEW;
+ dto = ls_locate(env, ls, fid, conf);
if (unlikely(IS_ERR(dto)))
RETURN(dto);
dt_declare_ref_add(env, parent, th);
}
- rc = dt_declare_insert(env, parent, (void *)fid, (void *)name, th);
+ rec->rec_fid = fid;
+ rec->rec_type = dto->do_lu.lo_header->loh_attr;
+ rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
+ (const struct dt_key *)name, th);
if (rc)
GOTO(trans_stop, rc);
if (dti->dti_dof.dof_type == DFT_DIR) {
if (!dt_try_as_dir(env, dto))
GOTO(destroy, rc = -ENOTDIR);
+
+ rec->rec_type = S_IFDIR;
+ rec->rec_fid = fid;
/* Add "." and ".." for newly created dir */
- rc = dt_insert(env, dto, (void *)fid, (void *)".", th,
- BYPASS_CAPA, 1);
- if (rc)
+ rc = dt_insert(env, dto, (const struct dt_rec *)rec,
+ (const struct dt_key *)".", th, BYPASS_CAPA, 1);
+ if (rc != 0)
GOTO(destroy, rc);
+
dt_ref_add(env, dto, th);
- rc = dt_insert(env, dto, (void *)lu_object_fid(&parent->do_lu),
- (void *)"..", th, BYPASS_CAPA, 1);
- if (rc)
+ rec->rec_fid = lu_object_fid(&parent->do_lu);
+ rc = dt_insert(env, dto, (const struct dt_rec *)rec,
+ (const struct dt_key *)"..", th, BYPASS_CAPA, 1);
+ if (rc != 0)
GOTO(destroy, rc);
}
+ rec->rec_fid = fid;
+ rec->rec_type = dto->do_lu.lo_header->loh_attr;
dt_write_lock(env, parent, 0);
- rc = dt_insert(env, parent, (const struct dt_rec *)fid,
+ rc = dt_insert(env, parent, (const struct dt_rec *)rec,
(const struct dt_key *)name, th, BYPASS_CAPA, 1);
if (dti->dti_dof.dof_type == DFT_DIR)
dt_ref_add(env, parent, th);
rc = dt_lookup_dir(env, parent, name, &dti->dti_fid);
if (rc == 0)
/* name is found, get the object */
- dto = ls_locate(env, dt2ls_dev(los->los_dev), &dti->dti_fid);
+ dto = ls_locate(env, dt2ls_dev(los->los_dev),
+ &dti->dti_fid, NULL);
else if (rc != -ENOENT)
dto = ERR_PTR(rc);
else {
rc = dt_lookup_dir(env, parent, name, &dti->dti_fid);
if (rc == 0) {
/* name is found, get the object */
- dto = ls_locate(env, dt2ls_dev(los->los_dev), &dti->dti_fid);
+ dto = ls_locate(env, dt2ls_dev(los->los_dev),
+ &dti->dti_fid, NULL);
} else if (rc != -ENOENT) {
dto = ERR_PTR(rc);
} else {
rc = dt_ref_del(env, dto, th);
if (rc < 0) {
- rc = dt_insert(env, parent,
- (const struct dt_rec *)&dti->dti_fid,
+ struct dt_insert_rec *rec = &dti->dti_dt_rec;
+
+ rec->rec_fid = &dti->dti_fid;
+ rec->rec_type = dto->do_lu.lo_header->loh_attr;
+ rc = dt_insert(env, parent, (const struct dt_rec *)rec,
(const struct dt_key *)name, th, BYPASS_CAPA, 1);
GOTO(unlock, rc);
}
{
struct local_oid_storage *los, *ret = NULL;
- cfs_list_for_each_entry(los, &ls->ls_los_list, los_list) {
+ list_for_each_entry(los, &ls->ls_los_list, los_list) {
if (los->los_seq == seq) {
- cfs_atomic_inc(&los->los_refcount);
+ atomic_inc(&los->los_refcount);
ret = los;
break;
}
void dt_los_put(struct local_oid_storage *los)
{
- if (cfs_atomic_dec_and_test(&los->los_refcount))
+ if (atomic_dec_and_test(&los->los_refcount))
/* should never happen, only local_oid_storage_fini should
* drop refcount to zero */
LBUG();
return;
}
+/* after Lustre 2.3 release there may be old file to store last generated FID
+ * If such file exists then we have to read its content
+ */
+int lastid_compat_check(const struct lu_env *env, struct dt_device *dev,
+ __u64 lastid_seq, __u32 *first_oid, struct ls_device *ls)
+{
+ struct dt_thread_info *dti = dt_info(env);
+ struct dt_object *root = NULL;
+ struct los_ondisk losd;
+ struct dt_object *o = NULL;
+ int rc = 0;
+
+ rc = dt_root_get(env, dev, &dti->dti_fid);
+ if (rc)
+ return rc;
+
+ root = ls_locate(env, ls, &dti->dti_fid, NULL);
+ if (IS_ERR(root))
+ return PTR_ERR(root);
+
+ /* find old last_id file */
+ snprintf(dti->dti_buf, sizeof(dti->dti_buf), "seq-"LPX64"-lastid",
+ lastid_seq);
+ rc = dt_lookup_dir(env, root, dti->dti_buf, &dti->dti_fid);
+ lu_object_put_nocache(env, &root->do_lu);
+ if (rc == -ENOENT) {
+ /* old llog lastid accessed by FID only */
+ if (lastid_seq != FID_SEQ_LLOG)
+ return 0;
+ dti->dti_fid.f_seq = FID_SEQ_LLOG;
+ dti->dti_fid.f_oid = 1;
+ dti->dti_fid.f_ver = 0;
+ o = ls_locate(env, ls, &dti->dti_fid, NULL);
+ if (IS_ERR(o))
+ return PTR_ERR(o);
+
+ if (!dt_object_exists(o)) {
+ lu_object_put_nocache(env, &o->do_lu);
+ return 0;
+ }
+ CDEBUG(D_INFO, "Found old llog lastid file\n");
+ } else if (rc < 0) {
+ return rc;
+ } else {
+ CDEBUG(D_INFO, "Found old lastid file for sequence "LPX64"\n",
+ lastid_seq);
+ o = ls_locate(env, ls, &dti->dti_fid, NULL);
+ if (IS_ERR(o))
+ return PTR_ERR(o);
+ }
+ /* let's read seq-NNNNNN-lastid file value */
+ LASSERT(dt_object_exists(o));
+ dti->dti_off = 0;
+ dti->dti_lb.lb_buf = &losd;
+ dti->dti_lb.lb_len = sizeof(losd);
+ dt_read_lock(env, o, 0);
+ rc = dt_record_read(env, o, &dti->dti_lb, &dti->dti_off);
+ dt_read_unlock(env, o);
+ if (rc == 0 && le32_to_cpu(losd.lso_magic) != LOS_MAGIC) {
+ CERROR("%s: wrong content of seq-"LPX64"-lastid file, magic %x\n",
+ o->do_lu.lo_dev->ld_obd->obd_name, lastid_seq,
+ le32_to_cpu(losd.lso_magic));
+ rc = -EINVAL;
+ } else if (rc < 0) {
+ CERROR("%s: failed to read seq-"LPX64"-lastid: rc = %d\n",
+ o->do_lu.lo_dev->ld_obd->obd_name, lastid_seq, rc);
+ }
+ lu_object_put_nocache(env, &o->do_lu);
+ if (rc == 0)
+ *first_oid = le32_to_cpu(losd.lso_next_oid);
+ return rc;
+}
+
/**
* Initialize local OID storage for required sequence.
* That may be needed for services that uses local files and requires
{
struct dt_thread_info *dti = dt_info(env);
struct ls_device *ls;
- struct los_ondisk losd;
- struct dt_object *root = NULL;
+ obd_id lastid;
struct dt_object *o = NULL;
struct thandle *th;
- int rc;
+ __u32 first_oid = fid_oid(first_fid);
+ int rc = 0;
ENTRY;
if (*los == NULL)
GOTO(out, rc = -ENOMEM);
- cfs_atomic_set(&(*los)->los_refcount, 1);
+ atomic_set(&(*los)->los_refcount, 1);
mutex_init(&(*los)->los_id_lock);
(*los)->los_dev = &ls->ls_top_dev;
- cfs_atomic_inc(&ls->ls_refcount);
- cfs_list_add(&(*los)->los_list, &ls->ls_los_list);
-
- rc = dt_root_get(env, dev, &dti->dti_fid);
- if (rc)
- GOTO(out_los, rc);
-
- root = ls_locate(env, ls, &dti->dti_fid);
- if (IS_ERR(root))
- GOTO(out_los, rc = PTR_ERR(root));
-
- /* initialize data allowing to generate new fids,
- * literally we need a sequence */
- snprintf(dti->dti_buf, sizeof(dti->dti_buf), "seq-%Lx-lastid",
- fid_seq(first_fid));
- rc = dt_lookup_dir(env, root, dti->dti_buf, &dti->dti_fid);
- if (rc == -ENOENT)
- dti->dti_fid = *first_fid;
- else if (rc < 0)
- GOTO(out_los, rc);
-
- o = ls_locate(env, ls, &dti->dti_fid);
+ atomic_inc(&ls->ls_refcount);
+ list_add(&(*los)->los_list, &ls->ls_los_list);
+
+ /* Use {seq, 0, 0} to create the LAST_ID file for every
+ * sequence. OIDs start at LUSTRE_FID_INIT_OID.
+ */
+ dti->dti_fid.f_seq = fid_seq(first_fid);
+ dti->dti_fid.f_oid = LUSTRE_FID_LASTID_OID;
+ dti->dti_fid.f_ver = 0;
+ o = ls_locate(env, ls, &dti->dti_fid, NULL);
if (IS_ERR(o))
GOTO(out_los, rc = PTR_ERR(o));
- LASSERT(fid_seq(&dti->dti_fid) == fid_seq(first_fid));
+
if (!dt_object_exists(o)) {
- LASSERT(rc == -ENOENT);
+ rc = lastid_compat_check(env, dev, fid_seq(first_fid),
+ &first_oid, ls);
+ if (rc < 0)
+ GOTO(out_los, rc);
th = dt_trans_create(env, dev);
if (IS_ERR(th))
- GOTO(out_lock, rc = PTR_ERR(th));
+ GOTO(out_los, rc = PTR_ERR(th));
dti->dti_attr.la_valid = LA_MODE | LA_TYPE;
dti->dti_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
if (rc)
GOTO(out_trans, rc);
- rc = dt_declare_insert(env, root,
- (const struct dt_rec *)&dti->dti_fid,
- (const struct dt_key *)dti->dti_buf,
- th);
- if (rc)
- GOTO(out_trans, rc);
+ lastid = cpu_to_le64(first_oid);
- rc = dt_declare_record_write(env, o, sizeof(losd), 0, th);
+ dti->dti_off = 0;
+ dti->dti_lb.lb_buf = &lastid;
+ dti->dti_lb.lb_len = sizeof(lastid);
+ rc = dt_declare_record_write(env, o, &dti->dti_lb, dti->dti_off,
+ th);
if (rc)
GOTO(out_trans, rc);
if (rc)
GOTO(out_trans, rc);
- dt_write_lock(env, root, 0);
dt_write_lock(env, o, 0);
if (dt_object_exists(o))
GOTO(out_lock, rc = 0);
if (rc)
GOTO(out_lock, rc);
- losd.lso_magic = cpu_to_le32(LOS_MAGIC);
- losd.lso_next_oid = cpu_to_le32(fid_oid(first_fid) + 1);
-
- dti->dti_off = 0;
- dti->dti_lb.lb_buf = &losd;
- dti->dti_lb.lb_len = sizeof(losd);
rc = dt_record_write(env, o, &dti->dti_lb, &dti->dti_off, th);
if (rc)
GOTO(out_lock, rc);
-#if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 3, 90, 0)
-#error "fix this before release"
-#endif
- /*
- * there is one technical debt left in Orion:
- * proper hanlding of named vs no-name objects.
- * Llog objects have name always as they are placed in O/d/...
- */
- if (fid_seq(&dti->dti_fid) != FID_SEQ_LLOG) {
- rc = dt_insert(env, root,
- (const struct dt_rec *)&dti->dti_fid,
- (const struct dt_key *)dti->dti_buf,
- th, BYPASS_CAPA, 1);
- if (rc)
- GOTO(out_lock, rc);
- }
out_lock:
dt_write_unlock(env, o);
- dt_write_unlock(env, root);
out_trans:
dt_trans_stop(env, dev, th);
} else {
dti->dti_off = 0;
- dti->dti_lb.lb_buf = &losd;
- dti->dti_lb.lb_len = sizeof(losd);
+ dti->dti_lb.lb_buf = &lastid;
+ dti->dti_lb.lb_len = sizeof(lastid);
dt_read_lock(env, o, 0);
rc = dt_record_read(env, o, &dti->dti_lb, &dti->dti_off);
dt_read_unlock(env, o);
- if (rc == 0 && le32_to_cpu(losd.lso_magic) != LOS_MAGIC) {
- CERROR("local storage file "DFID" is corrupted\n",
- PFID(first_fid));
+ if (rc == 0 && le64_to_cpu(lastid) > OBIF_MAX_OID) {
+ CERROR("%s: bad oid "LPU64" is read from LAST_ID\n",
+ o->do_lu.lo_dev->ld_obd->obd_name,
+ le64_to_cpu(lastid));
rc = -EINVAL;
}
}
out_los:
- if (root != NULL && !IS_ERR(root))
- lu_object_put_nocache(env, &root->do_lu);
-
if (rc != 0) {
- cfs_list_del(&(*los)->los_list);
- cfs_atomic_dec(&ls->ls_refcount);
+ list_del(&(*los)->los_list);
+ atomic_dec(&ls->ls_refcount);
OBD_FREE_PTR(*los);
*los = NULL;
if (o != NULL && !IS_ERR(o))
lu_object_put_nocache(env, &o->do_lu);
} else {
(*los)->los_seq = fid_seq(first_fid);
- (*los)->los_last_oid = le32_to_cpu(losd.lso_next_oid);
+ (*los)->los_last_oid = le64_to_cpu(lastid);
(*los)->los_obj = o;
+ /* Read value should not be less than initial one
+ * but possible after upgrade from older fs.
+ * In this case just switch to the first_oid in memory and
+ * it will be updated on disk with first object generated */
+ if ((*los)->los_last_oid < first_oid)
+ (*los)->los_last_oid = first_oid;
}
out:
mutex_unlock(&ls->ls_los_mutex);
EXPORT_SYMBOL(local_oid_storage_init);
void local_oid_storage_fini(const struct lu_env *env,
- struct local_oid_storage *los)
+ struct local_oid_storage *los)
{
struct ls_device *ls;
- if (!cfs_atomic_dec_and_test(&los->los_refcount))
- return;
-
LASSERT(env);
LASSERT(los->los_dev);
ls = dt2ls_dev(los->los_dev);
+ /* Take the mutex before decreasing the reference to avoid race
+ * conditions as described in LU-4721. */
mutex_lock(&ls->ls_los_mutex);
- if (cfs_atomic_read(&los->los_refcount) == 0) {
- if (los->los_obj)
- lu_object_put_nocache(env, &los->los_obj->do_lu);
- cfs_list_del(&los->los_list);
- OBD_FREE_PTR(los);
+ if (!atomic_dec_and_test(&los->los_refcount)) {
+ mutex_unlock(&ls->ls_los_mutex);
+ return;
}
+
+ if (los->los_obj)
+ lu_object_put_nocache(env, &los->los_obj->do_lu);
+ list_del(&los->los_list);
+ OBD_FREE_PTR(los);
mutex_unlock(&ls->ls_los_mutex);
ls_device_put(env, ls);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff