X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_lov.c;h=df20c2a21c621484ea949f7b1d8cd35dd39f69b8;hb=3b7d840a6f1ca207ec810272013bc7aa3907f96b;hp=5b6c65654bdd302af4ed855d576632bc16674601;hpb=eb7c28ff977f4e0a280558aa74e23f2a9ab0ea0c;p=fs%2Flustre-release.git diff --git a/lustre/mdd/mdd_lov.c b/lustre/mdd/mdd_lov.c index 5b6c656..df20c2a 100644 --- a/lustre/mdd/mdd_lov.c +++ b/lustre/mdd/mdd_lov.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -28,6 +26,8 @@ /* * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Whamcloud, Inc. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -41,9 +41,6 @@ * Author: wangdi */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_MDS #include @@ -118,11 +115,11 @@ int mdd_init_obd(const struct lu_env *env, struct mdd_device *mdd, if (!bufs) GOTO(cleanup_mem, rc = -ENOMEM); - snprintf(name, strlen(MDD_OBD_NAME) + 35, "%s-%s-%d", - MDD_OBD_NAME, dev, mds_id); + snprintf(name, strlen(MDD_OBD_NAME) + 35, "%s-%s", + MDD_OBD_NAME, dev); - snprintf(uuid, strlen(MDD_OBD_UUID) + 35, "%s-%s-%d", - MDD_OBD_UUID, dev, mds_id); + snprintf(uuid, strlen(MDD_OBD_UUID) + 35, "%s-%s", + MDD_OBD_UUID, dev); lustre_cfg_bufs_reset(bufs, name); lustre_cfg_bufs_set_string(bufs, 1, MDD_OBD_TYPE); @@ -145,8 +142,13 @@ int mdd_init_obd(const struct lu_env *env, struct mdd_device *mdd, LBUG(); } + cfs_spin_lock(&obd->obd_dev_lock); obd->obd_recovering = 1; + cfs_spin_unlock(&obd->obd_dev_lock); obd->u.mds.mds_id = mds_id; + obd->u.obt.obt_osd_properties.osd_max_ea_size = + mdd->mdd_dt_conf.ddp_max_ea_size; + rc = class_setup(obd, lcfg); if (rc) GOTO(class_detach, rc); @@ -158,6 +160,7 @@ int mdd_init_obd(const struct lu_env *env, struct mdd_device *mdd, obd->obd_upcall.onu_upcall = mdd_notify; obd->obd_upcall.onu_owner = mdd; mdd->mdd_obd_dev = obd; + EXIT; class_detach: if (rc) @@ -214,7 +217,8 @@ int mdd_get_md(const struct lu_env *env, struct mdd_object *obj, *md_size = 0; rc = 0; } else if (rc < 0) { - CERROR("Error %d reading eadata - %d\n", rc, *md_size); + CDEBUG(D_OTHER, "Error %d reading eadata - %d\n", + rc, *md_size); } else { /* XXX: Convert lov EA but fixed after verification test. */ *md_size = rc; @@ -272,10 +276,11 @@ static int mdd_lov_set_dir_md(const struct lu_env *env, LASSERT(S_ISDIR(mdd_object_type(obj))); lum = (struct lov_user_md*)buf->lb_buf; - /* if { size, offset, count } = { 0, -1, 0 } and no pool (i.e. all default - * values specified) then delete default striping from dir. */ - if (lum->lmm_stripe_size == 0 && lum->lmm_stripe_count == 0 && - lum->lmm_stripe_offset == (typeof(lum->lmm_stripe_offset))(-1) && + /* if { size, offset, count } = { 0, -1, 0 } and no pool + * (i.e. all default values specified) then delete default + * striping from dir. */ + if (LOVEA_DELETE_VALUES(lum->lmm_stripe_size, lum->lmm_stripe_count, + lum->lmm_stripe_offset) && lum->lmm_magic != LOV_USER_MAGIC_V3) { rc = mdd_xattr_set_txn(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV, 0, handle); @@ -339,7 +344,8 @@ int mdd_lov_set_md(const struct lu_env *env, struct mdd_object *pobj, if (rc > 0) { buf = mdd_buf_get(env, lmm, size); rc = mdd_xattr_set_txn(env, child, buf, - XATTR_NAME_LOV, 0, handle); + XATTR_NAME_LOV, 0, + handle); if (rc) CERROR("error on copy stripe info: rc " "= %d\n", rc); @@ -360,6 +366,24 @@ int mdd_lov_objid_prepare(struct mdd_device *mdd, struct lov_mds_md *lmm) return mds_lov_prepare_objids(mdd->mdd_obd_dev, lmm); } +int mdd_declare_lov_objid_update(const struct lu_env *env, + struct mdd_device *mdd, + struct thandle *handle) +{ + struct obd_device *obd = mdd2obd_dev(mdd); + int size; + + /* in prepare we create local files */ + if (unlikely(mdd->mdd_capa == NULL)) + return 0; + + /* XXX: this is a temporary solution to declare llog changes + * will be fixed in 2.3 with new llog implementation */ + + size = obd->u.mds.mds_lov_desc.ld_tgt_count * sizeof(obd_id); + return dt_declare_record_write(env, mdd->mdd_capa, size, 0, handle); +} + void mdd_lov_objid_update(struct mdd_device *mdd, struct lov_mds_md *lmm) { /* copy mds_lov code is using wrong layer */ @@ -371,13 +395,13 @@ void mdd_lov_create_finish(const struct lu_env *env, struct mdd_device *mdd, const struct md_op_spec *spec) { if (lmm && !spec->no_create) - OBD_FREE(lmm, lmm_size); + OBD_FREE_LARGE(lmm, lmm_size); } int mdd_lov_create(const struct lu_env *env, struct mdd_device *mdd, struct mdd_object *parent, struct mdd_object *child, struct lov_mds_md **lmm, int *lmm_size, - const struct md_op_spec *spec, struct lu_attr *la) + const struct md_op_spec *spec, struct md_attr *ma) { struct obd_device *obd = mdd2obd_dev(mdd); struct obd_export *lov_exp = obd->u.mds.mds_lov_exp; @@ -385,8 +409,9 @@ int mdd_lov_create(const struct lu_env *env, struct mdd_device *mdd, struct obdo *oa; struct lov_stripe_md *lsm = NULL; const void *eadata = spec->u.sp_ea.eadata; - __u32 create_flags = spec->sp_cr_flags; + __u64 create_flags = spec->sp_cr_flags; struct obd_trans_info *oti = &mdd_env_info(env)->mti_oti; + struct lu_attr *la = &ma->ma_attr; int rc = 0; ENTRY; @@ -431,31 +456,39 @@ int mdd_lov_create(const struct lu_env *env, struct mdd_device *mdd, 0, &lsm, (void*)eadata); if (rc) GOTO(out_oti, rc); - } else if (parent != NULL) { + } else { /* get lov ea from parent and set to lov */ struct lov_mds_md *_lmm; - int _lmm_size; + int _lmm_size = mdd_lov_mdsize(env, mdd); - _lmm_size = mdd_lov_mdsize(env, mdd); - _lmm = mdd_max_lmm_get(env, mdd); + LASSERT(parent != NULL); + /* + * can not create child's lov_mds_md by access it + * thru .lustre path + */ + if (mdd_object_obf(parent)) + GOTO(out_oti, rc = -EBADFD); + + _lmm = mdd_max_lmm_get(env, mdd); if (_lmm == NULL) GOTO(out_oti, rc = -ENOMEM); rc = mdd_get_md_locked(env, parent, _lmm, &_lmm_size, XATTR_NAME_LOV); - if (rc > 0) + if (rc > 0) { + _lmm_size = mdd_lov_mdsize(env, mdd); rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE, - lov_exp, *lmm_size, + lov_exp, _lmm_size, &lsm, _lmm); - + } if (rc) GOTO(out_oti, rc); } OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_OPEN_WAIT_CREATE, 10); - rc = obd_create(lov_exp, oa, &lsm, oti); + rc = obd_create(env, lov_exp, oa, &lsm, oti); if (rc) { if (rc > 0) { CERROR("Create error for "DFID": %d\n", @@ -464,6 +497,19 @@ int mdd_lov_create(const struct lu_env *env, struct mdd_device *mdd, } GOTO(out_oti, rc); } + + if (ma->ma_valid & MA_LAY_GEN) + /* If we already have a lsm, the file is not new and we + * are about to change the layout, so we have to bump + * the generation. It is worth noting that old versions + * will be confused by a non-zero gen, that's why + * OBD_INCOMPAT_LMM_VER has been introduced */ + lsm->lsm_layout_gen = ma->ma_layout_gen + 1; + else + /* Start with a null generation for backward + * compatiblity with old versions */ + lsm->lsm_layout_gen = 0; + LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq); } else { LASSERT(eadata != NULL); @@ -472,6 +518,10 @@ int mdd_lov_create(const struct lu_env *env, struct mdd_device *mdd, if (rc) GOTO(out_oti, rc); + if (ma->ma_valid & MA_LAY_GEN) + lsm->lsm_layout_gen = ma->ma_layout_gen; + else + lsm->lsm_layout_gen = 0; } lsm->lsm_object_id = fid_ver_oid(mdd_object_fid(child)); @@ -488,16 +538,14 @@ int mdd_lov_create(const struct lu_env *env, struct mdd_device *mdd, /* When setting attr to ost, FLBKSZ is not needed. */ oa->o_valid &= ~OBD_MD_FLBLKSZ; - obdo_from_la(oa, la, OBD_MD_FLTYPE | OBD_MD_FLATIME | - OBD_MD_FLMTIME | OBD_MD_FLCTIME | OBD_MD_FLSIZE); - + obdo_from_la(oa, la, LA_TYPE | LA_ATIME | LA_MTIME | + LA_CTIME | LA_SIZE); /* * XXX: Pack lustre id to OST, in OST, it will be packed by * filter_fid, but can not see what is the usages. So just pack * o_seq o_ver here, maybe fix it after this cycle. */ - obdo_from_inode(oa, NULL, - (struct lu_fid *)mdd_object_fid(child), 0); + obdo_set_parent_fid(oa, mdd_object_fid(child)); oinfo->oi_oa = oa; oinfo->oi_md = lsm; oinfo->oi_capa = NULL; @@ -547,12 +595,9 @@ out_ids: * used when destroying orphans and from mds_reint_unlink() when MDS wants to * destroy objects on OSS. */ -static int mdd_lovobj_unlink(const struct lu_env *env, struct mdd_device *mdd, - struct mdd_object *obj, struct lu_attr *la, - struct lov_mds_md *lmm, int lmm_size, - struct llog_cookie *logcookies, - int log_unlink) + struct mdd_object *obj, struct lu_attr *la, + struct md_attr *ma, int log_unlink) { struct obd_device *obd = mdd2obd_dev(mdd); struct obd_export *lov_exp = obd->u.mds.mds_lov_exp; @@ -560,7 +605,10 @@ int mdd_lovobj_unlink(const struct lu_env *env, struct mdd_device *mdd, struct obd_trans_info *oti = &mdd_env_info(env)->mti_oti; struct obdo *oa = &mdd_env_info(env)->mti_oa; struct lu_site *site = mdd2lu_dev(mdd)->ld_site; - int rc; + struct lov_mds_md *lmm = ma->ma_lmm; + int lmm_size = ma->ma_lmm_size; + struct llog_cookie *logcookies = ma->ma_cookie; + int rc; ENTRY; if (lmm_size == 0) @@ -586,17 +634,20 @@ int mdd_lovobj_unlink(const struct lu_env *env, struct mdd_device *mdd, oti->oti_logcookies = logcookies; } + if (!(ma->ma_attr_flags & MDS_UNLINK_DESTROY)) + oa->o_flags = OBD_FL_DELORPHAN; + CDEBUG(D_INFO, "destroying OSS object "LPU64":"LPU64"\n", oa->o_seq, oa->o_id); - rc = obd_destroy(lov_exp, oa, lsm, oti, NULL, NULL); + rc = obd_destroy(env, lov_exp, oa, lsm, oti, NULL, NULL); obd_free_memmd(lov_exp, &lsm); RETURN(rc); } /* - * called with obj locked. + * called with obj locked. */ int mdd_lov_destroy(const struct lu_env *env, struct mdd_device *mdd, struct mdd_object *obj, struct lu_attr *la) @@ -645,11 +696,52 @@ int mdd_lov_destroy(const struct lu_env *env, struct mdd_device *mdd, RETURN(rc); } - if (ma->ma_valid & MA_COOKIE) - rc = mdd_lovobj_unlink(env, mdd, obj, la, - ma->ma_lmm, ma->ma_lmm_size, - ma->ma_cookie, 1); - RETURN(rc); + if (ma->ma_valid & MA_COOKIE) + rc = mdd_lovobj_unlink(env, mdd, obj, la, ma, 1); + + RETURN(rc); +} + +int mdd_declare_unlink_log(const struct lu_env *env, struct mdd_object *obj, + struct md_attr *ma, struct thandle *handle) +{ + struct mdd_device *mdd = mdo2mdd(&obj->mod_obj); + int rc, i; + __u16 stripe; + + LASSERT(obj); + LASSERT(ma); + + if (!S_ISREG(lu_object_attr(&obj->mod_obj.mo_lu))) + return 0; + + rc = mdd_lmm_get_locked(env, obj, ma); + if (rc || !(ma->ma_valid & MA_LOV)) + return rc; + + LASSERT(ma->ma_lmm); + if (le32_to_cpu(ma->ma_lmm->lmm_magic) != LOV_MAGIC_V1 && + le32_to_cpu(ma->ma_lmm->lmm_magic) != LOV_MAGIC_V3) { + CERROR("%s: invalid LOV_MAGIC %08x on object "DFID"\n", + mdd->mdd_obd_dev->obd_name, + le32_to_cpu(ma->ma_lmm->lmm_magic), + PFID(lu_object_fid(&obj->mod_obj.mo_lu))); + return -EINVAL; + } + + stripe = le16_to_cpu(ma->ma_lmm->lmm_stripe_count); + if (stripe == LOV_ALL_STRIPES); + stripe = mdd2obd_dev(mdd)->u.mds.mds_lov_desc.ld_tgt_count; + + for (i = 0; i < stripe; i++) { + rc = mdd_declare_llog_record(env, mdd, + sizeof(struct llog_unlink_rec), + handle); + if (rc) + return rc; + } + + return rc; } int mdd_unlink_log(const struct lu_env *env, struct mdd_device *mdd, @@ -697,8 +789,8 @@ int mdd_log_op_setattr(struct obd_device *obd, __u32 uid, __u32 gid, /* write setattr log */ ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT); - rc = llog_add(ctxt, &lsr->lsr_hdr, lsm, logcookies, - cookies_size / sizeof(struct llog_cookie)); + rc = llog_add(NULL, ctxt, &lsr->lsr_hdr, lsm, logcookies, + cookies_size / sizeof(struct llog_cookie)); llog_ctxt_put(ctxt); @@ -768,7 +860,7 @@ static int mdd_osc_setattr_async(struct obd_device *obd, __u32 uid, __u32 gid, oti.oti_logcookies = logcookies; } - obdo_from_inode(oinfo.oi_oa, NULL, (struct lu_fid *)parent, 0); + obdo_set_parent_fid(oinfo.oi_oa, parent); oinfo.oi_capa = oc; /* do async setattr from mds to ost not waiting for responses. */ @@ -804,3 +896,201 @@ int mdd_lov_setattr_async(const struct lu_env *env, struct mdd_object *obj, lmm_size, logcookies, fid, NULL); RETURN(rc); } + +static int grouplock_blocking_ast(struct ldlm_lock *lock, + struct ldlm_lock_desc *desc, + void *data, int flag) +{ + struct md_attr *ma = data; + struct lustre_handle lockh; + int rc = 0; + ENTRY; + + switch (flag) + { + case LDLM_CB_BLOCKING : + /* lock is canceled */ + CDEBUG(D_DLMTRACE, "Lock %p is canceled\n", lock); + + ldlm_lock2handle(lock, &lockh); + rc = ldlm_cli_cancel(&lockh); + + break; + case LDLM_CB_CANCELING : + CDEBUG(D_DLMTRACE, + "Lock %p has been canceled, do cleaning\n", + lock); + + if (ma && ma->ma_som) + OBD_FREE_PTR(ma->ma_som); + if (ma) + OBD_FREE_PTR(ma); + break; + default: + LBUG(); + } + RETURN(rc); +} + +static int grouplock_glimpse_ast(struct ldlm_lock *lock, void *data) +{ + struct ptlrpc_request *req = data; + struct ost_lvb *lvb; + int rc; + struct md_attr *ma; + ENTRY; + + ma = lock->l_ast_data; + + req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK); + req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, + sizeof(*lvb)); + rc = req_capsule_server_pack(&req->rq_pill); + if (rc) { + CERROR("failed pack reply: %d\n", rc); + GOTO(out, rc); + } + + lvb = req_capsule_server_get(&req->rq_pill, &RMF_DLM_LVB); + + if ((ma) && (ma->ma_valid & MA_SOM)) { + lvb->lvb_size = ma->ma_som->msd_size; + lvb->lvb_blocks = ma->ma_som->msd_blocks; + } else if ((ma) && (ma->ma_valid & MA_INODE)) { + lvb->lvb_size = ma->ma_attr.la_size; + lvb->lvb_blocks = ma->ma_attr.la_blocks; + } else { + lvb->lvb_size = 0; + rc = -ELDLM_NO_LOCK_DATA; + } + + EXIT; +out: + if (rc == -ELDLM_NO_LOCK_DATA) + lustre_pack_reply(req, 1, NULL, NULL); + + req->rq_status = rc; + return rc; +} + +int mdd_file_lock(const struct lu_env *env, struct md_object *obj, + struct lov_mds_md *lmm, struct ldlm_extent *extent, + struct lustre_handle *lockh) +{ + struct ldlm_enqueue_info einfo = { 0 }; + struct obd_info oinfo = { { { 0 } } }; + struct obd_device *obd; + struct obd_export *lov_exp; + struct lov_stripe_md *lsm = NULL; + struct md_attr *ma = NULL; + int rc; + ENTRY; + + obd = mdo2mdd(obj)->mdd_obd_dev; + lov_exp = obd->u.mds.mds_lov_exp; + + obd_unpackmd(lov_exp, &lsm, lmm, + lov_mds_md_size(lmm->lmm_stripe_count, lmm->lmm_magic)); + + OBD_ALLOC_PTR(ma); + if (ma == NULL) + GOTO(out, rc = -ENOMEM); + + OBD_ALLOC_PTR(ma->ma_som); + if (ma->ma_som == NULL) + GOTO(out, rc = -ENOMEM); + + ma->ma_need = MA_SOM | MA_INODE; + mo_attr_get(env, obj, ma); + + einfo.ei_type = LDLM_EXTENT; + einfo.ei_mode = LCK_GROUP; + einfo.ei_cb_bl = grouplock_blocking_ast; + einfo.ei_cb_cp = ldlm_completion_ast; + einfo.ei_cb_gl = grouplock_glimpse_ast; + + if (ma->ma_valid & (MA_SOM | MA_INODE)) + einfo.ei_cbdata = ma; + else + einfo.ei_cbdata = NULL; + + memset(&oinfo.oi_policy, 0, sizeof(oinfo.oi_policy)); + oinfo.oi_policy.l_extent = *extent; + oinfo.oi_lockh = lockh; + oinfo.oi_md = lsm; + oinfo.oi_flags = 0; + + rc = obd_enqueue(lov_exp, &oinfo, &einfo, NULL); + /* ei_cbdata is used as a free flag at exit */ + if (rc) + einfo.ei_cbdata = NULL; + + obd_unpackmd(lov_exp, &lsm, NULL, 0); + +out: + /* ma is freed if not used as callback data */ + if ((einfo.ei_cbdata == NULL) && ma && ma->ma_som) + OBD_FREE_PTR(ma->ma_som); + if ((einfo.ei_cbdata == NULL) && ma) + OBD_FREE_PTR(ma); + + RETURN(rc); +} + +int mdd_file_unlock(const struct lu_env *env, struct md_object *obj, + struct lov_mds_md *lmm, struct lustre_handle *lockh) +{ + struct obd_device *obd; + struct obd_export *lov_exp; + struct lov_stripe_md *lsm = NULL; + int rc; + ENTRY; + + LASSERT(lustre_handle_is_used(lockh)); + + obd = mdo2mdd(obj)->mdd_obd_dev; + lov_exp = obd->u.mds.mds_lov_exp; + + obd_unpackmd(lov_exp, &lsm, lmm, + lov_mds_md_size(lmm->lmm_stripe_count, lmm->lmm_magic)); + + rc = obd_cancel(lov_exp, lsm, LCK_GROUP, lockh); + + obd_unpackmd(lov_exp, &lsm, NULL, 0); + + RETURN(rc); +} + +/* file lov is in ma->ma_lmm */ +/* requested lov is in info->mti_spec.u.sp_ea.eadata */ +int mdd_lum_lmm_cmp(const struct lu_env *env, struct md_object *cobj, + const struct md_op_spec *spec, struct md_attr *ma) +{ + struct obd_export *lov_exp = + mdd2obd_dev(mdo2mdd(cobj))->u.mds.mds_lov_exp; + struct lov_mds_md *lmm = ma->ma_lmm; + struct lov_user_md_v3 *lum = + (struct lov_user_md_v3 *)(spec->u.sp_ea.eadata); + struct lov_stripe_md *lsm = NULL; + int lmm_magic, rc; + ENTRY; + + rc = obd_unpackmd(lov_exp, &lsm, lmm, + lov_mds_md_size(lmm->lmm_stripe_count, + lmm->lmm_magic)); + ma->ma_layout_gen = lsm->lsm_layout_gen; + ma->ma_valid |= MA_LAY_GEN; + + rc = lov_lum_swab_if_needed(lum, &lmm_magic, NULL); + if (rc) + GOTO(out, rc); + + rc = lov_lum_lsm_cmp((struct lov_user_md *)lum, lsm); + if (rc) + GOTO(out, rc); /* keep GOTO to for traces */ + +out: + /* free lsm */ + obd_unpackmd(lov_exp, &lsm, NULL, 0); + return rc; +}