* Copyright 2009 Sun Microsystems, Inc. All rights reserved
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * 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"
#define TGT_BAVAIL(i) (OST_TGT(lod,i)->ltd_statfs.os_bavail * \
OST_TGT(lod,i)->ltd_statfs.os_bsize)
-int qos_add_tgt(struct lod_device *lod, struct lod_ost_desc *ost_desc)
+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;
+ struct list_head *list;
ENTRY;
down_write(&lod->lod_qos.lq_rw_sem);
* but there is no official API to access information like this
* with OSD API.
*/
- cfs_list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list) {
+ list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list) {
if (obd_uuid_equals(&oss->lqo_uuid,
&exp->exp_connection->c_remote_uuid)) {
found++;
sizeof(oss->lqo_uuid));
} else {
/* Assume we have to move this one */
- cfs_list_del(&oss->lqo_oss_list);
+ list_del(&oss->lqo_oss_list);
}
oss->lqo_ost_count++;
/* Add sorted by # of OSTs. Find the first entry that we're
bigger than... */
list = &lod->lod_qos.lq_oss_list;
- cfs_list_for_each_entry(temposs, list, lqo_oss_list) {
+ list_for_each_entry(temposs, list, lqo_oss_list) {
if (oss->lqo_ost_count > temposs->lqo_ost_count)
break;
}
/* ...and add before it. If we're the first or smallest, temposs
points to the list head, and we add to the end. */
- cfs_list_add_tail(&oss->lqo_oss_list, &temposs->lqo_oss_list);
+ list_add_tail(&oss->lqo_oss_list, &temposs->lqo_oss_list);
lod->lod_qos.lq_dirty = 1;
lod->lod_qos.lq_rr.lqr_dirty = 1;
RETURN(rc);
}
-int qos_del_tgt(struct lod_device *lod, struct lod_ost_desc *ost_desc)
+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 (oss->lqo_ost_count == 0) {
CDEBUG(D_QOS, "removing OSS %s\n",
obd_uuid2str(&oss->lqo_uuid));
- cfs_list_del(&oss->lqo_oss_list);
+ list_del(&oss->lqo_oss_list);
ost_desc->ltd_qos.ltq_oss = NULL;
OBD_FREE_PTR(oss);
}
static int lod_statfs_and_check(const struct lu_env *env, struct lod_device *d,
int index, struct obd_statfs *sfs)
{
- struct lod_ost_desc *ost;
+ struct lod_tgt_desc *ost;
int rc;
LASSERT(d);
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,
{
struct obd_device *obd = lod2obd(lod);
struct ost_pool *osts = &(lod->lod_pool_info);
- int i, idx, rc = 0;
+ unsigned int i;
+ int idx, rc = 0;
__u64 max_age, avail;
ENTRY;
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;
+ unsigned int i;
+ int rc, prio_wide;
time_t now, age;
ENTRY;
GOTO(out, rc = -EAGAIN);
/* find bavail on each OSS */
- cfs_list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list)
+ list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list)
oss->lqo_bavail = 0;
lod->lod_qos.lq_active_oss_count = 0;
/* 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) {
+ 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;
static int lod_qos_used(struct lod_device *lod, struct ost_pool *osts,
__u32 index, __u64 *total_wt)
{
- struct lod_ost_desc *ost;
- struct lov_qos_oss *oss;
- int j;
+ struct lod_tgt_desc *ost;
+ struct lod_qos_oss *oss;
+ unsigned int j;
ENTRY;
ost = OST_TGT(lod,index);
lod->lod_qos.lq_active_oss_count;
/* Decrease all OSS penalties */
- cfs_list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list) {
+ list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list) {
if (oss->lqo_penalty < oss->lqo_penalty_per_obj)
oss->lqo_penalty = 0;
else
#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_ost_desc *ost;
+ struct lod_qos_oss *oss;
+ struct lod_tgt_desc *ost;
unsigned placed, real_count;
- int i, rc;
+ unsigned int i;
+ int rc;
ENTRY;
if (!lqr->lqr_dirty) {
/* Place all the OSTs from 1 OSS at the same time. */
placed = 0;
- cfs_list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list) {
+ list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list) {
int j = 0;
for (i = 0; i < lqr->lqr_pool.op_count; i++) {
*/
static struct dt_object *lod_qos_declare_object_on(const struct lu_env *env,
struct lod_device *d,
- int ost_idx,
+ __u32 ost_idx,
struct thandle *th)
{
- struct lod_ost_desc *ost;
+ struct lod_tgt_desc *ost;
struct lu_object *o, *n;
struct lu_device *nd;
struct dt_object *dt;
ENTRY;
LASSERT(d);
- LASSERT(ost_idx >= 0);
LASSERT(ost_idx < d->lod_osts_size);
ost = OST_TGT(d,ost_idx);
LASSERT(ost);
RETURN(dt);
}
-static int min_stripe_count(int stripe_cnt, int flags)
+static int min_stripe_count(__u32 stripe_cnt, int flags)
{
return (flags & LOV_USES_DEFAULT_STRIPE ?
stripe_cnt - (stripe_cnt / 4) : stripe_cnt);
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)
+static inline int lod_qos_ost_in_use_clear(const struct lu_env *env,
+ __u32 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;
}
osts[idx] = ost;
}
-static int lod_qos_is_ost_used(const struct lu_env *env, int ost, int stripes)
+static int lod_qos_is_ost_used(const struct lu_env *env, int ost, __u32 stripes)
{
struct lod_thread_info *info = lod_env_info(env);
int *osts = info->lti_ea_store;
- int j;
+ __u32 j;
for (j = 0; j < stripes; j++) {
if (osts[j] == ost)
/* Allocate objects on OSTs with round-robin algorithm */
static int lod_alloc_rr(const struct lu_env *env, struct lod_object *lo,
- int flags, struct thandle *th)
+ struct dt_object **stripe, int flags,
+ struct thandle *th)
{
struct lod_device *m = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
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;
- int ost_start_idx_temp;
+ unsigned int i, array_idx;
+ int rc;
+ __u32 ost_start_idx_temp;
int speed = 0;
- int stripe_idx = 0;
- int stripe_cnt = lo->ldo_stripenr;
- int stripe_cnt_min = min_stripe_count(stripe_cnt, flags);
+ __u32 stripe_idx = 0;
+ __u32 stripe_cnt = lo->ldo_stripenr;
+ __u32 stripe_cnt_min = min_stripe_count(stripe_cnt, flags);
__u32 ost_idx;
ENTRY;
/*
* 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_idx, ost_idx);
- lo->ldo_stripe[stripe_idx] = o;
+ stripe[stripe_idx] = o;
stripe_idx++;
}
/* alloc objects on osts with specific stripe offset */
static int lod_alloc_specific(const struct lu_env *env, struct lod_object *lo,
- int flags, struct thandle *th)
+ struct dt_object **stripe, int flags,
+ struct thandle *th)
{
struct lod_device *m = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
struct obd_statfs *sfs = &lod_env_info(env)->lti_osfs;
struct dt_object *o;
- unsigned ost_idx, array_idx, ost_count;
- int i, rc, stripe_num = 0;
+ __u32 ost_idx;
+ unsigned int i, array_idx, ost_count;
+ int rc, stripe_num = 0;
int speed = 0;
struct pool_desc *pool = NULL;
struct ost_pool *osts;
/*
* We've successfuly declared (reserved) an object
*/
- lo->ldo_stripe[stripe_num] = o;
+ lod_qos_ost_in_use(env, stripe_num, ost_idx);
+ stripe[stripe_num] = o;
stripe_num++;
/* We have enough stripes */
- network resources (shared OSS's)
*/
static int lod_alloc_qos(const struct lu_env *env, struct lod_object *lo,
- int flags, struct thandle *th)
+ struct dt_object **stripe, int flags,
+ struct thandle *th)
{
struct lod_device *m = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
struct obd_statfs *sfs = &lod_env_info(env)->lti_osfs;
- struct lod_ost_desc *ost;
+ struct lod_tgt_desc *ost;
struct dt_object *o;
__u64 total_weight = 0;
- int nfound, good_osts, i, rc = 0;
- int stripe_cnt = lo->ldo_stripenr;
- int stripe_cnt_min;
+ unsigned int i;
+ int rc = 0;
+ __u32 nfound, good_osts;
+ __u32 stripe_cnt = lo->ldo_stripenr;
+ __u32 stripe_cnt_min;
struct pool_desc *pool = NULL;
struct ost_pool *osts;
ENTRY;
/* On average, this will hit larger-weighted osts more often.
0-weight osts will always get used last (only when rand=0) */
for (i = 0; i < osts->op_count; i++) {
- int idx = osts->op_array[i];
+ __u32 idx = osts->op_array[i];
if (!cfs_bitmap_check(m->lod_ost_bitmap, idx))
continue;
idx, (int) PTR_ERR(o));
continue;
}
- lo->ldo_stripe[nfound++] = o;
+ stripe[nfound++] = o;
lod_qos_used(m, osts, idx, &total_weight);
rc = 0;
break;
*/
LCONSOLE_INFO("wanted %d, found %d\n", stripe_cnt, nfound);
for (i = 0; i < nfound; i++) {
- LASSERT(lo->ldo_stripe[i]);
- lu_object_put(env, &lo->ldo_stripe[i]->do_lu);
- lo->ldo_stripe[i] = NULL;
+ LASSERT(stripe[i] != NULL);
+ lu_object_put(env, &stripe[i]->do_lu);
+ stripe[i] = NULL;
}
/* makes sense to rebalance next time */
/* 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);
v1 = buf->lb_buf;
magic = v1->lmm_magic;
- if (magic == __swab32(LOV_USER_MAGIC_V1))
+ if (magic == __swab32(LOV_USER_MAGIC_V1)) {
lustre_swab_lov_user_md_v1(v1);
- else if (magic == __swab32(LOV_USER_MAGIC_V3))
+ magic = v1->lmm_magic;
+ } else if (magic == __swab32(LOV_USER_MAGIC_V3)) {
+ v3 = buf->lb_buf;
lustre_swab_lov_user_md_v3(v3);
+ magic = v3->lmm_magic;
+ }
if (unlikely(magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)) {
/* try to use as fully defined striping */
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;
v3 = buf->lb_buf;
lod_object_set_pool(lo, v3->lmm_pool_name);
+ /* In the function below, .hs_keycmp resolves to
+ * pool_hashkey_keycmp() */
+ /* coverity[overrun-buffer-val] */
pool = lod_find_pool(d, v3->lmm_pool_name);
if (pool != NULL) {
if (lo->ldo_def_stripe_offset !=
} 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);
}
struct lu_attr *attr, const struct lu_buf *buf,
struct thandle *th)
{
- struct lod_device *d = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
- int flag = LOV_USES_ASSIGNED_STRIPE;
- int i, rc = 0;
+ struct lod_device *d = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
+ struct dt_object **stripe;
+ int stripe_len;
+ int flag = LOV_USES_ASSIGNED_STRIPE;
+ int i, rc;
ENTRY;
LASSERT(lo);
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);
- i = sizeof(struct dt_object *) * lo->ldo_stripenr;
- OBD_ALLOC(lo->ldo_stripe, i);
- if (lo->ldo_stripe == NULL)
+
+ stripe_len = lo->ldo_stripenr;
+ OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
+ if (stripe == NULL)
GOTO(out, rc = -ENOMEM);
- lo->ldo_stripes_allocated = lo->ldo_stripenr;
- lod_getref(d);
+ 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, flag, th);
+ rc = lod_alloc_qos(env, lo, stripe, flag, th);
if (rc == -EAGAIN)
- rc = lod_alloc_rr(env, lo, flag, th);
- } else
- rc = lod_alloc_specific(env, lo, flag, th);
- lod_putref(d);
+ rc = lod_alloc_rr(env, lo, stripe, flag, th);
+ } else {
+ rc = lod_alloc_specific(env, lo, stripe, flag, th);
+ }
+ lod_putref(d, &d->lod_ost_descs);
+
+ if (rc < 0) {
+ for (i = 0; i < stripe_len; i++)
+ if (stripe[i] != NULL)
+ lu_object_put(env, &stripe[i]->do_lu);
+
+ OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
+ lo->ldo_stripenr = 0;
+ } else {
+ lo->ldo_stripe = stripe;
+ lo->ldo_stripes_allocated = stripe_len;
+ }
} else {
/*
* lod_qos_parse_config() found supplied buf as a predefined