* Copyright 2009 Sun Microsystems, Inc. All rights reserved
* Use is subject to license terms.
*
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_LOV
+#include <asm/div64.h>
#include <libcfs/libcfs.h>
#include <obd_class.h>
-#include <obd_lov.h>
#include <lustre/lustre_idl.h>
#include "lod_internal.h"
int qos_add_tgt(struct lod_device *lod, struct lod_tgt_desc *ost_desc)
{
- struct lov_qos_oss *oss = NULL, *temposs;
+ struct lod_qos_oss *oss = NULL, *temposs;
struct obd_export *exp = ost_desc->ltd_exp;
int rc = 0, found = 0;
cfs_list_t *list;
int qos_del_tgt(struct lod_device *lod, struct lod_tgt_desc *ost_desc)
{
- struct lov_qos_oss *oss;
+ struct lod_qos_oss *oss;
int rc = 0;
ENTRY;
if (rc && rc != -ENOTCONN)
CERROR("%s: statfs: rc = %d\n", lod2obd(d)->obd_name, rc);
+ /* If the OST is readonly then we can't allocate objects there */
+ if (sfs->os_state & OS_STATE_READONLY)
+ rc = -EROFS;
+
/* check whether device has changed state (active, inactive) */
if (rc != 0 && ost->ltd_active) {
/* turned inactive? */
spin_unlock(&d->lod_desc_lock);
} else if (rc == 0 && ost->ltd_active == 0) {
/* turned active? */
- LASSERT(d->lod_desc.ld_active_tgt_count < d->lod_ostnr);
+ LASSERTF(d->lod_desc.ld_active_tgt_count < d->lod_ostnr,
+ "active tgt count %d, ost nr %d\n",
+ d->lod_desc.ld_active_tgt_count, d->lod_ostnr);
spin_lock(&d->lod_desc_lock);
if (ost->ltd_active == 0) {
ost->ltd_active = 1;
spin_unlock(&d->lod_desc_lock);
}
- return rc;
+ RETURN(rc);
}
static void lod_qos_statfs_update(const struct lu_env *env,
out:
up_write(&lod->lod_qos.lq_rw_sem);
+ EXIT;
}
/* Recalculate per-object penalties for OSSs and OSTs,
depends on size of each ost in an oss */
static int lod_qos_calc_ppo(struct lod_device *lod)
{
- struct lov_qos_oss *oss;
+ struct lod_qos_oss *oss;
__u64 ba_max, ba_min, temp;
__u32 num_active;
int rc, i, prio_wide;
/* per-OST penalty is prio * TGT_bavail / (num_ost - 1) / 2 */
temp >>= 1;
- lov_do_div64(temp, num_active);
+ do_div(temp, num_active);
OST_TGT(lod,i)->ltd_qos.ltq_penalty_per_obj =
(temp * prio_wide) >> 8;
/* Per-OSS penalty is prio * oss_avail / oss_osts / (num_oss - 1) / 2 */
cfs_list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list) {
temp = oss->lqo_bavail >> 1;
- lov_do_div64(temp, oss->lqo_ost_count * num_active);
+ do_div(temp, oss->lqo_ost_count * num_active);
oss->lqo_penalty_per_obj = (temp * prio_wide) >> 8;
age = (now - oss->lqo_used) >> 3;
__u32 index, __u64 *total_wt)
{
struct lod_tgt_desc *ost;
- struct lov_qos_oss *oss;
+ struct lod_qos_oss *oss;
int j;
ENTRY;
#define LOV_QOS_EMPTY ((__u32)-1)
/* compute optimal round-robin order, based on OSTs per OSS */
static int lod_qos_calc_rr(struct lod_device *lod, struct ost_pool *src_pool,
- struct lov_qos_rr *lqr)
+ struct lod_qos_rr *lqr)
{
- struct lov_qos_oss *oss;
+ struct lod_qos_oss *oss;
struct lod_tgt_desc *ost;
unsigned placed, real_count;
int i, rc;
return (msfs->os_bavail < used);
}
-int lod_ea_store_resize(struct lod_thread_info *info, int size);
-
static inline int lod_qos_ost_in_use_clear(const struct lu_env *env, int stripes)
{
struct lod_thread_info *info = lod_env_info(env);
CERROR("can't allocate memory for ost-in-use array\n");
return -ENOMEM;
}
- memset(info->lti_ea_store, 0, sizeof(int) * stripes);
+ memset(info->lti_ea_store, -1, sizeof(int) * stripes);
return 0;
}
struct obd_statfs *sfs = &lod_env_info(env)->lti_osfs;
struct pool_desc *pool = NULL;
struct ost_pool *osts;
- struct lov_qos_rr *lqr;
+ struct lod_qos_rr *lqr;
struct dt_object *o;
unsigned array_idx;
int i, rc;
/*
* try to use another OSP if this one is degraded
*/
- if (sfs->os_state == OS_STATE_DEGRADED && speed < 2) {
+ if (sfs->os_state & OS_STATE_DEGRADED && speed < 2) {
QOS_DEBUG("#%d: degraded\n", ost_idx);
continue;
}
/*
* We've successfuly declared (reserved) an object
*/
+ lod_qos_ost_in_use(env, stripe_num, ost_idx);
stripe[stripe_num] = o;
stripe_num++;
/* stripe count is based on whether OSD can handle larger EA sizes */
if (lod->lod_osd_max_easize > 0)
- max_stripes = lov_mds_md_stripecnt(lod->lod_osd_max_easize,
- magic);
+ max_stripes = lov_mds_md_max_stripe_count(
+ lod->lod_osd_max_easize, magic);
return (stripe_count < max_stripes) ? stripe_count : max_stripes;
}
struct lod_object *mo,
const struct lu_buf *buf)
{
- struct lod_device *d = lu2lod_dev(lod2lu_obj(mo)->lo_dev);
struct lov_mds_md_v1 *v1 = buf->lb_buf;
struct lov_mds_md_v3 *v3 = buf->lb_buf;
struct lov_ost_data_v1 *objs;
__u32 magic;
- int rc;
+ int rc = 0;
ENTRY;
- rc = lod_verify_striping(d, buf, 1);
- if (rc)
- RETURN(rc);
-
magic = le32_to_cpu(v1->lmm_magic);
if (magic == LOV_MAGIC_V1_DEF) {
+ magic = LOV_MAGIC_V1;
objs = &v1->lmm_objects[0];
} else if (magic == LOV_MAGIC_V3_DEF) {
+ magic = LOV_MAGIC_V3;
objs = &v3->lmm_objects[0];
lod_object_set_pool(mo, v3->lmm_pool_name);
} else {
GOTO(out, rc = -EINVAL);
}
- /*
- * LOD shouldn't be aware of recovery at all,
- * but this track recovery status (to some extent)
- * to be do additional checks like this one
- */
- LASSERT(d->lod_recovery_completed == 0);
-
+ mo->ldo_pattern = le32_to_cpu(v1->lmm_pattern);
mo->ldo_stripe_size = le32_to_cpu(v1->lmm_stripe_size);
mo->ldo_stripenr = le16_to_cpu(v1->lmm_stripe_count);
mo->ldo_layout_gen = le16_to_cpu(v1->lmm_layout_gen);
+
+ /* fixup for released file before object initialization */
+ if (mo->ldo_pattern & LOV_PATTERN_F_RELEASED) {
+ mo->ldo_released_stripenr = mo->ldo_stripenr;
+ mo->ldo_stripenr = 0;
+ }
+
LASSERT(buf->lb_len >= lov_mds_md_size(mo->ldo_stripenr, magic));
- rc = lod_initialize_objects(env, mo, objs);
+ if (mo->ldo_stripenr > 0)
+ rc = lod_initialize_objects(env, mo, objs);
out:
RETURN(rc);
RETURN(-EINVAL);
}
- if (v1->lmm_pattern != 0 && v1->lmm_pattern != LOV_PATTERN_RAID0) {
+ v1->lmm_magic = magic;
+ if (v1->lmm_pattern == 0)
+ v1->lmm_pattern = LOV_PATTERN_RAID0;
+ if (lov_pattern(v1->lmm_pattern) != LOV_PATTERN_RAID0) {
CERROR("invalid pattern: %x\n", v1->lmm_pattern);
RETURN(-EINVAL);
}
+ lo->ldo_pattern = v1->lmm_pattern;
if (v1->lmm_stripe_size)
lo->ldo_stripe_size = v1->lmm_stripe_size;
} else
lod_object_set_pool(lo, NULL);
+ /* fixup for released file */
+ if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED) {
+ lo->ldo_released_stripenr = lo->ldo_stripenr;
+ lo->ldo_stripenr = 0;
+ }
+
RETURN(0);
}
if (rc)
GOTO(out, rc);
+ /* A released file is being created */
+ if (lo->ldo_stripenr == 0)
+ GOTO(out, rc = 0);
+
if (likely(lo->ldo_stripe == NULL)) {
/*
* no striping has been created so far
*/
LASSERT(lo->ldo_stripenr > 0);
+ /*
+ * statfs and check OST targets now, since ld_active_tgt_count
+ * could be changed if some OSTs are [de]activated manually.
+ */
+ lod_qos_statfs_update(env, d);
lo->ldo_stripenr = lod_get_stripecnt(d, LOV_MAGIC,
lo->ldo_stripenr);
lod_getref(&d->lod_ost_descs);
/* XXX: support for non-0 files w/o objects */
+ CDEBUG(D_OTHER, "tgt_count %d stripenr %d\n",
+ d->lod_desc.ld_tgt_count, stripe_len);
if (lo->ldo_def_stripe_offset >= d->lod_desc.ld_tgt_count) {
- lod_qos_statfs_update(env, d);
rc = lod_alloc_qos(env, lo, stripe, flag, th);
if (rc == -EAGAIN)
rc = lod_alloc_rr(env, lo, stripe, flag, th);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff