X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=d9f0a40683c3c7612c56ad6f4e23b8e3adfb23f1;hp=8dbcf11a27517661d8465ae552e57a632b70f283;hb=e0702769f267dd009a6287bbc9da2760079a101d;hpb=7ef817c9af35fa8fe56ea911afa1e54d6db71c9b diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index 8dbcf11..d9f0a40 100644 --- a/lustre/osd-ldiskfs/osd_handler.c +++ b/lustre/osd-ldiskfs/osd_handler.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Intel Corporation. + * Copyright (c) 2011, 2013, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -73,6 +73,10 @@ int ldiskfs_pdo = 1; CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644, "ldiskfs with parallel directory operations"); +int ldiskfs_track_declares_assert; +CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644, + "LBUG during tracking of declares"); + static const char dot[] = "."; static const char dotdot[] = ".."; static const char remote_obj_dir[] = "REM_OBJ_DIR"; @@ -84,7 +88,6 @@ static const struct dt_object_operations osd_obj_otable_it_ops; static const struct dt_index_operations osd_index_iam_ops; static const struct dt_index_operations osd_index_ea_ops; -#ifdef OSD_TRACK_DECLARES int osd_trans_declare_op2rb[] = { [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET, [OSD_OT_PUNCH] = OSD_OT_MAX, @@ -96,9 +99,9 @@ int osd_trans_declare_op2rb[] = { [OSD_OT_WRITE] = OSD_OT_WRITE, [OSD_OT_INSERT] = OSD_OT_DELETE, [OSD_OT_DELETE] = OSD_OT_INSERT, + [OSD_OT_UPDATE] = OSD_OT_MAX, [OSD_OT_QUOTA] = OSD_OT_MAX, }; -#endif static int osd_has_index(const struct osd_object *obj) { @@ -168,41 +171,27 @@ static struct lu_object *osd_object_alloc(const struct lu_env *env, } } -static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry, - const char *name, void *buf, int len) -{ - dentry->d_inode = inode; - dentry->d_sb = inode->i_sb; - return inode->i_op->getxattr(dentry, name, buf, len); -} - int osd_get_lma(struct osd_thread_info *info, struct inode *inode, struct dentry *dentry, struct lustre_mdt_attrs *lma) { int rc; - rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, (void *)lma, - sizeof(*lma)); - if (rc == -ERANGE) { - /* try with old lma size */ - rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, - info->oti_mdt_attrs_old, - LMA_OLD_SIZE); - if (rc > 0) - memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma)); - } + CLASSERT(LMA_OLD_SIZE >= sizeof(*lma)); + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, + info->oti_mdt_attrs_old, LMA_OLD_SIZE); if (rc > 0) { + if ((void *)lma != (void *)info->oti_mdt_attrs_old) + memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma)); + rc = 0; + lustre_lma_swab(lma); /* Check LMA compatibility */ - if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) { - CWARN("%.16s: unsupported incompat LMA feature(s) " - "%lu/%#x\n", + if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) { + CWARN("%.16s: unsupported incompat LMA feature(s) %#x " + "for fid = "DFID", ino = %lu\n", LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - inode->i_ino, le32_to_cpu(lma->lma_incompat) & - ~LMA_INCOMPAT_SUPP); - rc = -ENOSYS; - } else { - lustre_lma_swab(lma); - rc = 0; + lma->lma_incompat & ~LMA_INCOMPAT_SUPP, + PFID(&lma->lma_self_fid), inode->i_ino); + rc = -EOPNOTSUPP; } } else if (rc == 0) { rc = -ENODATA; @@ -284,35 +273,91 @@ osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, return inode; } -static struct inode * -osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev, - struct osd_inode_id *id, const struct lu_fid *fid) +/** + * \retval +v: new filter_fid, does not contain self-fid + * \retval 0: filter_fid_old, contains self-fid + * \retval -v: other failure cases + */ +int osd_get_idif(struct osd_thread_info *info, struct inode *inode, + struct dentry *dentry, struct lu_fid *fid) { - struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; - struct inode *inode; + struct filter_fid_old *ff = &info->oti_ff; + struct ost_id *ostid = &info->oti_ostid; int rc; - inode = osd_iget(info, dev, id); - if (IS_ERR(inode)) - return inode; + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff)); + if (rc == sizeof(*ff)) { + rc = 0; + ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq)); + ostid_set_id(ostid, le64_to_cpu(ff->ff_objid)); + /* XXX: should use real OST index in the future. LU-3569 */ + ostid_to_fid(fid, ostid, 0); + } else if (rc == sizeof(struct filter_fid)) { + rc = 1; + } else if (rc >= 0) { + rc = -EINVAL; + } - rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); - if (rc == -ENODATA) - return inode; + return rc; +} - if (rc != 0) { - iput(inode); - return ERR_PTR(rc); +static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) +{ + struct osd_thread_info *info = osd_oti_get(env); + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + struct inode *inode = obj->oo_inode; + struct dentry *dentry = &info->oti_obj_dentry; + struct lu_fid *fid = NULL; + int rc; + ENTRY; + + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY)) + RETURN(0); + + CLASSERT(LMA_OLD_SIZE >= sizeof(*lma)); + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, + info->oti_mdt_attrs_old, LMA_OLD_SIZE); + if (rc == -ENODATA && !fid_is_igif(lu_object_fid(&obj->oo_dt.do_lu)) && + osd_obj2dev(obj)->od_check_ff) { + fid = &lma->lma_self_fid; + rc = osd_get_idif(info, inode, dentry, fid); + if (rc > 0) + RETURN(0); } - if (!lu_fid_eq(fid, &lma->lma_self_fid)) { - CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n", - PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid)); - iput(inode); - return ERR_PTR(-EREMCHG); + if (unlikely(rc == -ENODATA)) + RETURN(0); + + if (rc < 0) + RETURN(rc); + + if (rc > 0) { + rc = 0; + lustre_lma_swab(lma); + if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) || + CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) { + CWARN("%s: unsupported incompat LMA feature(s) %#x for " + "fid = "DFID", ino = %lu\n", + osd_obj2dev(obj)->od_svname, + lma->lma_incompat & ~LMA_INCOMPAT_SUPP, + PFID(lu_object_fid(&obj->oo_dt.do_lu)), + inode->i_ino); + rc = -EOPNOTSUPP; + } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) { + fid = &lma->lma_self_fid; + } } - return inode; + if (fid != NULL && + unlikely(!lu_fid_eq(lu_object_fid(&obj->oo_dt.do_lu), fid))) { + CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n", + osd_obj2dev(obj)->od_svname, + PFID(lu_object_fid(&obj->oo_dt.do_lu)), + PFID(&lma->lma_self_fid)); + rc = -EREMCHG; + } + + RETURN(rc); } static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, @@ -328,7 +373,9 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, struct osd_scrub *scrub; struct scrub_file *sf; int result; - int verify = 0; + int saved = 0; + bool in_oi = false; + bool triggered = false; ENTRY; LINVRNT(osd_invariant(obj)); @@ -346,7 +393,8 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, RETURN(-ENOENT); /* Search order: 1. per-thread cache. */ - if (lu_fid_eq(fid, &oic->oic_fid)) { + if (lu_fid_eq(fid, &oic->oic_fid) && + likely(oic->oic_dev == dev)) { id = &oic->oic_lid; goto iget; } @@ -359,9 +407,6 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, goto iget; } - if (sf->sf_flags & SF_INCONSISTENT) - verify = 1; - /* * Objects are created as locking anchors or place holders for objects * yet to be created. No need to osd_oi_lookup() at here because FID @@ -373,9 +418,10 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, GOTO(out, result = 0); /* Search order: 3. OI files. */ - result = osd_oi_lookup(info, dev, fid, id, true); + result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); if (result == -ENOENT) { - if (!fid_is_norm(fid) || fid_is_on_ost(info, dev, fid) || + if (!fid_is_norm(fid) || + fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) || !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid), sf->sf_oi_bitmap)) GOTO(out, result = 0); @@ -386,19 +432,44 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (result != 0) GOTO(out, result); + in_oi = true; + iget: - if (verify == 0) - inode = osd_iget(info, dev, id); - else - inode = osd_iget_verify(info, dev, id, fid); + inode = osd_iget(info, dev, id); if (IS_ERR(inode)) { result = PTR_ERR(inode); if (result == -ENOENT || result == -ESTALE) { - fid_zero(&oic->oic_fid); - result = 0; + if (!in_oi) { + fid_zero(&oic->oic_fid); + GOTO(out, result = 0); + } + + /* XXX: There are three possible cases: + * 1. Backup/restore caused the OI invalid. + * 2. Someone unlinked the object but NOT removed + * the OI mapping, such as mount target device + * as ldiskfs, and modify something directly. + * 3. Someone just removed the object between the + * former oi_lookup and the iget. It is normal. + * + * It is diffcult to distinguish the 2nd from the + * 1st case. Relatively speaking, the 1st case is + * common than the 2nd case, trigger OI scrub. */ + result = osd_oi_lookup(info, dev, fid, id, true); + if (result == 0) + /* It is the case 1 or 2. */ + goto trigger; + + if (result == -ENOENT) + /* It is the case 3. */ + result = 0; } else if (result == -EREMCHG) { trigger: + if (unlikely(triggered)) + GOTO(out, result = saved); + + triggered = true; if (thread_is_running(&scrub->os_thread)) { result = -EINPROGRESS; } else if (!dev->od_noscrub) { @@ -412,6 +483,31 @@ trigger: else result = -EREMCHG; } + + /* We still have chance to get the valid inode: for the + * object which is referenced by remote name entry, the + * object on the local MDT will be linked under the dir + * of "/REMOTE_PARENT_DIR" with its FID string as name. + * + * We do not know whether the object for the given FID + * is referenced by some remote name entry or not, and + * especially for DNE II, a multiple-linked object may + * have many name entries reside on many MDTs. + * + * To simplify the operation, OSD will not distinguish + * more, just lookup "/REMOTE_PARENT_DIR". Usually, it + * only happened for the RPC from other MDT during the + * OI scrub, or for the client side RPC with FID only, + * such as FID to path, or from old connected client. */ + saved = result; + result = osd_lookup_in_remote_parent(info, dev, + fid, id); + if (result == 0) { + in_oi = false; + goto iget; + } + + result = saved; } GOTO(out, result); @@ -420,6 +516,16 @@ trigger: obj->oo_inode = inode; LASSERT(obj->oo_inode->i_sb == osd_sb(dev)); + result = osd_check_lma(env, obj); + if (result != 0) { + iput(inode); + obj->oo_inode = NULL; + if (result == -EREMCHG) + goto trigger; + + GOTO(out, result); + } + obj->oo_compat_dot_created = 1; obj->oo_compat_dotdot_created = 1; @@ -646,7 +752,7 @@ static void osd_trans_commit_cb(struct super_block *sb, dcb->dcb_func(NULL, th, dcb, error); } - lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th); + lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th); lu_device_put(lud); th->th_dev = NULL; @@ -668,7 +774,7 @@ static struct thandle *osd_trans_create(const struct lu_env *env, LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0); th = ERR_PTR(-ENOMEM); - OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO); + OBD_ALLOC_GFP(oh, sizeof *oh, __GFP_IO); if (oh != NULL) { oh->ot_quota_trans = &oti->oti_quota_trans; memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans)); @@ -681,7 +787,6 @@ static struct thandle *osd_trans_create(const struct lu_env *env, CFS_INIT_LIST_HEAD(&oh->ot_dcb_list); osd_th_alloced(oh); -#ifdef OSD_TRACK_DECLARES memset(oti->oti_declare_ops, 0, sizeof(oti->oti_declare_ops)); memset(oti->oti_declare_ops_rb, 0, @@ -689,7 +794,6 @@ static struct thandle *osd_trans_create(const struct lu_env *env, memset(oti->oti_declare_ops_cred, 0, sizeof(oti->oti_declare_ops_cred)); oti->oti_rollback = false; -#endif } RETURN(th); } @@ -719,16 +823,13 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, GOTO(out, rc); if (unlikely(osd_param_is_not_sane(dev, th))) { -#ifdef OSD_TRACK_DECLARES static unsigned long last_printed; static int last_credits; -#endif CWARN("%.16s: too many transaction credits (%d > %d)\n", LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, oh->ot_credits, osd_journal(dev)->j_max_transaction_buffers); -#ifdef OSD_TRACK_DECLARES CWARN(" create: %u/%u, delete: %u/%u, destroy: %u/%u\n", oti->oti_declare_ops[OSD_OT_CREATE], oti->oti_declare_ops_cred[OSD_OT_CREATE], @@ -765,7 +866,6 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, last_credits = oh->ot_credits; last_printed = jiffies; } -#endif /* XXX Limit the credits to 'max_transaction_buffers', and * let the underlying filesystem to catch the error if * we really need so many credits. @@ -788,8 +888,8 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, lu_context_enter(&th->th_ctx); lu_device_get(&d->dd_lu_dev); - oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference, - "osd-tx", th); + lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link, + "osd-tx", th); oti->oti_txns++; rc = 0; } else { @@ -1320,48 +1420,50 @@ static int capa_is_sane(const struct lu_env *env, } int osd_object_auth(const struct lu_env *env, struct dt_object *dt, - struct lustre_capa *capa, __u64 opc) + struct lustre_capa *capa, __u64 opc) { - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_device *dev = osd_dev(dt->do_lu.lo_dev); - struct md_capainfo *ci; - int rc; + const struct lu_fid *fid = lu_object_fid(&dt->do_lu); + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + struct lu_capainfo *lci; + int rc; - if (!dev->od_fl_capa) - return 0; + if (!osd->od_fl_capa) + return 0; - if (capa == BYPASS_CAPA) - return 0; + if (capa == BYPASS_CAPA) + return 0; - ci = md_capainfo(env); - if (unlikely(!ci)) - return 0; + lci = lu_capainfo_get(env); + if (unlikely(lci == NULL)) + return 0; - if (ci->mc_auth == LC_ID_NONE) - return 0; + if (lci->lci_auth == LC_ID_NONE) + return 0; - if (!capa) { - CERROR("no capability is provided for fid "DFID"\n", PFID(fid)); - return -EACCES; - } + if (capa == NULL) { + CERROR("%s: no capability provided for FID "DFID": rc = %d\n", + osd_name(osd), PFID(fid), -EACCES); + return -EACCES; + } - if (!lu_fid_eq(fid, &capa->lc_fid)) { - DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with", - PFID(fid)); - return -EACCES; - } + if (!lu_fid_eq(fid, &capa->lc_fid)) { + DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with", + PFID(fid)); + return -EACCES; + } - if (!capa_opc_supported(capa, opc)) { - DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc); - return -EACCES; - } + if (!capa_opc_supported(capa, opc)) { + DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc); + return -EACCES; + } - if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) { - DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc); - return -EACCES; - } + rc = capa_is_sane(env, osd, capa, osd->od_capa_keys); + if (rc != 0) { + DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc); + return -EACCES; + } - return 0; + return 0; } static struct timespec *osd_inode_time(const struct lu_env *env, @@ -1697,9 +1799,8 @@ struct dentry *osd_child_dentry_get(const struct lu_env *env, } static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj, - cfs_umode_t mode, - struct dt_allocation_hint *hint, - struct thandle *th) + umode_t mode, struct dt_allocation_hint *hint, + struct thandle *th) { int result; struct osd_device *osd = osd_obj2dev(obj); @@ -1832,16 +1933,16 @@ static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj, } static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *th) + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *th) { - cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX); - int result; + umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX); + int result; - LINVRNT(osd_invariant(obj)); - LASSERT(obj->oo_inode == NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(obj->oo_inode == NULL); LASSERT(S_ISCHR(mode) || S_ISBLK(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)); @@ -1896,7 +1997,7 @@ static osd_obj_type_f osd_create_type_f(enum dt_format_type type) static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah, struct dt_object *parent, struct dt_object *child, - cfs_umode_t child_mode) + umode_t child_mode) { LASSERT(ah); @@ -1992,22 +2093,22 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj, LASSERT(obj->oo_inode != NULL); osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation); - return osd_oi_insert(info, osd, fid, id, th); + return osd_oi_insert(info, osd, fid, id, th, OI_CHECK_FLD); } int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, - const struct lu_fid *fid, struct lu_seq_range *range) + obd_seq seq, struct lu_seq_range *range) { struct seq_server_site *ss = osd_seq_site(osd); int rc; - if (fid_is_idif(fid)) { + if (fid_seq_is_idif(seq)) { fld_range_set_ost(range); - range->lsr_index = fid_idif_ost_idx(fid); + range->lsr_index = idif_ost_idx(seq); return 0; } - if (!fid_seq_in_fldb(fid_seq(fid))) { + if (!fid_seq_in_fldb(seq)) { fld_range_set_mdt(range); if (ss != NULL) /* FIXME: If ss is NULL, it suppose not get lsr_index @@ -2018,10 +2119,10 @@ int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, LASSERT(ss != NULL); fld_range_set_any(range); - rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range); + rc = fld_server_lookup(env, ss->ss_server_fld, seq, range); if (rc != 0) { - CERROR("%s: cannot find FLD range for "DFID": rc = %d\n", - osd_name(osd), PFID(fid), rc); + CERROR("%s: cannot find FLD range for "LPX64": rc = %d\n", + osd_name(osd), seq, rc); } return rc; } @@ -2049,7 +2150,7 @@ static int osd_declare_object_create(const struct lu_env *env, osd_trans_declare_op(env, oh, OSD_OT_CREATE, osd_dto_credits_noquota[DTO_OBJECT_CREATE]); if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev), - lu_object_fid(&dt->do_lu))) + lu_object_fid(&dt->do_lu), OI_CHECK_FLD)) /* Reuse idle OI block may cause additional one OI block * to be changed. */ osd_trans_declare_op(env, oh, OSD_OT_INSERT, @@ -2081,7 +2182,7 @@ static int osd_declare_object_create(const struct lu_env *env, if (fid_is_norm(lu_object_fid(&dt->do_lu)) && !fid_is_last_id(lu_object_fid(&dt->do_lu))) osd_fld_lookup(env, osd_dt_dev(handle->th_dev), - lu_object_fid(&dt->do_lu), range); + fid_seq(lu_object_fid(&dt->do_lu)), range); RETURN(rc); @@ -2180,9 +2281,6 @@ static int osd_object_destroy(const struct lu_env *env, if (unlikely(fid_is_acct(fid))) RETURN(-EPERM); - /* Parallel control for OI scrub. For most of cases, there is no - * lock contention. So it will not affect unlink performance. */ - mutex_lock(&inode->i_mutex); if (S_ISDIR(inode->i_mode)) { LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1); /* it will check/delete the inode from remote parent, @@ -2200,8 +2298,7 @@ static int osd_object_destroy(const struct lu_env *env, osd_trans_exec_op(env, th, OSD_OT_DESTROY); - result = osd_oi_delete(osd_oti_get(env), osd, fid, th); - mutex_unlock(&inode->i_mutex); + result = osd_oi_delete(osd_oti_get(env), osd, fid, th, OI_CHECK_FLD); /* XXX: add to ext3 orphan list */ /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */ @@ -2223,23 +2320,45 @@ static int osd_object_destroy(const struct lu_env *env, * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here */ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, - const struct lu_fid *fid, __u64 flags) + const struct lu_fid *fid, __u32 compat, __u32 incompat) { struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; int rc; + ENTRY; if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA)) - return 0; + RETURN(0); - lustre_lma_init(lma, fid, flags); + lustre_lma_init(lma, fid, compat, incompat); lustre_lma_swab(lma); rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma), XATTR_CREATE); - /* Someone may created the EA by race. */ - if (unlikely(rc == -EEXIST)) - rc = 0; - return rc; + /* LMA may already exist, but we need to check that all the + * desired compat/incompat flags have been added. */ + if (unlikely(rc == -EEXIST)) { + if (compat == 0 && incompat == 0) + RETURN(0); + + rc = __osd_xattr_get(inode, &info->oti_obj_dentry, + XATTR_NAME_LMA, info->oti_mdt_attrs_old, + LMA_OLD_SIZE); + if (rc <= 0) + RETURN(-EINVAL); + + lustre_lma_swab(lma); + if (!(~lma->lma_compat & compat) && + !(~lma->lma_incompat & incompat)) + RETURN(0); + + lma->lma_compat |= compat; + lma->lma_incompat |= incompat; + lustre_lma_swab(lma); + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, + sizeof(*lma), XATTR_REPLACE); + } + + RETURN(rc); } /** @@ -2336,7 +2455,7 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, } /* Set special LMA flag for local agent inode */ - rc = osd_ea_fid_set(info, local, fid, LMAI_AGENT); + rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT); if (rc != 0) { CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n", osd_name(osd), PFID(fid), rc); @@ -2421,7 +2540,10 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, result = __osd_object_create(info, obj, attr, hint, dof, th); if (result == 0) - result = osd_ea_fid_set(info, obj->oo_inode, fid, 0); + result = osd_ea_fid_set(info, obj->oo_inode, fid, + fid_is_on_ost(info, osd_obj2dev(obj), + fid, OI_CHECK_FLD) ? + LMAC_FID_ON_OST : 0, 0); if (result == 0) result = __osd_oi_insert(env, obj, fid, th); @@ -2456,8 +2578,10 @@ static int osd_declare_object_ref_add(const struct lu_env *env, static int osd_object_ref_add(const struct lu_env *env, struct dt_object *dt, struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + bool need_dirty = false; + int rc = 0; LINVRNT(osd_invariant(obj)); LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); @@ -2466,33 +2590,44 @@ static int osd_object_ref_add(const struct lu_env *env, osd_trans_exec_op(env, th, OSD_OT_REF_ADD); - /* - * DIR_NLINK feature is set for compatibility reasons if: - * 1) nlinks > LDISKFS_LINK_MAX, or - * 2) nlinks == 2, since this indicates i_nlink was previously 1. + /* This based on ldiskfs_inc_count(), which is not exported. + * + * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX + * (65000) subdirectories by storing "1" in i_nlink if the link count + * would otherwise overflow. Directory tranversal tools understand + * that (st_nlink == 1) indicates that the filesystem dose not track + * hard links count on the directory, and will not abort subdirectory + * scanning early once (st_nlink - 2) subdirs have been found. * - * It is easier to always set this flag (rather than check and set), - * since it has less overhead, and the superblock will be dirtied - * at some point. Both e2fsprogs and any Lustre-supported ldiskfs - * do not actually care whether this flag is set or not. + * This also has to properly handle the case of inodes with nlink == 0 + * in case they are being linked into the PENDING directory */ spin_lock(&obj->oo_guard); - /* inc_nlink from 0 may cause WARN_ON */ - if(inode->i_nlink == 0) + if (unlikely(!S_ISDIR(inode->i_mode) && + inode->i_nlink >= LDISKFS_LINK_MAX)) { + /* MDD should have checked this, but good to be safe */ + rc = -EMLINK; + } else if (unlikely(inode->i_nlink == 0 || + (S_ISDIR(inode->i_mode) && + inode->i_nlink >= LDISKFS_LINK_MAX))) { + /* inc_nlink from 0 may cause WARN_ON */ set_nlink(inode, 1); - else + need_dirty = true; + } else if (!S_ISDIR(inode->i_mode) || + (S_ISDIR(inode->i_mode) && inode->i_nlink >= 2)) { inc_nlink(inode); - if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) { - if (inode->i_nlink >= LDISKFS_LINK_MAX || - inode->i_nlink == 2) - set_nlink(inode, 1); - } + need_dirty = true; + } /* else (S_ISDIR(inode->i_mode) && inode->i_nlink == 1) { ; } */ + LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX); spin_unlock(&obj->oo_guard); - ll_dirty_inode(inode, I_DIRTY_DATASYNC); + + if (need_dirty) + ll_dirty_inode(inode, I_DIRTY_DATASYNC); + LINVRNT(osd_invariant(obj)); - return 0; + return rc; } static int osd_declare_object_ref_del(const struct lu_env *env, @@ -2531,15 +2666,24 @@ static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt, spin_lock(&obj->oo_guard); LASSERT(inode->i_nlink > 0); - drop_nlink(inode); - /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX) - * then the nlink count is 1. Don't let it be set to 0 or the directory - * inode will be deleted incorrectly. */ - if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0) - set_nlink(inode, 1); - spin_unlock(&obj->oo_guard); - ll_dirty_inode(inode, I_DIRTY_DATASYNC); - LINVRNT(osd_invariant(obj)); + + /* This based on ldiskfs_dec_count(), which is not exported. + * + * If a directory already has nlink == 1, then do not drop the nlink + * count to 0, even temporarily, to avoid race conditions with other + * threads not holding oo_guard seeing i_nlink == 0 in rare cases. + * + * nlink == 1 means the directory has/had > EXT4_LINK_MAX subdirs. + * */ + if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 1) { + drop_nlink(inode); + + spin_unlock(&obj->oo_guard); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); + LINVRNT(osd_invariant(obj)); + } else { + spin_unlock(&obj->oo_guard); + } return 0; } @@ -2736,90 +2880,89 @@ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, } static struct obd_capa *osd_capa_get(const struct lu_env *env, - struct dt_object *dt, - struct lustre_capa *old, - __u64 opc) + struct dt_object *dt, + struct lustre_capa *old, __u64 opc) { - struct osd_thread_info *info = osd_oti_get(env); - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_device *dev = osd_obj2dev(obj); - struct lustre_capa_key *key = &info->oti_capa_key; - struct lustre_capa *capa = &info->oti_capa; - struct obd_capa *oc; - struct md_capainfo *ci; - int rc; - ENTRY; + struct osd_thread_info *info = osd_oti_get(env); + const struct lu_fid *fid = lu_object_fid(&dt->do_lu); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_device *osd = osd_obj2dev(obj); + struct lustre_capa_key *key = &info->oti_capa_key; + struct lustre_capa *capa = &info->oti_capa; + struct obd_capa *oc; + struct lu_capainfo *lci; + int rc; + ENTRY; - if (!dev->od_fl_capa) - RETURN(ERR_PTR(-ENOENT)); + if (!osd->od_fl_capa) + RETURN(ERR_PTR(-ENOENT)); LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - /* renewal sanity check */ - if (old && osd_object_auth(env, dt, old, opc)) - RETURN(ERR_PTR(-EACCES)); - - ci = md_capainfo(env); - if (unlikely(!ci)) - RETURN(ERR_PTR(-ENOENT)); - - switch (ci->mc_auth) { - case LC_ID_NONE: - RETURN(NULL); - case LC_ID_PLAIN: - capa->lc_uid = obj->oo_inode->i_uid; - capa->lc_gid = obj->oo_inode->i_gid; - capa->lc_flags = LC_ID_PLAIN; - break; - case LC_ID_CONVERT: { - __u32 d[4], s[4]; - - s[0] = obj->oo_inode->i_uid; - cfs_get_random_bytes(&(s[1]), sizeof(__u32)); - s[2] = obj->oo_inode->i_gid; - cfs_get_random_bytes(&(s[3]), sizeof(__u32)); - rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN); - if (unlikely(rc)) - RETURN(ERR_PTR(rc)); - - capa->lc_uid = ((__u64)d[1] << 32) | d[0]; - capa->lc_gid = ((__u64)d[3] << 32) | d[2]; - capa->lc_flags = LC_ID_CONVERT; - break; - } - default: - RETURN(ERR_PTR(-EINVAL)); + /* renewal sanity check */ + if (old && osd_object_auth(env, dt, old, opc)) + RETURN(ERR_PTR(-EACCES)); + + lci = lu_capainfo_get(env); + if (unlikely(lci == NULL)) + RETURN(ERR_PTR(-ENOENT)); + + switch (lci->lci_auth) { + case LC_ID_NONE: + RETURN(NULL); + case LC_ID_PLAIN: + capa->lc_uid = obj->oo_inode->i_uid; + capa->lc_gid = obj->oo_inode->i_gid; + capa->lc_flags = LC_ID_PLAIN; + break; + case LC_ID_CONVERT: { + __u32 d[4], s[4]; + + s[0] = obj->oo_inode->i_uid; + cfs_get_random_bytes(&(s[1]), sizeof(__u32)); + s[2] = obj->oo_inode->i_gid; + cfs_get_random_bytes(&(s[3]), sizeof(__u32)); + rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN); + if (unlikely(rc)) + RETURN(ERR_PTR(rc)); + + capa->lc_uid = ((__u64)d[1] << 32) | d[0]; + capa->lc_gid = ((__u64)d[3] << 32) | d[2]; + capa->lc_flags = LC_ID_CONVERT; + break; } + default: + RETURN(ERR_PTR(-EINVAL)); + } - capa->lc_fid = *fid; - capa->lc_opc = opc; - capa->lc_flags |= dev->od_capa_alg << 24; - capa->lc_timeout = dev->od_capa_timeout; - capa->lc_expiry = 0; + capa->lc_fid = *fid; + capa->lc_opc = opc; + capa->lc_flags |= osd->od_capa_alg << 24; + capa->lc_timeout = osd->od_capa_timeout; + capa->lc_expiry = 0; - oc = capa_lookup(dev->od_capa_hash, capa, 1); - if (oc) { - LASSERT(!capa_is_expired(oc)); - RETURN(oc); - } + oc = capa_lookup(osd->od_capa_hash, capa, 1); + if (oc) { + LASSERT(!capa_is_expired(oc)); + RETURN(oc); + } spin_lock(&capa_lock); - *key = dev->od_capa_keys[1]; + *key = osd->od_capa_keys[1]; spin_unlock(&capa_lock); - capa->lc_keyid = key->lk_keyid; - capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout; + capa->lc_keyid = key->lk_keyid; + capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout; - rc = capa_hmac(capa->lc_hmac, capa, key->lk_key); - if (rc) { - DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc); - RETURN(ERR_PTR(rc)); - } + rc = capa_hmac(capa->lc_hmac, capa, key->lk_key); + if (rc) { + DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc); + RETURN(ERR_PTR(rc)); + } - oc = capa_add(dev->od_capa_hash, capa); - RETURN(oc); + oc = capa_add(osd->od_capa_hash, capa); + RETURN(oc); } static int osd_object_sync(const struct lu_env *env, struct dt_object *dt) @@ -3180,26 +3323,45 @@ static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de, return rc; } -static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, - struct lu_fid *fid) +static int osd_mdt_seq_exists(const struct lu_env *env, + struct osd_device *osd, obd_seq seq) { struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; struct seq_server_site *ss = osd_seq_site(osd); int rc; ENTRY; - /* Those FID seqs, which are not in FLDB, must be local seq */ - if (unlikely(!fid_seq_in_fldb(fid_seq(fid)) || ss == NULL)) - RETURN(0); + if (ss == NULL) + RETURN(1); - rc = osd_fld_lookup(env, osd, fid, range); + /* XXX: currently, each MDT only store avaible sequence on disk, and no + * allocated sequences information on disk, so we have to lookup FLDB, + * but it probably makes more sense also store allocated sequence + * locally, so we do not need do remote FLDB lookup in OSD */ + rc = osd_fld_lookup(env, osd, seq, range); if (rc != 0) { - CERROR("%s: Can not lookup fld for "DFID"\n", - osd_name(osd), PFID(fid)); - RETURN(rc); + CERROR("%s: Can not lookup fld for "LPX64"\n", + osd_name(osd), seq); + RETURN(0); } - RETURN(ss->ss_node_id != range->lsr_index); + RETURN(ss->ss_node_id == range->lsr_index); +} + +static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, + struct lu_fid *fid) +{ + ENTRY; + + /* FID seqs not in FLDB, must be local seq */ + if (unlikely(!fid_seq_in_fldb(fid_seq(fid)))) + RETURN(0); + + /* Currently only check this for FID on MDT */ + if (osd_mdt_seq_exists(env, osd, fid_seq(fid))) + RETURN(0); + + RETURN(1); } /** @@ -3672,7 +3834,7 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, RETURN_EXIT; again: - rc = osd_oi_lookup(oti, dev, fid, id, true); + rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD); if (rc != 0 && rc != -ENOENT) RETURN_EXIT; @@ -3731,6 +3893,18 @@ static int osd_fail_fid_lookup(struct osd_thread_info *oti, return rc; } +int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd, + struct osd_inode_id *id, const struct lu_fid *fid) +{ + CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid), + id->oii_ino, id->oii_gen, info); + info->oti_cache.oic_lid = *id; + info->oti_cache.oic_fid = *fid; + info->oti_cache.oic_dev = osd; + + return 0; +} + /** * Calls ->lookup() to find dentry. From dentry get inode and * read inode's ea to get fid. This is required for interoperability @@ -3789,13 +3963,15 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, rc = osd_ea_fid_get(env, obj, ino, fid, id); else osd_id_gen(id, ino, OSD_OII_NOGEN); - if (rc != 0 || osd_remote_fid(env, dev, fid)) { + if (rc != 0) { fid_zero(&oic->oic_fid); GOTO(out, rc); } - oic->oic_lid = *id; - oic->oic_fid = *fid; + rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, + fid); + if (rc != 0) + GOTO(out, rc); if ((scrub->os_pos_current <= ino) && ((sf->sf_flags & SF_INCONSISTENT) || (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) || @@ -4814,8 +4990,14 @@ again: GOTO(out_journal, rc); } + /* skip the REMOTE_PARENT_DIR. */ + if (inode == dev->od_mdt_map->omm_remote_parent->d_inode) + GOTO(out_inode, rc = 0); + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); if (rc == 0) { + LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI)); + if (fid_is_sane(fid)) { /* FID-in-dirent is valid. */ if (lu_fid_eq(fid, &lma->lma_self_fid)) @@ -4899,7 +5081,7 @@ again: if (unlikely(fid_is_sane(fid))) { /* FID-in-dirent exists, but FID-in-LMA is lost. * Trust the FID-in-dirent, and add FID-in-LMA. */ - rc = osd_ea_fid_set(info, inode, fid, 0); + rc = osd_ea_fid_set(info, inode, fid, 0, 0); if (rc == 0) *attr |= LUDA_REPAIR; } else { @@ -4968,10 +5150,10 @@ static inline int osd_it_ea_rec(const struct lu_env *env, if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) { attr |= LUDA_IGNORE; rc = 0; - goto pack; + } else { + rc = osd_dirent_check_repair(env, obj, it, fid, id, + &attr); } - - rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr); } else { attr &= ~LU_DIRENT_ATTRS_MASK; if (!fid_is_sane(fid)) { @@ -4984,22 +5166,20 @@ static inline int osd_it_ea_rec(const struct lu_env *env, } } - if (rc < 0) - RETURN(rc); - -pack: + /* Pack the entry anyway, at least the offset is right. */ osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off, it->oie_dirent->oied_name, it->oie_dirent->oied_namelen, it->oie_dirent->oied_type, attr); + if (rc < 0) + RETURN(rc); + if (osd_remote_fid(env, dev, fid)) RETURN(0); - if (likely(!(attr & LUDA_IGNORE))) { - oic->oic_lid = *id; - oic->oic_fid = *fid; - } + if (likely(!(attr & LUDA_IGNORE))) + rc = osd_add_oi_cache(oti, dev, id, fid); if (!(attr & LUDA_VERIFY) && (scrub->os_pos_current <= ino) && @@ -5182,20 +5362,33 @@ static int osd_shutdown(const struct lu_env *env, struct osd_device *o) { ENTRY; - osd_scrub_cleanup(env, o); + /* shutdown quota slave instance associated with the device */ + if (o->od_quota_slave != NULL) { + qsd_fini(env, o->od_quota_slave); + o->od_quota_slave = NULL; + } + + RETURN(0); +} + +static void osd_umount(const struct lu_env *env, struct osd_device *o) +{ + ENTRY; if (o->od_fsops) { fsfilt_put_ops(o->od_fsops); o->od_fsops = NULL; } - /* shutdown quota slave instance associated with the device */ - if (o->od_quota_slave != NULL) { - qsd_fini(env, o->od_quota_slave); - o->od_quota_slave = NULL; + if (o->od_mnt != NULL) { + shrink_dcache_sb(osd_sb(o)); + osd_sync(env, &o->od_dt_dev); + + mntput(o->od_mnt); + o->od_mnt = NULL; } - RETURN(0); + EXIT; } static int osd_mount(const struct lu_env *env, @@ -5209,7 +5402,10 @@ static int osd_mount(const struct lu_env *env, struct file_system_type *type; char *options = NULL; char *str; - int rc = 0; + struct osd_thread_info *info = osd_oti_get(env); + struct lu_fid *fid = &info->oti_fid; + struct inode *inode; + int rc = 0; ENTRY; if (o->od_mnt != NULL) @@ -5219,13 +5415,14 @@ static int osd_mount(const struct lu_env *env, RETURN(-E2BIG); strcpy(o->od_mntdev, dev); - o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS)); - if (o->od_fsops == NULL) { - CERROR("Can't find fsfilt_ldiskfs\n"); - RETURN(-ENOTSUPP); - } + o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS)); + if (IS_ERR(o->od_fsops)) { + CERROR("%s: Can't find fsfilt_ldiskfs\n", name); + o->od_fsops = NULL; + RETURN(-ENOTSUPP); + } - OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD); + OBD_PAGE_ALLOC(__page, GFP_IOFS); if (__page == NULL) GOTO(out, rc = -ENOMEM); @@ -5235,7 +5432,7 @@ static int osd_mount(const struct lu_env *env, if (str) lmd_flags = simple_strtoul(str + 1, NULL, 0); opts = lustre_cfg_string(cfg, 3); - page = (unsigned long)cfs_page_address(__page); + page = (unsigned long)page_address(__page); options = (char *)page; *options = '\0'; if (opts == NULL) @@ -5246,7 +5443,7 @@ static int osd_mount(const struct lu_env *env, /* Glom up mount options */ if (*options != '\0') strcat(options, ","); - strlcat(options, "no_mbcache", CFS_PAGE_SIZE); + strlcat(options, "no_mbcache", PAGE_CACHE_SIZE); type = get_fs_type("ldiskfs"); if (!type) { @@ -5259,8 +5456,8 @@ static int osd_mount(const struct lu_env *env, if (IS_ERR(o->od_mnt)) { rc = PTR_ERR(o->od_mnt); - CERROR("%s: can't mount %s: %d\n", name, dev, rc); o->od_mnt = NULL; + CERROR("%s: can't mount %s: %d\n", name, dev, rc); GOTO(out, rc); } @@ -5268,59 +5465,56 @@ static int osd_mount(const struct lu_env *env, if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) { CERROR("%s: underlying device %s is marked as read-only. " "Setup failed\n", name, dev); - mntput(o->od_mnt); - o->od_mnt = NULL; - GOTO(out, rc = -EROFS); + GOTO(out_mnt, rc = -EROFS); } #endif if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb, LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) { CERROR("%s: device %s is mounted w/o journal\n", name, dev); - mntput(o->od_mnt); - o->od_mnt = NULL; - GOTO(out, rc = -EINVAL); + GOTO(out_mnt, rc = -EINVAL); + } + + inode = osd_sb(o)->s_root->d_inode; + ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI); + lu_local_obj_fid(fid, OSD_FS_ROOT_OID); + rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0); + if (rc != 0) { + CERROR("%s: failed to set lma on %s root inode\n", name, dev); + GOTO(out_mnt, rc); } - ldiskfs_set_inode_state(osd_sb(o)->s_root->d_inode, - LDISKFS_STATE_LUSTRE_NO_OI); if (lmd_flags & LMD_FLG_NOSCRUB) o->od_noscrub = 1; + GOTO(out, rc = 0); + +out_mnt: + mntput(o->od_mnt); + o->od_mnt = NULL; + out: if (__page) OBD_PAGE_FREE(__page); if (rc) fsfilt_put_ops(o->od_fsops); - RETURN(rc); + return rc; } static struct lu_device *osd_device_fini(const struct lu_env *env, - struct lu_device *d) + struct lu_device *d) { - int rc; - ENTRY; - - rc = osd_shutdown(env, osd_dev(d)); - - osd_obj_map_fini(osd_dev(d)); - - shrink_dcache_sb(osd_sb(osd_dev(d))); - osd_sync(env, lu2dt_dev(d)); - - rc = osd_procfs_fini(osd_dev(d)); - if (rc) { - CERROR("proc fini error %d \n", rc); - RETURN (ERR_PTR(rc)); - } + struct osd_device *o = osd_dev(d); + ENTRY; - if (osd_dev(d)->od_mnt) { - mntput(osd_dev(d)->od_mnt); - osd_dev(d)->od_mnt = NULL; - } + osd_procfs_fini(o); + osd_shutdown(env, o); + osd_scrub_cleanup(env, o); + osd_obj_map_fini(o); + osd_umount(env, o); - RETURN(NULL); + RETURN(NULL); } static int osd_device_init0(const struct lu_env *env, @@ -5358,12 +5552,6 @@ static int osd_device_init0(const struct lu_env *env, if (rc) GOTO(out_capa, rc); - CFS_INIT_LIST_HEAD(&o->od_ios_list); - /* setup scrub, including OI files initialization */ - rc = osd_scrub_setup(env, o); - if (rc < 0) - GOTO(out_mnt, rc); - cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4), sizeof(o->od_svname)); if (cplen >= sizeof(o->od_svname)) { @@ -5371,24 +5559,33 @@ static int osd_device_init0(const struct lu_env *env, GOTO(out_mnt, rc); } + if (server_name_is_ost(o->od_svname)) + o->od_is_ost = 1; + rc = osd_obj_map_init(env, o); if (rc != 0) - GOTO(out_scrub, rc); + GOTO(out_mnt, rc); rc = lu_site_init(&o->od_site, l); - if (rc) + if (rc != 0) GOTO(out_compat, rc); o->od_site.ls_bottom_dev = l; rc = lu_site_init_finish(&o->od_site); - if (rc) + if (rc != 0) + GOTO(out_site, rc); + + CFS_INIT_LIST_HEAD(&o->od_ios_list); + /* setup scrub, including OI files initialization */ + rc = osd_scrub_setup(env, o); + if (rc < 0) GOTO(out_site, rc); rc = osd_procfs_init(o, o->od_svname); if (rc != 0) { CERROR("%s: can't initialize procfs: rc = %d\n", o->od_svname, rc); - GOTO(out_site, rc); + GOTO(out_scrub, rc); } LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev); @@ -5403,23 +5600,21 @@ static int osd_device_init0(const struct lu_env *env, } RETURN(0); + out_procfs: osd_procfs_fini(o); +out_scrub: + osd_scrub_cleanup(env, o); out_site: lu_site_fini(&o->od_site); out_compat: osd_obj_map_fini(o); -out_scrub: - osd_scrub_cleanup(env, o); out_mnt: - osd_oi_fini(info, o); - osd_shutdown(env, o); - mntput(o->od_mnt); - o->od_mnt = NULL; + osd_umount(env, o); out_capa: cleanup_capa_hash(o->od_capa_hash); out: - RETURN(rc); + return rc; } static struct lu_device *osd_device_alloc(const struct lu_env *env, @@ -5566,15 +5761,6 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, int result = 0; ENTRY; - if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) { - /* MDT/MDD still use old infrastructure to create - * special files */ - result = llo_local_objects_setup(env, lu2md_dev(pdev), - lu2dt_dev(dev)); - if (result) - RETURN(result); - } - if (osd->od_quota_slave != NULL) /* set up quota slave objects */ result = qsd_prepare(env, osd->od_quota_slave);