X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Flod%2Flod_qos.c;h=fa7eed8a1b34e0b0a132109e32b3b5d6788b2aaf;hb=c944e6eaa7f8b98230701266641bfd9dfcb56da8;hp=216555a328e0eeb8f8bd1ebc9ac1dd4fd3450bf4;hpb=349de1e30b10ffb8501e6e23dda75963baf8c063;p=fs%2Flustre-release.git diff --git a/lustre/lod/lod_qos.c b/lustre/lod/lod_qos.c index 216555a..fa7eed8 100644 --- a/lustre/lod/lod_qos.c +++ b/lustre/lod/lod_qos.c @@ -23,7 +23,7 @@ * 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/ @@ -59,7 +59,7 @@ #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 obd_export *exp = ost_desc->ltd_exp; @@ -67,7 +67,7 @@ int qos_add_tgt(struct lod_device *lod, struct lod_ost_desc *ost_desc) cfs_list_t *list; ENTRY; - cfs_down_write(&lod->lod_qos.lq_rw_sem); + down_write(&lod->lod_qos.lq_rw_sem); /* * a bit hacky approach to learn NID of corresponding connection * but there is no official API to access information like this @@ -114,17 +114,17 @@ int qos_add_tgt(struct lod_device *lod, struct lod_ost_desc *ost_desc) lod->lod_qos.lq_rr.lqr_dirty = 1; out: - cfs_up_write(&lod->lod_qos.lq_rw_sem); + up_write(&lod->lod_qos.lq_rw_sem); 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; int rc = 0; ENTRY; - cfs_down_write(&lod->lod_qos.lq_rw_sem); + down_write(&lod->lod_qos.lq_rw_sem); oss = ost_desc->ltd_qos.ltq_oss; if (!oss) GOTO(out, rc = -ENOENT); @@ -141,14 +141,14 @@ int qos_del_tgt(struct lod_device *lod, struct lod_ost_desc *ost_desc) lod->lod_qos.lq_dirty = 1; lod->lod_qos.lq_rr.lqr_dirty = 1; out: - cfs_up_write(&lod->lod_qos.lq_rw_sem); + up_write(&lod->lod_qos.lq_rw_sem); RETURN(rc); } 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); @@ -162,7 +162,7 @@ static int lod_statfs_and_check(const struct lu_env *env, struct lod_device *d, /* check whether device has changed state (active, inactive) */ if (rc != 0 && ost->ltd_active) { /* turned inactive? */ - cfs_spin_lock(&d->lod_desc_lock); + spin_lock(&d->lod_desc_lock); if (ost->ltd_active) { ost->ltd_active = 0; LASSERT(d->lod_desc.ld_active_tgt_count > 0); @@ -172,11 +172,13 @@ static int lod_statfs_and_check(const struct lu_env *env, struct lod_device *d, CDEBUG(D_CONFIG, "%s: turns inactive\n", ost->ltd_exp->exp_obd->obd_name); } - cfs_spin_unlock(&d->lod_desc_lock); + 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); - cfs_spin_lock(&d->lod_desc_lock); + 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; d->lod_desc.ld_active_tgt_count++; @@ -185,10 +187,10 @@ static int lod_statfs_and_check(const struct lu_env *env, struct lod_device *d, CDEBUG(D_CONFIG, "%s: turns active\n", ost->ltd_exp->exp_obd->obd_name); } - cfs_spin_unlock(&d->lod_desc_lock); + spin_unlock(&d->lod_desc_lock); } - return rc; + RETURN(rc); } static void lod_qos_statfs_update(const struct lu_env *env, @@ -206,7 +208,7 @@ static void lod_qos_statfs_update(const struct lu_env *env, /* statfs data are quite recent, don't need to refresh it */ RETURN_EXIT; - cfs_down_write(&lod->lod_qos.lq_rw_sem); + down_write(&lod->lod_qos.lq_rw_sem); if (cfs_time_beforeq_64(max_age, obd->obd_osfs_age)) GOTO(out, rc = 0); @@ -224,7 +226,8 @@ static void lod_qos_statfs_update(const struct lu_env *env, obd->obd_osfs_age = cfs_time_current_64(); out: - cfs_up_write(&lod->lod_qos.lq_rw_sem); + up_write(&lod->lod_qos.lq_rw_sem); + EXIT; } /* Recalculate per-object penalties for OSSs and OSTs, @@ -363,7 +366,7 @@ static int lod_qos_calc_weight(struct lod_device *lod, int i) static int lod_qos_used(struct lod_device *lod, struct ost_pool *osts, __u32 index, __u64 *total_wt) { - struct lod_ost_desc *ost; + struct lod_tgt_desc *ost; struct lov_qos_oss *oss; int j; ENTRY; @@ -443,7 +446,7 @@ static int lod_qos_calc_rr(struct lod_device *lod, struct ost_pool *src_pool, struct lov_qos_rr *lqr) { struct lov_qos_oss *oss; - struct lod_ost_desc *ost; + struct lod_tgt_desc *ost; unsigned placed, real_count; int i, rc; ENTRY; @@ -454,7 +457,7 @@ static int lod_qos_calc_rr(struct lod_device *lod, struct ost_pool *src_pool, } /* Do actual allocation. */ - cfs_down_write(&lod->lod_qos.lq_rw_sem); + down_write(&lod->lod_qos.lq_rw_sem); /* * Check again. While we were sleeping on @lq_rw_sem something could @@ -462,7 +465,7 @@ static int lod_qos_calc_rr(struct lod_device *lod, struct ost_pool *src_pool, */ if (!lqr->lqr_dirty) { LASSERT(lqr->lqr_pool.op_size); - cfs_up_write(&lod->lod_qos.lq_rw_sem); + up_write(&lod->lod_qos.lq_rw_sem); RETURN(0); } @@ -475,7 +478,7 @@ static int lod_qos_calc_rr(struct lod_device *lod, struct ost_pool *src_pool, lqr->lqr_pool.op_count = real_count; rc = lod_ost_pool_extend(&lqr->lqr_pool, real_count); if (rc) { - cfs_up_write(&lod->lod_qos.lq_rw_sem); + up_write(&lod->lod_qos.lq_rw_sem); RETURN(rc); } for (i = 0; i < lqr->lqr_pool.op_count; i++) @@ -510,7 +513,7 @@ static int lod_qos_calc_rr(struct lod_device *lod, struct ost_pool *src_pool, } lqr->lqr_dirty = 0; - cfs_up_write(&lod->lod_qos.lq_rw_sem); + up_write(&lod->lod_qos.lq_rw_sem); if (placed != real_count) { /* This should never happen */ @@ -549,7 +552,7 @@ static struct dt_object *lod_qos_declare_object_on(const struct lu_env *env, int 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; @@ -613,7 +616,7 @@ static int inline lod_qos_dev_is_full(struct obd_statfs *msfs) /* the minimum of 0.1% used blocks and 1GB bytes. */ used = min_t(__u64, (msfs->os_blocks - msfs->os_bfree) >> 10, - 1 << (31 - cfs_ffs(bs))); + 1 << (31 - ffs(bs))); return (msfs->os_bavail < used); } @@ -629,7 +632,7 @@ static inline int lod_qos_ost_in_use_clear(const struct lu_env *env, int stripes 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; } @@ -657,7 +660,8 @@ static int lod_qos_is_ost_used(const struct lu_env *env, int ost, int stripes) /* 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; @@ -679,7 +683,7 @@ static int lod_alloc_rr(const struct lu_env *env, struct lod_object *lo, pool = lod_find_pool(m, lo->ldo_pool); if (pool != NULL) { - cfs_down_read(&pool_tgt_rw_sem(pool)); + down_read(&pool_tgt_rw_sem(pool)); osts = &(pool->pool_obds); lqr = &(pool->pool_rr); } else { @@ -709,7 +713,7 @@ static int lod_alloc_rr(const struct lu_env *env, struct lod_object *lo, if (stripe_cnt > 1 && (osts->op_count % stripe_cnt) != 1) ++lqr->lqr_offset_idx; } - cfs_down_read(&m->lod_qos.lq_rw_sem); + down_read(&m->lod_qos.lq_rw_sem); ost_start_idx_temp = lqr->lqr_start_idx; repeat_find: @@ -790,7 +794,7 @@ repeat_find: * 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++; } @@ -801,7 +805,7 @@ repeat_find: goto repeat_find; } - cfs_up_read(&m->lod_qos.lq_rw_sem); + up_read(&m->lod_qos.lq_rw_sem); if (stripe_idx) { lo->ldo_stripenr = stripe_idx; @@ -814,7 +818,7 @@ repeat_find: out: if (pool != NULL) { - cfs_up_read(&pool_tgt_rw_sem(pool)); + up_read(&pool_tgt_rw_sem(pool)); /* put back ref got by lod_find_pool() */ lod_pool_putref(pool); } @@ -824,7 +828,8 @@ out: /* 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; @@ -844,7 +849,7 @@ static int lod_alloc_specific(const struct lu_env *env, struct lod_object *lo, pool = lod_find_pool(m, lo->ldo_pool); if (pool != NULL) { - cfs_down_read(&pool_tgt_rw_sem(pool)); + down_read(&pool_tgt_rw_sem(pool)); osts = &(pool->pool_obds); } else { osts = &(m->lod_pool_info); @@ -916,7 +921,8 @@ repeat_find: /* * 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 */ @@ -941,7 +947,7 @@ repeat_find: rc = -EFBIG; out: if (pool != NULL) { - cfs_up_read(&pool_tgt_rw_sem(pool)); + up_read(&pool_tgt_rw_sem(pool)); /* put back ref got by lod_find_pool() */ lod_pool_putref(pool); } @@ -971,11 +977,12 @@ static inline int lod_qos_is_usable(struct lod_device *lod) - 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; @@ -993,7 +1000,7 @@ static int lod_alloc_qos(const struct lu_env *env, struct lod_object *lo, pool = lod_find_pool(m, lo->ldo_pool); if (pool != NULL) { - cfs_down_read(&pool_tgt_rw_sem(pool)); + down_read(&pool_tgt_rw_sem(pool)); osts = &(pool->pool_obds); } else { osts = &(m->lod_pool_info); @@ -1004,7 +1011,7 @@ static int lod_alloc_qos(const struct lu_env *env, struct lod_object *lo, GOTO(out_nolock, rc = -EAGAIN); /* Do actual allocation, use write lock here. */ - cfs_down_write(&m->lod_qos.lq_rw_sem); + down_write(&m->lod_qos.lq_rw_sem); /* * Check again, while we were sleeping on @lq_rw_sem things could @@ -1132,11 +1139,16 @@ static int lod_alloc_qos(const struct lu_env *env, struct lod_object *lo, 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; } + + if (rc) { + /* no OST found on this iteration, give up */ + break; + } } if (unlikely(nfound != stripe_cnt)) { @@ -1149,9 +1161,9 @@ static int lod_alloc_qos(const struct lu_env *env, struct lod_object *lo, */ 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 */ @@ -1162,11 +1174,11 @@ static int lod_alloc_qos(const struct lu_env *env, struct lod_object *lo, } out: - cfs_up_write(&m->lod_qos.lq_rw_sem); + up_write(&m->lod_qos.lq_rw_sem); out_nolock: if (pool != NULL) { - cfs_up_read(&pool_tgt_rw_sem(pool)); + up_read(&pool_tgt_rw_sem(pool)); /* put back ref got by lod_find_pool() */ lod_pool_putref(pool); } @@ -1221,13 +1233,6 @@ static int lod_use_defined_striping(const struct lu_env *env, 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_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); @@ -1257,10 +1262,14 @@ static int lod_qos_parse_config(const struct lu_env *env, 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 */ @@ -1306,6 +1315,9 @@ static int lod_qos_parse_config(const struct lu_env *env, 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 != @@ -1337,9 +1349,11 @@ int lod_qos_prep_create(const struct lu_env *env, struct lod_object *lo, 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); @@ -1367,24 +1381,42 @@ int lod_qos_prep_create(const struct lu_env *env, struct lod_object *lo, * 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); + } else { + lo->ldo_stripe = stripe; + lo->ldo_stripes_allocated = stripe_len; + } } else { /* * lod_qos_parse_config() found supplied buf as a predefined