X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=c13c682c10acb2f4c2334c78fa27424f12cac194;hp=99436312104da1c5e9e0c4ca49eebf0c90a55a59;hb=cd09dd47819737604c2fb5109abd90995afeb968;hpb=f5491211ed7251a45e4e6916539cd2a2ab5e6038 diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index 9943631..c13c682 100644 --- a/lustre/osd-ldiskfs/osd_handler.c +++ b/lustre/osd-ldiskfs/osd_handler.c @@ -64,6 +64,7 @@ #include #include "osd_internal.h" +#include "osd_dynlocks.h" /* llo_* api support */ #include @@ -77,6 +78,20 @@ int ldiskfs_track_declares_assert; CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644, "LBUG during tracking of declares"); +/* Slab to allocate dynlocks */ +struct kmem_cache *dynlock_cachep; + +static struct lu_kmem_descr ldiskfs_caches[] = { + { + .ckd_cache = &dynlock_cachep, + .ckd_name = "dynlock_cache", + .ckd_size = sizeof(struct dynlock_handle) + }, + { + .ckd_cache = NULL + } +}; + static const char dot[] = "."; static const char dotdot[] = ".."; static const char remote_obj_dir[] = "REM_OBJ_DIR"; @@ -99,6 +114,7 @@ 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, }; @@ -170,14 +186,6 @@ 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) { @@ -280,35 +288,120 @@ 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 osd_device *osd = osd_obj2dev(obj); + 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->od_check_ff) { + fid = &lma->lma_self_fid; + rc = osd_get_idif(info, inode, dentry, fid); + if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) { + handle_t *jh; + + /* For the given OST-object, if it has neither LMA nor + * FID in XATTR_NAME_FID, then the given FID (which is + * contained in the @obj, from client RPC for locating + * the OST-object) is trusted. We use it to generate + * the LMA. */ + + LASSERT(current->journal_info == NULL); + + jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, + osd_dto_credits_noquota[DTO_XATTR_SET]); + if (IS_ERR(jh)) { + CWARN("%s: cannot start journal for " + "lma_self_repair: rc = %ld\n", + osd_name(osd), PTR_ERR(jh)); + RETURN(0); + } + + rc = osd_ea_fid_set(info, inode, + lu_object_fid(&obj->oo_dt.do_lu), + fid_is_on_ost(info, osd, + lu_object_fid(&obj->oo_dt.do_lu), + OI_CHECK_FLD) ? + LMAC_FID_ON_OST : 0, 0); + if (rc != 0) + CWARN("%s: cannot self repair the LMA: " + "rc = %d\n", osd_name(osd), rc); + ldiskfs_journal_stop(jh); + 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_name(osd), + 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_name(osd), 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, @@ -324,7 +417,9 @@ 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)); @@ -341,8 +436,17 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) RETURN(-ENOENT); + /* For the object is created as locking anchor, or for the object to + * be created on disk. No need to osd_oi_lookup() at here because FID + * shouldn't never be re-used, if it's really a duplicate FID from + * unexpected reason, we should be able to detect it later by calling + * do_create->osd_oi_insert(). */ + if (conf != NULL && conf->loc_flags & LOC_F_NEW) + GOTO(out, result = 0); + /* 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; } @@ -355,23 +459,11 @@ 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 - * shouldn't never be re-used, if it's really a duplicate FID from - * unexpected reason, we should be able to detect it later by calling - * do_create->osd_oi_insert() - */ - if (conf != NULL && conf->loc_flags & LOC_F_NEW) - 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); @@ -382,19 +474,40 @@ 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 = -ENOENT); + } + + /* 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; } 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) { @@ -408,6 +521,31 @@ 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); @@ -416,6 +554,16 @@ 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; @@ -447,50 +595,6 @@ static void osd_object_init0(struct osd_object *obj) (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT)); } -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; - int rc; - ENTRY; - - CLASSERT(LMA_OLD_SIZE >= sizeof(*lma)); - rc = __osd_xattr_get(obj->oo_inode, &info->oti_obj_dentry, - XATTR_NAME_LMA, info->oti_mdt_attrs_old, - LMA_OLD_SIZE); - 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))) { - rc = -EOPNOTSUPP; - CWARN("%s: unsupported incompat LMA feature(s) %#x for " - "fid = "DFID", ino = %lu: rc = %d\n", - osd_obj2dev(obj)->od_svname, - lma->lma_incompat & ~LMA_INCOMPAT_SUPP, - PFID(lu_object_fid(&obj->oo_dt.do_lu)), - obj->oo_inode->i_ino, rc); - } - if (unlikely(!lu_fid_eq(lu_object_fid(&obj->oo_dt.do_lu), - &lma->lma_self_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)); - if (obj->oo_inode != NULL) { - iput(obj->oo_inode); - obj->oo_inode = NULL; - } - rc = -ESTALE; - } - } else if (rc == -ENODATA) { - /* haven't initialize LMA xattr */ - rc = 0; - } - - RETURN(rc); -} - /* * Concurrency: no concurrent access is possible that early in object * life-cycle. @@ -511,13 +615,8 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l, result = osd_fid_lookup(env, obj, lu_object_fid(l), conf); obj->oo_dt.do_body_ops = &osd_body_ops_new; - if (result == 0 && obj->oo_inode != NULL) { - result = osd_check_lma(env, obj); - if (result != 0) - return result; - + if (result == 0 && obj->oo_inode != NULL) osd_object_init0(obj); - } LINVRNT(osd_invariant(obj)); return result; @@ -713,7 +812,7 @@ 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)); @@ -818,7 +917,7 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, * XXX temporary stuff. Some abstraction layer should * be used. */ - jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits); + jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits); osd_th_started(oh); if (!IS_ERR(jh)) { oh->ot_handle = jh; @@ -838,6 +937,28 @@ out: RETURN(rc); } +static int osd_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; + + if (ss == NULL) + RETURN(1); + + rc = osd_fld_lookup(env, osd, seq, range); + if (rc != 0) { + if (rc != -ENOENT) + CERROR("%s: can't lookup FLD sequence "LPX64 + ": rc = %d\n", osd_name(osd), seq, rc); + RETURN(0); + } + + RETURN(ss->ss_node_id == range->lsr_index); +} + /* * Concurrency: shouldn't matter. */ @@ -887,21 +1008,21 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th) OBD_FREE_PTR(oh); } - /* as we want IO to journal and data IO be concurrent, we don't block - * awaiting data IO completion in osd_do_bio(), instead we wait here - * once transaction is submitted to the journal. all reqular requests - * don't do direct IO (except read/write), thus this wait_event becomes - * no-op for them. - * - * IMPORTANT: we have to wait till any IO submited by the thread is - * completed otherwise iobuf may be corrupted by different request - */ - cfs_wait_event(iobuf->dr_wait, - cfs_atomic_read(&iobuf->dr_numreqs) == 0); - if (!rc) - rc = iobuf->dr_error; + /* as we want IO to journal and data IO be concurrent, we don't block + * awaiting data IO completion in osd_do_bio(), instead we wait here + * once transaction is submitted to the journal. all reqular requests + * don't do direct IO (except read/write), thus this wait_event becomes + * no-op for them. + * + * IMPORTANT: we have to wait till any IO submited by the thread is + * completed otherwise iobuf may be corrupted by different request + */ + wait_event(iobuf->dr_wait, + cfs_atomic_read(&iobuf->dr_numreqs) == 0); + if (!rc) + rc = iobuf->dr_error; - RETURN(rc); + RETURN(rc); } static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb) @@ -1054,7 +1175,6 @@ static void osd_conf_get(const struct lu_env *env, /* * XXX should be taken from not-yet-existing fs abstraction layer. */ - param->ddp_mnt = osd_dt_dev(dev)->od_mnt; param->ddp_max_name_len = LDISKFS_NAME_LEN; param->ddp_max_nlink = LDISKFS_LINK_MAX; param->ddp_block_shift = sb->s_blocksize_bits; @@ -1359,48 +1479,50 @@ 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, @@ -1736,9 +1858,8 @@ 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); @@ -1871,16 +1992,16 @@ 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)); @@ -1935,7 +2056,7 @@ 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); @@ -2022,31 +2143,34 @@ static int __osd_object_create(struct osd_thread_info *info, * \retval 0, on success */ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj, - const struct lu_fid *fid, struct thandle *th) + const struct lu_fid *fid, struct thandle *th) { - struct osd_thread_info *info = osd_oti_get(env); - struct osd_inode_id *id = &info->oti_id; - struct osd_device *osd = osd_obj2dev(obj); + struct osd_thread_info *info = osd_oti_get(env); + struct osd_inode_id *id = &info->oti_id; + struct osd_device *osd = osd_obj2dev(obj); + struct osd_thandle *oh; - LASSERT(obj->oo_inode != NULL); + LASSERT(obj->oo_inode != NULL); + + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle); 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, oh->ot_handle, 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 @@ -2057,12 +2181,8 @@ 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); - if (rc != 0) { - CERROR("%s: cannot find FLD range for "DFID": rc = %d\n", - osd_name(osd), PFID(fid), rc); - } - return rc; + /* OSD will only do local fld lookup */ + return fld_local_lookup(env, ss->ss_server_fld, seq, range); } /* @@ -2075,7 +2195,6 @@ static int osd_declare_object_create(const struct lu_env *env, struct dt_object_format *dof, struct thandle *handle) { - struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; struct osd_thandle *oh; int rc; ENTRY; @@ -2088,7 +2207,7 @@ 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, @@ -2113,16 +2232,6 @@ static int osd_declare_object_create(const struct lu_env *env, if (rc != 0) RETURN(rc); - /* It does fld look up inside declare, and the result will be - * added to fld cache, so the following fld lookup inside insert - * does not need send RPC anymore, so avoid send rpc with holding - * transaction */ - 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); - - RETURN(rc); } @@ -2220,7 +2329,8 @@ static int osd_object_destroy(const struct lu_env *env, RETURN(-EPERM); if (S_ISDIR(inode->i_mode)) { - LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1); + LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 || + inode->i_nlink == 2); /* it will check/delete the inode from remote parent, * how to optimize it? unlink performance impaction XXX */ result = osd_delete_from_remote_parent(env, osd, obj, oh); @@ -2236,7 +2346,8 @@ 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); + result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle, + OI_CHECK_FLD); /* XXX: add to ext3 orphan list */ /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */ @@ -2258,23 +2369,45 @@ 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); } /** @@ -2371,7 +2504,7 @@ 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); @@ -2456,7 +2589,10 @@ 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); @@ -2491,20 +2627,22 @@ 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; - bool need_dirty = false; - int rc = 0; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thandle *oh; + int rc = 0; LINVRNT(osd_invariant(obj)); LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(osd_write_locked(env, obj)); LASSERT(th != NULL); + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); + osd_trans_exec_op(env, th, OSD_OT_REF_ADD); - /* 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 @@ -2516,28 +2654,11 @@ static int osd_object_ref_add(const struct lu_env *env, * in case they are being linked into the PENDING directory */ spin_lock(&obj->oo_guard); - 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); - need_dirty = true; - } else if (!S_ISDIR(inode->i_mode) || - (S_ISDIR(inode->i_mode) && inode->i_nlink >= 2)) { - inc_nlink(inode); - need_dirty = true; - } /* else (S_ISDIR(inode->i_mode) && inode->i_nlink == 1) { ; } */ - + ldiskfs_inc_count(oh->ot_handle, inode); LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX); spin_unlock(&obj->oo_guard); - if (need_dirty) - ll_dirty_inode(inode, I_DIRTY_DATASYNC); - + ll_dirty_inode(inode, I_DIRTY_DATASYNC); LINVRNT(osd_invariant(obj)); return rc; @@ -2567,37 +2688,40 @@ static int osd_declare_object_ref_del(const struct lu_env *env, static int osd_object_ref_del(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; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + struct osd_thandle *oh; - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(osd_write_locked(env, obj)); - LASSERT(th != NULL); + LASSERT(osd_write_locked(env, obj)); + LASSERT(th != NULL); + + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); osd_trans_exec_op(env, th, OSD_OT_REF_DEL); spin_lock(&obj->oo_guard); - LASSERT(inode->i_nlink > 0); - - /* 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 { + /* That can be result of upgrade from old Lustre version and + * applied only to local files. Just skip this ref_del call. + * ext4_unlink() only treats this as a warning, don't LASSERT here.*/ + if (inode->i_nlink == 0) { + CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ? + D_ERROR : D_INODE, "%s: nlink == 0 on "DFID + ", maybe an upgraded file? (LU-3915)\n", + osd_name(osd), PFID(lu_object_fid(&dt->do_lu))); spin_unlock(&obj->oo_guard); + return 0; } + ldiskfs_dec_count(oh->ot_handle, inode); + spin_unlock(&obj->oo_guard); + + ll_dirty_inode(inode, I_DIRTY_DATASYNC); + LINVRNT(osd_invariant(obj)); + return 0; } @@ -2793,90 +2917,89 @@ 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) @@ -2895,6 +3018,7 @@ static int osd_object_sync(const struct lu_env *env, struct dt_object *dt) file->f_dentry = dentry; file->f_mapping = inode->i_mapping; file->f_op = inode->i_fop; + set_file_inode(file, inode); #ifndef HAVE_FILE_FSYNC_4ARGS mutex_lock(&inode->i_mutex); #endif @@ -3240,23 +3364,16 @@ static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de, static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, struct lu_fid *fid) { - 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)) + /* FID seqs not in FLDB, must be local seq */ + if (unlikely(!fid_seq_in_fldb(fid_seq(fid)))) RETURN(0); - rc = osd_fld_lookup(env, osd, fid, range); - if (rc != 0) { - CERROR("%s: Can not lookup fld for "DFID"\n", - osd_name(osd), PFID(fid)); - RETURN(rc); - } + if (osd_seq_exists(env, osd, fid_seq(fid))) + RETURN(0); - RETURN(ss->ss_node_id != range->lsr_index); + RETURN(1); } /** @@ -3311,7 +3428,7 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, down_write(&obj->oo_ext_idx_sem); } - bh = ldiskfs_find_entry(dir, &dentry->d_name, &de, hlock); + bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); if (bh) { __u32 ino = 0; @@ -3729,7 +3846,7 @@ 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; @@ -3788,15 +3905,14 @@ static int osd_fail_fid_lookup(struct osd_thread_info *oti, return rc; } -static int osd_add_oi_cache(struct osd_thread_info *info, - struct osd_device *osd, - struct osd_inode_id *id, - struct lu_fid *fid) +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; } @@ -3835,7 +3951,7 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, down_read(&obj->oo_ext_idx_sem); } - bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock); + bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); if (bh) { struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; @@ -3859,7 +3975,7 @@ 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); } @@ -3938,10 +4054,11 @@ struct osd_object *osd_object_find(const struct lu_env *env, lu_object_put(env, luch); child = ERR_PTR(-ENOENT); } - } else - child = (void *)luch; + } else { + child = ERR_CAST(luch); + } - return child; + return child; } /** @@ -4090,8 +4207,8 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n", osd_name(osd), PFID(fid), id->oii_ino, id->oii_gen, - (int)PTR_ERR(child_inode)); - RETURN(PTR_ERR(child_inode)); + (int)PTR_ERR(child)); + RETURN(PTR_ERR(child)); } child_inode = igrab(child->oo_inode); } @@ -4423,6 +4540,8 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, file->f_dentry = obj_dentry; file->f_mapping = obj->oo_inode->i_mapping; file->f_op = obj->oo_inode->i_fop; + set_file_inode(file, obj->oo_inode); + lu_object_get(lo); RETURN((struct dt_it *) it); } @@ -4831,7 +4950,7 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, again: if (dev->od_dirent_journal) { - jh = ldiskfs_journal_start_sb(sb, credits); + jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits); if (IS_ERR(jh)) { rc = PTR_ERR(jh); CERROR("%.16s: fail to start trans for dirent " @@ -4861,7 +4980,7 @@ again: } } - bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock); + bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); /* For dot/dotdot entry, if there is not enough space to hold the * FID-in-dirent, just keep them there. It only happens when the * device upgraded from 1.8 or restored from MDT file-level backup. @@ -4892,6 +5011,8 @@ again: 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)) @@ -4975,7 +5096,7 @@ 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 { @@ -5269,11 +5390,6 @@ 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; - } - if (o->od_mnt != NULL) { shrink_dcache_sb(osd_sb(o)); osd_sync(env, &o->od_dt_dev); @@ -5296,7 +5412,10 @@ 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) @@ -5306,14 +5425,7 @@ 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 (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); @@ -5323,7 +5435,7 @@ 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) @@ -5334,7 +5446,7 @@ 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) { @@ -5343,12 +5455,12 @@ static int osd_mount(const struct lu_env *env, } o->od_mnt = vfs_kern_mount(type, s_flags, dev, options); - cfs_module_put(type->owner); + module_put(type->owner); 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); } @@ -5356,32 +5468,39 @@ 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, @@ -5390,8 +5509,8 @@ static struct lu_device *osd_device_fini(const struct lu_env *env, struct osd_device *o = osd_dev(d); ENTRY; - osd_procfs_fini(o); osd_shutdown(env, o); + osd_procfs_fini(o); osd_scrub_cleanup(env, o); osd_obj_map_fini(o); osd_umount(env, o); @@ -5441,6 +5560,9 @@ 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_mnt, rc); @@ -5454,6 +5576,9 @@ static int osd_device_init0(const struct lu_env *env, if (rc != 0) GOTO(out_site, rc); + /* self-repair LMA by default */ + o->od_lma_self_repair = 1; + CFS_INIT_LIST_HEAD(&o->od_ios_list); /* setup scrub, including OI files initialization */ rc = osd_scrub_setup(env, o); @@ -5696,16 +5821,29 @@ static struct obd_ops osd_obd_device_ops = { static int __init osd_mod_init(void) { struct lprocfs_static_vars lvars; + int rc; + + osd_oi_mod_init(); + lprocfs_osd_init_vars(&lvars); - osd_oi_mod_init(); - lprocfs_osd_init_vars(&lvars); - return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars, - LUSTRE_OSD_LDISKFS_NAME, &osd_device_type); + rc = lu_kmem_init(ldiskfs_caches); + if (rc) + return rc; + + rc = class_register_type(&osd_obd_device_ops, NULL, NULL, +#ifndef HAVE_ONLY_PROCFS_SEQ + lvars.module_vars, +#endif + LUSTRE_OSD_LDISKFS_NAME, &osd_device_type); + if (rc) + lu_kmem_fini(ldiskfs_caches); + return rc; } static void __exit osd_mod_exit(void) { class_unregister_type(LUSTRE_OSD_LDISKFS_NAME); + lu_kmem_fini(ldiskfs_caches); } MODULE_AUTHOR("Sun Microsystems, Inc. ");