X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=b542546424c59cda86408f0fe3205b3ef079fcf4;hb=020941416419ab282f3d9b694014b2059d299d51;hp=a66d90ef1b054d55808d7cdb32ee7f91a94bedf0;hpb=2dc1bb8b7a53077fce8632aabe65b2ce8048a550;p=fs%2Flustre-release.git diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index a66d90e..b542546 100644 --- a/lustre/osd-ldiskfs/osd_handler.c +++ b/lustre/osd-ldiskfs/osd_handler.c @@ -41,7 +41,7 @@ * Pravin Shelar : Added fid in dirent */ -#define DEBUG_SUBSYSTEM S_MDS +#define DEBUG_SUBSYSTEM S_OSD #include @@ -62,13 +62,19 @@ /* struct ptlrpc_thread */ #include #include +/* process_config */ +#include #include "osd_internal.h" +#include "osd_dynlocks.h" /* llo_* api support */ #include #include +#include +#include + int ldiskfs_pdo = 1; CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644, "ldiskfs with parallel directory operations"); @@ -77,6 +83,20 @@ 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"; @@ -116,10 +136,6 @@ static int osd_object_invariant(const struct lu_object *l) /* * Concurrency: doesn't matter */ -static int osd_read_locked(const struct lu_env *env, struct osd_object *o) -{ - return osd_oti_get(env)->oti_r_locks > 0; -} /* * Concurrency: doesn't matter @@ -214,8 +230,8 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, id->oii_ino, PTR_ERR(inode)); } else if (id->oii_gen != OSD_OII_NOGEN && inode->i_generation != id->oii_gen) { - CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, " - "gen1 = %u\n", + CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, " + "i_generation = %u\n", id->oii_ino, id->oii_gen, inode->i_generation); iput(inode); inode = ERR_PTR(-ESTALE); @@ -223,7 +239,6 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, /* due to parallel readdir and unlink, * we can have dead inode here. */ CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino); - make_bad_inode(inode); iput(inode); inode = ERR_PTR(-ESTALE); } else if (is_bad_inode(inode)) { @@ -273,6 +288,119 @@ osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, return inode; } +static struct inode *osd_iget_check(struct osd_thread_info *info, + struct osd_device *dev, + const struct lu_fid *fid, + struct osd_inode_id *id, + bool in_oi) +{ + struct inode *inode; + int rc = 0; + ENTRY; + + inode = ldiskfs_iget(osd_sb(dev), id->oii_ino); + if (IS_ERR(inode)) { + rc = PTR_ERR(inode); + if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) { + CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n", + id->oii_ino, rc); + + GOTO(put, rc); + } + + goto check_oi; + } + + if (is_bad_inode(inode)) { + rc = -ENOENT; + if (!in_oi) { + CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino); + + GOTO(put, rc); + } + + goto check_oi; + } + + if (id->oii_gen != OSD_OII_NOGEN && + inode->i_generation != id->oii_gen) { + rc = -ESTALE; + if (!in_oi) { + CDEBUG(D_INODE, "unmatched inode: ino = %u, " + "oii_gen = %u, i_generation = %u\n", + id->oii_ino, id->oii_gen, inode->i_generation); + + GOTO(put, rc); + } + + goto check_oi; + } + + if (inode->i_nlink == 0) { + rc = -ENOENT; + if (!in_oi) { + CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino); + + GOTO(put, rc); + } + + goto check_oi; + } + +check_oi: + if (rc != 0) { + LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc); + + rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); + /* XXX: There are some possible cases: + * 1. rc = 0. + * Backup/restore caused the OI invalid. + * 2. rc = 0. + * Someone unlinked the object but NOT removed + * the OI mapping, such as mount target device + * as ldiskfs, and modify something directly. + * 3. rc = -ENOENT. + * Someone just removed the object between the + * former oi_lookup and the iget. It is normal. + * 4. Other failure cases. + * + * Generally, when the device is mounted, it will + * auto check whether the system is restored from + * file-level backup or not. We trust such detect + * to distinguish the 1st case from the 2nd case. */ + if (rc == 0) { + if (!IS_ERR(inode) && inode->i_generation != 0 && + inode->i_generation == id->oii_gen) + rc = -ENOENT; + else + rc = -EREMCHG; + } + } else { + if (id->oii_gen == OSD_OII_NOGEN) + osd_id_gen(id, inode->i_ino, inode->i_generation); + + /* Do not update file c/mtime in ldiskfs. + * NB: we don't have any lock to protect this because we don't + * have reference on osd_object now, but contention with + * another lookup + attr_set can't happen in the tiny window + * between if (...) and set S_NOCMTIME. */ + if (!(inode->i_flags & S_NOCMTIME)) + inode->i_flags |= S_NOCMTIME; + } + + GOTO(put, rc); + +put: + if (rc != 0) { + if (!IS_ERR(inode)) + iput(inode); + + inode = ERR_PTR(rc); + } + + return inode; +} + /** * \retval +v: new filter_fid, does not contain self-fid * \retval 0: filter_fid_old, contains self-fid @@ -301,28 +429,61 @@ int osd_get_idif(struct osd_thread_info *info, struct inode *inode, return rc; } +static int osd_lma_self_repair(struct osd_thread_info *info, + struct osd_device *osd, struct inode *inode, + const struct lu_fid *fid, __u32 compat) +{ + handle_t *jh; + int rc; + + 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)) { + rc = PTR_ERR(jh); + CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n", + osd_name(osd), rc); + return rc; + } + + rc = osd_ea_fid_set(info, inode, fid, compat, 0); + if (rc != 0) + CWARN("%s: cannot self repair the LMA: rc = %d\n", + osd_name(osd), rc); + ldiskfs_journal_stop(jh); + return 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; + const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu); 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) { + if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) { fid = &lma->lma_self_fid; rc = osd_get_idif(info, inode, dentry, fid); - if (rc > 0) + if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) { + /* 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. */ + osd_lma_self_repair(info, osd, inode, rfid, + fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ? + LMAC_FID_ON_OST : 0); RETURN(0); + } } if (unlikely(rc == -ENODATA)) @@ -337,23 +498,41 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) 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, + "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); + PFID(rfid), inode->i_ino); rc = -EOPNOTSUPP; } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) { fid = &lma->lma_self_fid; } } - if (fid != NULL && - unlikely(!lu_fid_eq(lu_object_fid(&obj->oo_dt.do_lu), fid))) { + if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) { + if (fid_is_idif(rfid) && fid_is_idif(fid)) { + struct ost_id *oi = &info->oti_ostid; + struct lu_fid *fid1 = &info->oti_fid3; + __u32 idx = fid_idif_ost_idx(rfid); + + /* For old IDIF, the OST index is not part of the IDIF, + * Means that different OSTs may have the same IDIFs. + * Under such case, we need to make some compatible + * check to make sure to trigger OI scrub properly. */ + if (idx != 0 && fid_idif_ost_idx(fid) == 0) { + /* Given @rfid is new, LMA is old. */ + fid_to_ostid(fid, oi); + ostid_to_fid(fid1, oi, idx); + if (lu_fid_eq(fid1, rfid)) { + if (osd->od_lma_self_repair) + osd_lma_self_repair(info, osd, + inode, rfid, + LMAC_FID_ON_OST); + RETURN(0); + } + } + } + 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)); + osd_name(osd), PFID(rfid), PFID(fid)); rc = -EREMCHG; } @@ -380,7 +559,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, LINVRNT(osd_invariant(obj)); LASSERT(obj->oo_inode == NULL); - LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid)); + LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid)); dev = osd_dev(ldev); scrub = &dev->od_scrub; @@ -392,30 +571,29 @@ 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; } id = &info->oti_id; - if (!cfs_list_empty(&scrub->os_inconsistent_items)) { + if (!list_empty(&scrub->os_inconsistent_items)) { /* Search order: 2. OI scrub pending list. */ result = osd_oii_lookup(dev, fid, id); if (result == 0) goto iget; } - /* - * 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, OI_CHECK_FLD); if (result == -ENOENT) { @@ -434,37 +612,20 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, in_oi = true; iget: - inode = osd_iget(info, dev, id); + inode = osd_iget_check(info, dev, fid, id, in_oi); if (IS_ERR(inode)) { result = PTR_ERR(inode); if (result == -ENOENT || result == -ESTALE) { - if (!in_oi) { + 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; + GOTO(out, result = -ENOENT); } else if (result == -EREMCHG) { trigger: + if (!in_oi) + fid_zero(&oic->oic_fid); + if (unlikely(triggered)) GOTO(out, result = saved); @@ -473,10 +634,9 @@ trigger: result = -EINPROGRESS; } else if (!dev->od_noscrub) { result = osd_scrub_start(dev); - LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC " - "for "DFID", rc = %d [1]\n", - LDISKFS_SB(osd_sb(dev))->s_es->\ - s_volume_name,PFID(fid), result); + LCONSOLE_WARN("%.16s: trigger OI scrub by RPC " + "for "DFID", rc = %d [1]\n", + osd_name(dev), PFID(fid),result); if (result == 0 || result == -EALREADY) result = -EINPROGRESS; else @@ -519,8 +679,18 @@ trigger: if (result != 0) { iput(inode); obj->oo_inode = NULL; - if (result == -EREMCHG) + if (result == -EREMCHG) { + if (!in_oi) { + result = osd_oi_lookup(info, dev, fid, id, + OI_CHECK_FLD); + if (result != 0) { + fid_zero(&oic->oic_fid); + GOTO(out, result); + } + } + goto trigger; + } GOTO(out, result); } @@ -721,7 +891,7 @@ static int osd_param_is_not_sane(const struct osd_device *dev, { struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super); - return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers; + return oh->ot_credits > osd_transaction_size(dev); } /* @@ -743,11 +913,11 @@ static void osd_trans_commit_cb(struct super_block *sb, dt_txn_hook_commit(th); /* call per-transaction callbacks if any */ - cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) { + list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) { LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC, "commit callback entry: magic=%x name='%s'\n", dcb->dcb_magic, dcb->dcb_name); - cfs_list_del_init(&dcb->dcb_linkage); + list_del_init(&dcb->dcb_linkage); dcb->dcb_func(NULL, th, dcb, error); } @@ -757,34 +927,36 @@ static void osd_trans_commit_cb(struct super_block *sb, lu_context_exit(&th->th_ctx); lu_context_fini(&th->th_ctx); - OBD_FREE_PTR(oh); + thandle_put(th); } static struct thandle *osd_trans_create(const struct lu_env *env, - struct dt_device *d) + struct dt_device *d) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_iobuf *iobuf = &oti->oti_iobuf; - struct osd_thandle *oh; - struct thandle *th; - ENTRY; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_iobuf *iobuf = &oti->oti_iobuf; + struct osd_thandle *oh; + struct thandle *th; + ENTRY; - /* on pending IO in this thread should left from prev. request */ - LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0); + /* on pending IO in this thread should left from prev. request */ + LASSERT(atomic_read(&iobuf->dr_numreqs) == 0); - th = ERR_PTR(-ENOMEM); - OBD_ALLOC_GFP(oh, sizeof *oh, __GFP_IO); - if (oh != NULL) { + th = ERR_PTR(-ENOMEM); + OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS); + if (oh != NULL) { oh->ot_quota_trans = &oti->oti_quota_trans; memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans)); - th = &oh->ot_super; - th->th_dev = d; - th->th_result = 0; - th->th_tags = LCT_TX_HANDLE; - oh->ot_credits = 0; - oti->oti_dev = osd_dt_dev(d); - CFS_INIT_LIST_HEAD(&oh->ot_dcb_list); - osd_th_alloced(oh); + th = &oh->ot_super; + th->th_dev = d; + th->th_result = 0; + th->th_tags = LCT_TX_HANDLE; + oh->ot_credits = 0; + atomic_set(&th->th_refc, 1); + th->th_alloc_size = sizeof(*oh); + oti->oti_dev = osd_dt_dev(d); + INIT_LIST_HEAD(&oh->ot_dcb_list); + osd_th_alloced(oh); memset(oti->oti_declare_ops, 0, sizeof(oti->oti_declare_ops)); @@ -793,8 +965,8 @@ 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; - } - RETURN(th); + } + RETURN(th); } /* @@ -829,11 +1001,9 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, oh->ot_credits, osd_journal(dev)->j_max_transaction_buffers); - CWARN(" create: %u/%u, delete: %u/%u, destroy: %u/%u\n", + CWARN(" create: %u/%u, destroy: %u/%u\n", oti->oti_declare_ops[OSD_OT_CREATE], oti->oti_declare_ops_cred[OSD_OT_CREATE], - oti->oti_declare_ops[OSD_OT_DELETE], - oti->oti_declare_ops_cred[OSD_OT_DELETE], oti->oti_declare_ops[OSD_OT_DESTROY], oti->oti_declare_ops_cred[OSD_OT_DESTROY]); CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n", @@ -851,8 +1021,8 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, CWARN(" insert: %u/%u, delete: %u/%u\n", oti->oti_declare_ops[OSD_OT_INSERT], oti->oti_declare_ops_cred[OSD_OT_INSERT], - oti->oti_declare_ops[OSD_OT_DESTROY], - oti->oti_declare_ops_cred[OSD_OT_DESTROY]); + oti->oti_declare_ops[OSD_OT_DELETE], + oti->oti_declare_ops_cred[OSD_OT_DELETE]); CWARN(" ref_add: %u/%u, ref_del: %u/%u\n", oti->oti_declare_ops[OSD_OT_REF_ADD], oti->oti_declare_ops_cred[OSD_OT_REF_ADD], @@ -871,14 +1041,14 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, * * This should be removed when we can calculate the * credits precisely. */ - oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers; + oh->ot_credits = osd_transaction_size(dev); } /* * 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; @@ -898,24 +1068,45 @@ 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; + + LASSERT(ss != NULL); + LASSERT(ss->ss_server_fld != NULL); + + 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. */ -static int osd_trans_stop(const struct lu_env *env, struct thandle *th) +static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, + struct thandle *th) { - int rc = 0; - struct osd_thandle *oh; - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_iobuf *iobuf = &oti->oti_iobuf; + int rc = 0; + struct osd_thandle *oh; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_iobuf *iobuf = &oti->oti_iobuf; struct qsd_instance *qsd = oti->oti_dev->od_quota_slave; - ENTRY; + struct lquota_trans *qtrans; + ENTRY; - oh = container_of0(th, struct osd_thandle, ot_super); + oh = container_of0(th, struct osd_thandle, ot_super); - if (qsd != NULL) - /* inform the quota slave device that the transaction is - * stopping */ - qsd_op_end(env, qsd, oh->ot_quota_trans); + qtrans = oh->ot_quota_trans; oh->ot_quota_trans = NULL; if (oh->ot_handle != NULL) { @@ -944,24 +1135,28 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th) if (rc != 0) CERROR("Failure to stop transaction: %d\n", rc); } else { - OBD_FREE_PTR(oh); + thandle_put(&oh->ot_super); } - /* 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; + /* inform the quota slave device that the transaction is stopping */ + qsd_op_end(env, qsd, qtrans); - RETURN(rc); + /* 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, + atomic_read(&iobuf->dr_numreqs) == 0); + osd_fini_iobuf(oti->oti_dev, iobuf); + if (!rc) + rc = iobuf->dr_error; + + RETURN(rc); } static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb) @@ -971,7 +1166,7 @@ static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb) LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC); LASSERT(&dcb->dcb_func != NULL); - cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list); + list_add(&dcb->dcb_linkage, &oh->ot_dcb_list); return 0; } @@ -997,8 +1192,8 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l) osd_index_fini(obj); if (inode != NULL) { struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave; - qid_t uid = inode->i_uid; - qid_t gid = inode->i_gid; + qid_t uid = i_uid_read(inode); + qid_t gid = i_gid_read(inode); iput(inode); obj->oo_inode = NULL; @@ -1046,6 +1241,8 @@ static int osd_object_print(const struct lu_env *env, void *cookie, d ? d->id_ops->id_name : "plain"); } +#define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */ + /* * Concurrency: shouldn't matter. */ @@ -1070,24 +1267,28 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, } spin_lock(&osd->od_osfs_lock); - /* cache 1 second */ - if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) { - result = sb->s_op->statfs(sb->s_root, ksfs); - if (likely(result == 0)) { /* N.B. statfs can't really fail */ - osd->od_osfs_age = cfs_time_current_64(); - statfs_pack(&osd->od_statfs, ksfs); - if (sb->s_flags & MS_RDONLY) - sfs->os_state = OS_STATE_READONLY; - } + result = sb->s_op->statfs(sb->s_root, ksfs); + if (likely(result == 0)) { /* N.B. statfs can't really fail */ + statfs_pack(sfs, ksfs); + if (sb->s_flags & MS_RDONLY) + sfs->os_state = OS_STATE_READONLY; } - if (likely(result == 0)) - *sfs = osd->od_statfs; spin_unlock(&osd->od_osfs_lock); - if (unlikely(env == NULL)) + if (unlikely(env == NULL)) OBD_FREE_PTR(ksfs); + /* Reserve a small amount of space for local objects like last_rcvd, + * llog, quota files, ... */ + if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) { + sfs->os_bavail = 0; + } else { + sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS; + /** Take out metadata overhead for indirect blocks */ + sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3); + } + return result; } @@ -1110,11 +1311,11 @@ static void osd_conf_get(const struct lu_env *env, struct dt_device_param *param) { struct super_block *sb = osd_sb(osd_dt_dev(dev)); + int ea_overhead; /* * 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; @@ -1134,13 +1335,19 @@ static void osd_conf_get(const struct lu_env *env, if (test_opt(sb, POSIX_ACL)) param->ddp_mntopts |= MNTOPT_ACL; + /* LOD might calculate the max stripe count based on max_ea_size, + * so we need take account in the overhead as well, + * xattr_header + magic + xattr_entry_head */ + ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) + + LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN); + #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE) - if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE)) - param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE; - else + if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE)) + param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE - + ea_overhead; + else #endif - param->ddp_max_ea_size = sb->s_blocksize; - + param->ddp_max_ea_size = sb->s_blocksize - ea_overhead; } /* @@ -1229,55 +1436,57 @@ static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d, * If we mount with --data_journal we may need more. */ const int osd_dto_credits_noquota[DTO_NR] = { - /** - * Insert/Delete. - * INDEX_EXTRA_TRANS_BLOCKS(8) + - * SINGLEDATA_TRANS_BLOCKS(8) - * XXX Note: maybe iam need more, since iam have more level than - * EXT3 htree. - */ - [DTO_INDEX_INSERT] = 16, - [DTO_INDEX_DELETE] = 16, - /** + /** + * Insert. + * INDEX_EXTRA_TRANS_BLOCKS(8) + + * SINGLEDATA_TRANS_BLOCKS(8) + * XXX Note: maybe iam need more, since iam have more level than + * EXT3 htree. + */ + [DTO_INDEX_INSERT] = 16, + /** + * Delete + * just modify a single entry, probably merge few within a block + */ + [DTO_INDEX_DELETE] = 1, + /** * Used for OI scrub - */ - [DTO_INDEX_UPDATE] = 16, - /** - * Create a object. The same as create object in EXT3. - * DATA_TRANS_BLOCKS(14) + - * INDEX_EXTRA_BLOCKS(8) + - * 3(inode bits, groups, GDT) - */ - [DTO_OBJECT_CREATE] = 25, - /** - * XXX: real credits to be fixed - */ - [DTO_OBJECT_DELETE] = 25, - /** - * Attr set credits (inode) - */ - [DTO_ATTR_SET_BASE] = 1, - /** - * Xattr set. The same as xattr of EXT3. - * DATA_TRANS_BLOCKS(14) - * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS - * are also counted in. Do not know why? - */ - [DTO_XATTR_SET] = 14, - [DTO_LOG_REC] = 14, - /** - * credits for inode change during write. - */ - [DTO_WRITE_BASE] = 3, - /** - * credits for single block write. - */ - [DTO_WRITE_BLOCK] = 14, - /** - * Attr set credits for chown. - * This is extra credits for setattr, and it is null without quota - */ - [DTO_ATTR_SET_CHOWN]= 0 + */ + [DTO_INDEX_UPDATE] = 16, + /** + * 4(inode, inode bits, groups, GDT) + * notice: OI updates are counted separately with DTO_INDEX_INSERT + */ + [DTO_OBJECT_CREATE] = 4, + /** + * 4(inode, inode bits, groups, GDT) + * notice: OI updates are counted separately with DTO_INDEX_DELETE + */ + [DTO_OBJECT_DELETE] = 4, + /** + * Attr set credits (inode) + */ + [DTO_ATTR_SET_BASE] = 1, + /** + * Xattr set. The same as xattr of EXT3. + * DATA_TRANS_BLOCKS(14) + * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS + * are also counted in. Do not know why? + */ + [DTO_XATTR_SET] = 14, + /** + * credits for inode change during write. + */ + [DTO_WRITE_BASE] = 3, + /** + * credits for single block write. + */ + [DTO_WRITE_BLOCK] = 14, + /** + * Attr set credits for chown. + * This is extra credits for setattr, and it is null without quota + */ + [DTO_ATTR_SET_CHOWN] = 0 }; static const struct dt_device_operations osd_dt_ops = { @@ -1479,26 +1688,26 @@ static struct timespec *osd_inode_time(const struct lu_env *env, static void osd_inode_getattr(const struct lu_env *env, - struct inode *inode, struct lu_attr *attr) -{ - attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | - LA_SIZE | LA_BLOCKS | LA_UID | LA_GID | - LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE | - LA_TYPE; - - attr->la_atime = LTIME_S(inode->i_atime); - attr->la_mtime = LTIME_S(inode->i_mtime); - attr->la_ctime = LTIME_S(inode->i_ctime); - attr->la_mode = inode->i_mode; - attr->la_size = i_size_read(inode); - attr->la_blocks = inode->i_blocks; - attr->la_uid = inode->i_uid; - attr->la_gid = inode->i_gid; - attr->la_flags = LDISKFS_I(inode)->i_flags; - attr->la_nlink = inode->i_nlink; - attr->la_rdev = inode->i_rdev; - attr->la_blksize = 1 << inode->i_blkbits; - attr->la_blkbits = inode->i_blkbits; + struct inode *inode, struct lu_attr *attr) +{ + attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | + LA_SIZE | LA_BLOCKS | LA_UID | LA_GID | + LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE | + LA_TYPE; + + attr->la_atime = LTIME_S(inode->i_atime); + attr->la_mtime = LTIME_S(inode->i_mtime); + attr->la_ctime = LTIME_S(inode->i_ctime); + attr->la_mode = inode->i_mode; + attr->la_size = i_size_read(inode); + attr->la_blocks = inode->i_blocks; + attr->la_uid = i_uid_read(inode); + attr->la_gid = i_gid_read(inode); + attr->la_flags = LDISKFS_I(inode)->i_flags; + attr->la_nlink = inode->i_nlink; + attr->la_rdev = inode->i_rdev; + attr->la_blksize = 1 << inode->i_blkbits; + attr->la_blkbits = inode->i_blkbits; } static int osd_attr_get(const struct lu_env *env, @@ -1529,9 +1738,11 @@ static int osd_declare_attr_set(const struct lu_env *env, struct osd_object *obj; struct osd_thread_info *info = osd_oti_get(env); struct lquota_id_info *qi = &info->oti_qi; + qid_t uid; + qid_t gid; long long bspace; int rc = 0; - bool allocated; + bool enforce; ENTRY; LASSERT(dt != NULL); @@ -1559,27 +1770,28 @@ static int osd_declare_attr_set(const struct lu_env *env, * We still need to call the osd_declare_qid() to calculate the journal * credits for updating quota accounting files and to trigger quota * space adjustment once the operation is completed.*/ - if ((attr->la_valid & LA_UID) != 0 && - attr->la_uid != obj->oo_inode->i_uid) { + if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) { + /* USERQUOTA */ + uid = i_uid_read(obj->oo_inode); qi->lqi_type = USRQUOTA; - + enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid); /* inode accounting */ qi->lqi_is_blk = false; - /* one more inode for the new owner ... */ + /* one more inode for the new uid ... */ qi->lqi_id.qid_uid = attr->la_uid; qi->lqi_space = 1; - allocated = (attr->la_uid == 0) ? true : false; - rc = osd_declare_qid(env, oh, qi, allocated, NULL); + /* Reserve credits for the new uid */ + rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); /* and one less inode for the current uid */ - qi->lqi_id.qid_uid = obj->oo_inode->i_uid; + qi->lqi_id.qid_uid = uid; qi->lqi_space = -1; - rc = osd_declare_qid(env, oh, qi, true, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) @@ -1588,47 +1800,49 @@ static int osd_declare_attr_set(const struct lu_env *env, /* block accounting */ qi->lqi_is_blk = true; - /* more blocks for the new owner ... */ + /* more blocks for the new uid ... */ qi->lqi_id.qid_uid = attr->la_uid; qi->lqi_space = bspace; - allocated = (attr->la_uid == 0) ? true : false; - rc = osd_declare_qid(env, oh, qi, allocated, NULL); + /* + * Credits for the new uid has been reserved, re-use "obj" + * to save credit reservation. + */ + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); - /* and finally less blocks for the current owner */ - qi->lqi_id.qid_uid = obj->oo_inode->i_uid; + /* and finally less blocks for the current uid */ + qi->lqi_id.qid_uid = uid; qi->lqi_space = -bspace; - rc = osd_declare_qid(env, oh, qi, true, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); - } - if (attr->la_valid & LA_GID && - attr->la_gid != obj->oo_inode->i_gid) { + /* GROUP QUOTA */ + gid = i_gid_read(obj->oo_inode); qi->lqi_type = GRPQUOTA; + enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid); /* inode accounting */ qi->lqi_is_blk = false; - /* one more inode for the new group owner ... */ + /* one more inode for the new gid ... */ qi->lqi_id.qid_gid = attr->la_gid; qi->lqi_space = 1; - allocated = (attr->la_gid == 0) ? true : false; - rc = osd_declare_qid(env, oh, qi, allocated, NULL); + rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); /* and one less inode for the current gid */ - qi->lqi_id.qid_gid = obj->oo_inode->i_gid; + qi->lqi_id.qid_gid = gid; qi->lqi_space = -1; - rc = osd_declare_qid(env, oh, qi, true, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) @@ -1637,20 +1851,19 @@ static int osd_declare_attr_set(const struct lu_env *env, /* block accounting */ qi->lqi_is_blk = true; - /* more blocks for the new owner ... */ + /* more blocks for the new gid ... */ qi->lqi_id.qid_gid = attr->la_gid; qi->lqi_space = bspace; - allocated = (attr->la_gid == 0) ? true : false; - rc = osd_declare_qid(env, oh, qi, allocated, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); - /* and finally less blocks for the current owner */ - qi->lqi_id.qid_gid = obj->oo_inode->i_gid; + /* and finally less blocks for the current gid */ + qi->lqi_id.qid_gid = gid; qi->lqi_space = -bspace; - rc = osd_declare_qid(env, oh, qi, true, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) @@ -1661,11 +1874,16 @@ static int osd_declare_attr_set(const struct lu_env *env, } static int osd_inode_setattr(const struct lu_env *env, - struct inode *inode, const struct lu_attr *attr) + struct inode *inode, const struct lu_attr *attr) { - __u64 bits; + __u64 bits = attr->la_valid; + + /* Only allow set size for regular file */ + if (!S_ISREG(inode->i_mode)) + bits &= ~(LA_SIZE | LA_BLOCKS); - bits = attr->la_valid; + if (bits == 0) + return 0; if (bits & LA_ATIME) inode->i_atime = *osd_inode_time(env, inode, attr->la_atime); @@ -1684,17 +1902,17 @@ static int osd_inode_setattr(const struct lu_env *env, if (bits & LA_BLOCKS) inode->i_blocks = attr->la_blocks; #endif - if (bits & LA_MODE) - inode->i_mode = (inode->i_mode & S_IFMT) | - (attr->la_mode & ~S_IFMT); - if (bits & LA_UID) - inode->i_uid = attr->la_uid; - if (bits & LA_GID) - inode->i_gid = attr->la_gid; - if (bits & LA_NLINK) + if (bits & LA_MODE) + inode->i_mode = (inode->i_mode & S_IFMT) | + (attr->la_mode & ~S_IFMT); + if (bits & LA_UID) + i_uid_write(inode, attr->la_uid); + if (bits & LA_GID) + i_gid_write(inode, attr->la_gid); + if (bits & LA_NLINK) set_nlink(inode, attr->la_nlink); - if (bits & LA_RDEV) - inode->i_rdev = attr->la_rdev; + if (bits & LA_RDEV) + inode->i_rdev = attr->la_rdev; if (bits & LA_FLAGS) { /* always keep S_NOCMTIME */ @@ -1706,24 +1924,25 @@ static int osd_inode_setattr(const struct lu_env *env, static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) { - if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) || - (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) { + if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) || + (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) { struct iattr iattr; int rc; + ll_vfs_dq_init(inode); iattr.ia_valid = 0; if (attr->la_valid & LA_UID) iattr.ia_valid |= ATTR_UID; if (attr->la_valid & LA_GID) iattr.ia_valid |= ATTR_GID; - iattr.ia_uid = attr->la_uid; - iattr.ia_gid = attr->la_gid; + iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid); + iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid); rc = ll_vfs_dq_transfer(inode, &iattr); if (rc) { CERROR("%s: quota transfer failed: rc = %d. Is quota " - "enforcement enabled on the ldiskfs filesystem?", - inode->i_sb->s_id, rc); + "enforcement enabled on the ldiskfs " + "filesystem?\n", inode->i_sb->s_id, rc); return rc; } } @@ -1775,7 +1994,6 @@ static int osd_attr_set(const struct lu_env *env, } inode = obj->oo_inode; - ll_vfs_dq_init(inode); rc = osd_quota_transfer(inode, attr); if (rc) @@ -2000,7 +2218,6 @@ static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah, { LASSERT(ah); - memset(ah, 0, sizeof(*ah)); ah->dah_parent = parent; ah->dah_mode = child_mode; } @@ -2036,7 +2253,7 @@ static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, * should not happen since quota enforcement is no longer * enabled on ldiskfs (lquota takes care of it). */ - LASSERTF(result == 0, "%d", result); + LASSERTF(result == 0, "%d\n", result); ll_dirty_inode(inode, I_DIRTY_DATASYNC); } @@ -2083,31 +2300,34 @@ 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, OI_CHECK_FLD); + 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 @@ -2118,12 +2338,8 @@ 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); } /* @@ -2136,7 +2352,6 @@ 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; @@ -2148,12 +2363,10 @@ 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), 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, - osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); + /* Reuse idle OI block may cause additional one OI block + * to be changed. */ + osd_trans_declare_op(env, oh, OSD_OT_INSERT, + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); /* If this is directory, then we expect . and .. to be inserted as * well. The one directory block always needs to be created for the @@ -2170,20 +2383,10 @@ static int osd_declare_object_create(const struct lu_env *env, RETURN(0); rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh, - false, false, NULL, false); + osd_dt_obj(dt), false, NULL, false); 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); } @@ -2250,13 +2453,13 @@ static int osd_declare_object_destroy(const struct lu_env *env, osd_trans_declare_op(env, oh, OSD_OT_DELETE, osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3); /* one less inode */ - rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh, - false, true, NULL, false); + rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), + -1, oh, obj, false, NULL, false); if (rc) RETURN(rc); /* data to be truncated */ - rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, - true, true, NULL, false); + rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), + 0, oh, obj, true, NULL, false); RETURN(rc); } @@ -2281,7 +2484,8 @@ 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); @@ -2297,7 +2501,8 @@ 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, OI_CHECK_FLD); + 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) */ @@ -2369,9 +2574,9 @@ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, * its inmemory API. */ void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param, - const struct dt_rec *fid) + const struct lu_fid *fid) { - if (!fid_is_namespace_visible((const struct lu_fid *)fid) || + if (!fid_is_namespace_visible(fid) || OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) { param->edp_magic = 0; return; @@ -2408,9 +2613,9 @@ static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj, static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, struct inode *dir, - struct inode *parent_dir, - const struct dt_rec *dot_fid, - const struct dt_rec *dot_dot_fid, + struct inode *parent_dir, + const struct lu_fid *dot_fid, + const struct lu_fid *dot_dot_fid, struct osd_thandle *oth) { struct ldiskfs_dentry_param *dot_ldp; @@ -2432,6 +2637,7 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, struct osd_device *osd, struct osd_object *pobj, const struct lu_fid *fid, + __u32 type, struct thandle *th) { struct osd_thread_info *info = osd_oti_get(env); @@ -2444,9 +2650,7 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, oh = container_of(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle->h_transaction != NULL); - /* FIXME: Insert index api needs to know the mode of - * the remote object. Just use S_IFDIR for now */ - local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR); + local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type); if (IS_ERR(local)) { CERROR("%s: create local error %d\n", osd_name(osd), (int)PTR_ERR(local)); @@ -2461,9 +2665,12 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, RETURN(ERR_PTR(rc)); } + if (!S_ISDIR(type)) + RETURN(local); + rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode, - (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu), - (const struct dt_rec *)fid, oh); + lu_object_fid(&pobj->oo_dt.do_lu), + fid, oh); if (rc != 0) { CERROR("%s: "DFID" add dot dotdot error: rc = %d\n", osd_name(osd), PFID(fid), rc); @@ -2577,20 +2784,24 @@ 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. - * + CDEBUG(D_INODE, DFID" increase nlink %d\n", + PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink); + /* * 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 @@ -2602,28 +2813,17 @@ 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))) { + if (unlikely(inode->i_nlink == 0)) /* 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) { ; } */ - - LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX); + else { + ldiskfs_inc_count(oh->ot_handle, inode); + if (!S_ISDIR(inode->i_mode)) + 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; @@ -2653,37 +2853,43 @@ 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; } + CDEBUG(D_INODE, DFID" decrease nlink %d\n", + PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink); + + 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; } @@ -2738,16 +2944,36 @@ static int osd_declare_xattr_set(const struct lu_env *env, int fl, struct thandle *handle) { struct osd_thandle *oh; + int credits; + struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, - strcmp(name, XATTR_NAME_VERSION) == 0 ? - osd_dto_credits_noquota[DTO_ATTR_SET_BASE] : - osd_dto_credits_noquota[DTO_XATTR_SET]); + /* optimistic optimization: LMA is set first and usually fit inode */ + if (strcmp(name, XATTR_NAME_LMA) == 0) { + if (dt_object_exists(dt)) + credits = 0; + else + credits = 1; + } else if (strcmp(name, XATTR_NAME_VERSION) == 0) { + credits = 1; + } else { + credits = osd_dto_credits_noquota[DTO_XATTR_SET]; + if (buf && buf->lb_len > sb->s_blocksize) { + credits *= (buf->lb_len + sb->s_blocksize - 1) >> + sb->s_blocksize_bits; + } + /* + * xattr set may involve inode quota change, reserve credits for + * dquot_initialize() + */ + oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb); + } + + osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits); return 0; } @@ -2797,6 +3023,9 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) return -EACCES; + CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n", + PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len); + osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); if (fl & LU_XATTR_REPLACE) fs_flags |= XATTR_REPLACE; @@ -2804,6 +3033,22 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, if (fl & LU_XATTR_CREATE) fs_flags |= XATTR_CREATE; + if (strcmp(name, XATTR_NAME_LMV) == 0) { + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + int rc; + + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); + if (rc != 0) + RETURN(rc); + + lma->lma_incompat |= LMAI_STRIPED; + lustre_lma_swab(lma); + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, + sizeof(*lma), XATTR_REPLACE); + if (rc != 0) + RETURN(rc); + } + return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len, fs_flags); } @@ -2821,7 +3066,6 @@ static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL); - LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj)); if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) return -EACCES; @@ -2836,6 +3080,7 @@ static int osd_declare_xattr_del(const struct lu_env *env, struct thandle *handle) { struct osd_thandle *oh; + struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev)); LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(handle != NULL); @@ -2845,6 +3090,11 @@ static int osd_declare_xattr_del(const struct lu_env *env, osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, osd_dto_credits_noquota[DTO_XATTR_SET]); + /* + * xattr del may involve inode quota change, reserve credits for + * dquot_initialize() + */ + oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb); return 0; } @@ -2911,16 +3161,16 @@ static struct obd_capa *osd_capa_get(const struct lu_env *env, 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_uid = i_uid_read(obj->oo_inode); + capa->lc_gid = i_gid_read(obj->oo_inode); capa->lc_flags = LC_ID_PLAIN; break; case LC_ID_CONVERT: { __u32 d[4], s[4]; - s[0] = obj->oo_inode->i_uid; + s[0] = i_uid_read(obj->oo_inode); cfs_get_random_bytes(&(s[1]), sizeof(__u32)); - s[2] = obj->oo_inode->i_gid; + s[2] = i_uid_read(obj->oo_inode); 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)) @@ -2964,7 +3214,8 @@ static struct obd_capa *osd_capa_get(const struct lu_env *env, RETURN(oc); } -static int osd_object_sync(const struct lu_env *env, struct dt_object *dt) +static int osd_object_sync(const struct lu_env *env, struct dt_object *dt, + __u64 start, __u64 end) { struct osd_object *obj = osd_dt_obj(dt); struct inode *inode = obj->oo_inode; @@ -2980,13 +3231,20 @@ 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; -#ifndef HAVE_FILE_FSYNC_4ARGS + set_file_inode(file, inode); + +#ifdef HAVE_FILE_FSYNC_4ARGS + rc = file->f_op->fsync(file, start, end, 0); +#elif defined(HAVE_FILE_FSYNC_2ARGS) mutex_lock(&inode->i_mutex); -#endif - rc = do_fsync(file, 0); -#ifndef HAVE_FILE_FSYNC_4ARGS + rc = file->f_op->fsync(file, 0); + mutex_unlock(&inode->i_mutex); +#else + mutex_lock(&inode->i_mutex); + rc = file->f_op->fsync(file, dentry, 0); mutex_unlock(&inode->i_mutex); #endif + RETURN(rc); } @@ -3108,7 +3366,8 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, * recheck under lock. */ if (!osd_has_index(obj)) - result = osd_iam_container_init(env, obj, dir); + result = osd_iam_container_init(env, obj, + obj->oo_dir); else result = 0; up_write(&obj->oo_ext_idx_sem); @@ -3304,8 +3563,8 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, inode = osd_dt_obj(dt)->oo_inode; LASSERT(inode); - rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, - true, true, NULL, false); + rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), + 0, oh, osd_dt_obj(dt), true, NULL, false); RETURN(rc); } @@ -3323,25 +3582,26 @@ 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) + const 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 FLD is not being initialized yet, it only happens during the + * initialization, likely during mgs initialization, and we assume + * this is local FID. */ + if (ss == NULL || ss->ss_server_fld == NULL) + RETURN(0); - RETURN(ss->ss_node_id != range->lsr_index); + /* Only check the local FLDB here */ + if (osd_seq_exists(env, osd, fid_seq(fid))) + RETURN(0); + + RETURN(1); } /** @@ -3396,7 +3656,7 @@ 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; @@ -3662,9 +3922,9 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, * \retval -ve, on error */ static int __osd_ea_add_rec(struct osd_thread_info *info, - struct osd_object *pobj, struct inode *cinode, - const char *name, const struct dt_rec *fid, - struct htree_lock *hlock, struct thandle *th) + struct osd_object *pobj, struct inode *cinode, + const char *name, const struct lu_fid *fid, + struct htree_lock *hlock, struct thandle *th) { struct ldiskfs_dentry_param *ldp; struct dentry *child; @@ -3703,11 +3963,11 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, * \retval -ve, on error */ static int osd_add_dot_dotdot(struct osd_thread_info *info, - struct osd_object *dir, - struct inode *parent_dir, const char *name, - const struct dt_rec *dot_fid, - const struct dt_rec *dot_dot_fid, - struct thandle *th) + struct osd_object *dir, + struct inode *parent_dir, const char *name, + const struct lu_fid *dot_fid, + const struct lu_fid *dot_dot_fid, + struct thandle *th) { struct inode *inode = dir->oo_inode; struct osd_thandle *oth; @@ -3735,8 +3995,8 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, } result = osd_add_dot_dotdot_internal(info, dir->oo_inode, - parent_dir, dot_fid, - dot_dot_fid, oth); + parent_dir, dot_fid, + dot_dot_fid, oth); if (result == 0) dir->oo_compat_dotdot_created = 1; } @@ -3750,8 +4010,8 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, * value, return by respective functions. */ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, - struct inode *cinode, const char *name, - const struct dt_rec *fid, struct thandle *th) + struct inode *cinode, const char *name, + const struct lu_fid *fid, struct thandle *th) { struct osd_thread_info *info = osd_oti_get(env); struct htree_lock *hlock; @@ -3766,9 +4026,10 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, pobj->oo_inode, 0); } else { down_write(&pobj->oo_ext_idx_sem); - } - rc = osd_add_dot_dotdot(info, pobj, cinode, name, - (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu), + } + + rc = osd_add_dot_dotdot(info, pobj, cinode, name, + lu_object_fid(&pobj->oo_dt.do_lu), fid, th); } else { if (hlock != NULL) { @@ -3781,11 +4042,10 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) { struct lu_fid *tfid = &info->oti_fid; - *tfid = *(const struct lu_fid *)fid; + *tfid = *fid; tfid->f_ver = ~0; rc = __osd_ea_add_rec(info, pobj, cinode, name, - (const struct dt_rec *)tfid, - hlock, th); + tfid, hlock, th); } else { rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, hlock, th); @@ -3834,13 +4094,11 @@ again: } if (!dev->od_noscrub && ++once == 1) { - CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n", - PFID(fid)); rc = osd_scrub_start(dev); - LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID - ", rc = %d [2]\n", - LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, - PFID(fid), rc); + LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID + ", rc = %d [2]\n", + LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, + PFID(fid), rc); if (rc == 0) goto again; } @@ -3880,11 +4138,89 @@ int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd, 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; } /** + * Get parent FID from the linkEA. + * + * For a directory which parent resides on remote MDT, to satisfy the + * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On + * the other hand, to make the lookup(..) on the directory can return + * the real parent FID, we append the real parent FID after its ".." + * name entry in the /REMOTE_PARENT_DIR. + * + * Unfortunately, such PFID-in-dirent cannot be preserved via file-level + * backup. So after the restore, we cannot get the right parent FID from + * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since + * we have stored the real parent FID in the directory object's linkEA, + * we can parse the linkEA for the real parent FID. + * + * \param[in] env pointer to the thread context + * \param[in] obj pointer to the object to be handled + * \param[out]fid pointer to the buffer to hold the parent FID + * + * \retval 0 for getting the real parent FID successfully + * \retval negative error number on failure + */ +static int osd_get_pfid_from_linkea(const struct lu_env *env, + struct osd_object *obj, + struct lu_fid *fid) +{ + struct osd_thread_info *oti = osd_oti_get(env); + struct lu_buf *buf = &oti->oti_big_buf; + struct dentry *dentry = &oti->oti_obj_dentry; + struct inode *inode = obj->oo_inode; + struct linkea_data ldata = { 0 }; + int rc; + ENTRY; + + fid_zero(fid); + if (!S_ISDIR(inode->i_mode)) + RETURN(-EIO); + +again: + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, + buf->lb_buf, buf->lb_len); + if (rc == -ERANGE) { + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, + NULL, 0); + if (rc > 0) { + lu_buf_realloc(buf, rc); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + + goto again; + } + } + + if (unlikely(rc == 0)) + RETURN(-ENODATA); + + if (rc < 0) + RETURN(rc); + + if (unlikely(buf->lb_buf == NULL)) { + lu_buf_realloc(buf, rc); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + + goto again; + } + + ldata.ld_buf = buf; + rc = linkea_init(&ldata); + if (rc == 0) { + linkea_first_entry(&ldata); + linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid); + } + + RETURN(rc); +} + +/** * Calls ->lookup() to find dentry. From dentry get inode and * read inode's ea to get fid. This is required for interoperability * mode (b11826) @@ -3918,7 +4254,7 @@ 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; @@ -3938,15 +4274,26 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, /* done with de, release bh */ brelse(bh); - if (rc != 0) - rc = osd_ea_fid_get(env, obj, ino, fid, id); - else + if (rc != 0) { + if (unlikely(ino == osd_remote_parent_ino(dev))) + /* If the parent is on remote MDT, and there + * is no FID-in-dirent, then we have to get + * the parent FID from the linkEA. */ + rc = osd_get_pfid_from_linkea(env, obj, fid); + else + 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); } + if (osd_remote_fid(env, dev, fid)) + GOTO(out, rc = 0); + rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid); if (rc != 0) @@ -3998,15 +4345,17 @@ struct osd_object *osd_object_find(const struct lu_env *env, * in the cache, otherwise lu_object_alloc() crashes * -bzzz */ - luch = lu_object_find_at(env, ludev, fid, NULL); - if (!IS_ERR(luch)) { - if (lu_object_exists(luch)) { - lo = lu_object_locate(luch->lo_header, ludev->ld_type); - if (lo != NULL) - child = osd_obj(lo); - else - LU_OBJECT_DEBUG(D_ERROR, env, luch, - "lu_object can't be located" + luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ? + ludev : ludev->ld_site->ls_top_dev, + fid, NULL); + if (!IS_ERR(luch)) { + if (lu_object_exists(luch)) { + lo = lu_object_locate(luch->lo_header, ludev->ld_type); + if (lo != NULL) + child = osd_obj(lo); + else + LU_OBJECT_DEBUG(D_ERROR, env, luch, + "lu_object can't be located" DFID"\n", PFID(fid)); if (child == NULL) { @@ -4021,10 +4370,11 @@ 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; } /** @@ -4050,7 +4400,7 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, int rc; ENTRY; - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); + LASSERT(!dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); @@ -4059,22 +4409,15 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, osd_trans_declare_op(env, oh, OSD_OT_INSERT, osd_dto_credits_noquota[DTO_INDEX_INSERT]); - if (osd_dt_obj(dt)->oo_inode == NULL) { - const char *name = (const char *)key; - /* Object is not being created yet. Only happens when - * 1. declare directory create - * 2. declare insert . - * 3. declare insert .. - */ - LASSERT(strcmp(name, dotdot) == 0 || strcmp(name, dot) == 0); - } else { + if (osd_dt_obj(dt)->oo_inode != NULL) { struct inode *inode = osd_dt_obj(dt)->oo_inode; /* We ignore block quota on meta pool (MDTs), so needn't * calculate how many blocks will be consumed by this index * insert */ - rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, - oh, true, true, NULL, false); + rc = osd_declare_inode_qid(env, i_uid_read(inode), + i_gid_read(inode), 0, oh, + osd_dt_obj(dt), true, NULL, false); } if (fid == NULL) @@ -4114,7 +4457,8 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, { struct osd_object *obj = osd_dt_obj(dt); struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct lu_fid *fid = (struct lu_fid *) rec; + struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; + const struct lu_fid *fid = rec1->rec_fid; const char *name = (const char *)key; struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; @@ -4123,16 +4467,16 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, int rc; ENTRY; - LASSERT(osd_invariant(obj)); + LASSERT(osd_invariant(obj)); LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(th != NULL); + LASSERT(th != NULL); osd_trans_exec_op(env, th, OSD_OT_INSERT); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) - RETURN(-EACCES); + if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) + RETURN(-EACCES); - LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid)); + LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid)); rc = osd_remote_fid(env, osd, fid); if (rc < 0) { @@ -4161,8 +4505,7 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, child_inode = igrab(omm->omm_remote_parent->d_inode); } else { child_inode = osd_create_local_agent_inode(env, osd, - obj, fid, - th); + obj, fid, rec1->rec_type & S_IFMT, th); if (IS_ERR(child_inode)) RETURN(PTR_ERR(child_inode)); } @@ -4173,13 +4516,16 @@ 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); } - rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th); + rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th); + + CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n", + obj->oo_inode->i_ino, name, child_inode->i_ino, rc); iput(child_inode); if (child != NULL) @@ -4506,6 +4852,8 @@ 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); } @@ -4559,6 +4907,12 @@ static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di) { } +struct osd_filldir_cbs { +#ifdef HAVE_DIR_CONTEXT + struct dir_context ctx; +#endif + struct osd_it_ea *it; +}; /** * It is called internally by ->readdir(). It fills the * iterator's in-memory data structure with required @@ -4570,11 +4924,11 @@ static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di) * \retval 0 on success * \retval 1 on buffer full */ -static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen, +static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen, loff_t offset, __u64 ino, unsigned d_type) { - struct osd_it_ea *it = (struct osd_it_ea *)buf; + struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it; struct osd_object *obj = it->oie_obj; struct osd_it_ea_dirent *ent = it->oie_dirent; struct lu_fid *fid = &ent->oied_fid; @@ -4629,6 +4983,7 @@ static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen, * * \retval 0 on success * \retval -ve on error + * \retval +1 reach the end of entry */ static int osd_ldiskfs_it_fill(const struct lu_env *env, const struct dt_it *di) @@ -4637,7 +4992,14 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, struct osd_object *obj = it->oie_obj; struct inode *inode = obj->oo_inode; struct htree_lock *hlock = NULL; - int result = 0; + struct file *filp = &it->oie_file; + int rc = 0; + struct osd_filldir_cbs buf = { +#ifdef HAVE_DIR_CONTEXT + .ctx.actor = osd_ldiskfs_filldir, +#endif + .it = it + }; ENTRY; it->oie_dirent = it->oie_buf; @@ -4651,22 +5013,31 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, down_read(&obj->oo_ext_idx_sem); } - result = inode->i_fop->readdir(&it->oie_file, it, - (filldir_t) osd_ldiskfs_filldir); +#ifdef HAVE_DIR_CONTEXT + buf.ctx.pos = filp->f_pos; + rc = inode->i_fop->iterate(filp, &buf.ctx); + filp->f_pos = buf.ctx.pos; +#else + rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir); +#endif if (hlock != NULL) ldiskfs_htree_unlock(hlock); else up_read(&obj->oo_ext_idx_sem); - if (it->oie_rd_dirent == 0) { - result = -EIO; - } else { - it->oie_dirent = it->oie_buf; - it->oie_it_dirent = 1; - } + if (it->oie_rd_dirent == 0) { + /*If it does not get any dirent, it means it has been reached + *to the end of the dir */ + it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file); + if (rc == 0) + rc = 1; + } else { + it->oie_dirent = it->oie_buf; + it->oie_it_dirent = 1; + } - RETURN(result); + RETURN(rc); } /** @@ -4746,22 +5117,12 @@ osd_dirent_update(handle_t *jh, struct super_block *sb, LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack)); rc = ldiskfs_journal_get_write_access(jh, bh); - if (rc != 0) { - CERROR("%.16s: fail to write access for update dirent: " - "name = %.*s, rc = %d\n", - LDISKFS_SB(sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + if (rc != 0) RETURN(rc); - } rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid); rc = ldiskfs_journal_dirty_metadata(jh, bh); - if (rc != 0) - CERROR("%.16s: fail to dirty metadata for update dirent: " - "name = %.*s, rc = %d\n", - LDISKFS_SB(sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); RETURN(rc); } @@ -4810,13 +5171,8 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, if (osd_dirent_has_space(de->rec_len, ent->oied_namelen, dir->i_sb->s_blocksize)) { rc = ldiskfs_journal_get_write_access(jh, bh); - if (rc != 0) { - CERROR("%.16s: fail to write access for reinsert " - "dirent: name = %.*s, rc = %d\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + if (rc != 0) RETURN(rc); - } de->name[de->name_len] = 0; rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); @@ -4825,38 +5181,29 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, de->file_type |= LDISKFS_DIRENT_LUFID; rc = ldiskfs_journal_dirty_metadata(jh, bh); - if (rc != 0) - CERROR("%.16s: fail to dirty metadata for reinsert " - "dirent: name = %.*s, rc = %d\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); RETURN(rc); } rc = ldiskfs_delete_entry(jh, dir, de, bh); - if (rc != 0) { - CERROR("%.16s: fail to delete entry for reinsert dirent: " - "name = %.*s, rc = %d\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + if (rc != 0) RETURN(rc); - } dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name, ent->oied_namelen); ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp; - osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid); + osd_get_ldiskfs_dirent_param(ldp, fid); dentry->d_fsdata = (void *)ldp; ll_vfs_dq_init(dir); rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock); /* It is too bad, we cannot reinsert the name entry back. * That means we lose it! */ if (rc != 0) - CERROR("%.16s: fail to insert entry for reinsert dirent: " - "name = %.*s, rc = %d\n", + CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, " + "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n", LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, PFID(fid), rc); RETURN(rc); } @@ -4914,13 +5261,14 @@ 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 " - "check_repair: credits %d, name %.*s, rc %d\n", - devname, credits, ent->oied_namelen, - ent->oied_name, rc); + CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent " + "check_repair, dir = %lu/%u, credits = %d, " + "name = %.*s: rc = %d\n", + devname, dir->i_ino, dir->i_generation, credits, + ent->oied_namelen, ent->oied_name, rc); RETURN(rc); } @@ -4944,7 +5292,7 @@ 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. @@ -4964,15 +5312,17 @@ again: if (rc == -ENOENT || rc == -ESTALE) { *attr |= LUDA_IGNORE; rc = 0; + } else { + CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent " + "check_repair, dir = %lu/%u, name = %.*s: " + "rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, rc); } 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)); @@ -5005,6 +5355,13 @@ again: rc = osd_dirent_update(jh, sb, ent, fid, bh, de); if (rc == 0) *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%.16s: fail to update FID " + "in the dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } else { /* Do not repair under dryrun mode. */ if (*attr & LUDA_VERIFY_DRYRUN) { @@ -5031,6 +5388,13 @@ again: fid, bh, de, hlock); if (rc == 0) *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%.16s: fail to append FID " + "after the dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } } else if (rc == -ENODATA) { /* Do not repair under dryrun mode. */ @@ -5063,6 +5427,13 @@ again: rc = osd_ea_fid_set(info, inode, fid, 0, 0); if (rc == 0) *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%.16s: fail to set LMA for " + "update dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } else { lu_igif_build(fid, inode->i_ino, inode->i_generation); /* It is probably IGIF object. Only aappend the @@ -5071,6 +5442,13 @@ again: fid, bh, de, hlock); if (rc == 0) *attr |= LUDA_UPGRADE; + else + CDEBUG(D_LFSCK, "%.16s: fail to append IGIF " + "after the dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } } @@ -5124,11 +5502,15 @@ static inline int osd_it_ea_rec(const struct lu_env *env, int rc = 0; ENTRY; + LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode); + if (attr & LUDA_VERIFY) { - attr |= LUDA_TYPE; - if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) { + if (unlikely(ino == osd_remote_parent_ino(dev))) { attr |= LUDA_IGNORE; - rc = 0; + /* If the parent is on remote MDT, and there + * is no FID-in-dirent, then we have to get + * the parent FID from the linkEA. */ + osd_get_pfid_from_linkea(env, obj, fid); } else { rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr); @@ -5136,10 +5518,19 @@ static inline int osd_it_ea_rec(const struct lu_env *env, } else { attr &= ~LU_DIRENT_ATTRS_MASK; if (!fid_is_sane(fid)) { - if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) + if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) && + likely(it->oie_dirent->oied_namelen != 2 || + it->oie_dirent->oied_name[0] != '.' || + it->oie_dirent->oied_name[1] != '.')) RETURN(-ENOENT); - rc = osd_ea_fid_get(env, obj, ino, fid, id); + if (unlikely(ino == osd_remote_parent_ino(dev))) + /* If the parent is on remote MDT, and there + * is no FID-in-dirent, then we have to get + * the parent FID from the linkEA. */ + rc = osd_get_pfid_from_linkea(env, obj, fid); + else + rc = osd_ea_fid_get(env, obj, ino, fid, id); } else { osd_id_gen(id, ino, OSD_OII_NOGEN); } @@ -5171,6 +5562,27 @@ static inline int osd_it_ea_rec(const struct lu_env *env, } /** + * Returns the record size size at current position. + * + * This function will return record(lu_dirent) size in bytes. + * + * \param[in] env execution environment + * \param[in] di iterator's in memory structure + * \param[in] attr attribute of the entry, only requires LUDA_TYPE to + * calculate the lu_dirent size. + * + * \retval record size(in bytes & in memory) of the current lu_dirent + * entry. + */ +static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di, + __u32 attr) +{ + struct osd_it_ea *it = (struct osd_it_ea *)di; + + return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr); +} + +/** * Returns a cookie for current position of the iterator head, so that * user can use this cookie to load/start the iterator next time. * @@ -5205,6 +5617,9 @@ static int osd_it_ea_load(const struct lu_env *env, it->oie_file.f_pos = hash; rc = osd_ldiskfs_it_fill(env, di); + if (rc > 0) + rc = -ENODATA; + if (rc == 0) rc = +1; @@ -5245,23 +5660,24 @@ static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt, * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826) */ static const struct dt_index_operations osd_index_ea_ops = { - .dio_lookup = osd_index_ea_lookup, - .dio_declare_insert = osd_index_declare_ea_insert, - .dio_insert = osd_index_ea_insert, - .dio_declare_delete = osd_index_declare_ea_delete, - .dio_delete = osd_index_ea_delete, - .dio_it = { - .init = osd_it_ea_init, - .fini = osd_it_ea_fini, - .get = osd_it_ea_get, - .put = osd_it_ea_put, - .next = osd_it_ea_next, - .key = osd_it_ea_key, - .key_size = osd_it_ea_key_size, - .rec = osd_it_ea_rec, - .store = osd_it_ea_store, - .load = osd_it_ea_load - } + .dio_lookup = osd_index_ea_lookup, + .dio_declare_insert = osd_index_declare_ea_insert, + .dio_insert = osd_index_ea_insert, + .dio_declare_delete = osd_index_declare_ea_delete, + .dio_delete = osd_index_ea_delete, + .dio_it = { + .init = osd_it_ea_init, + .fini = osd_it_ea_fini, + .get = osd_it_ea_get, + .put = osd_it_ea_put, + .next = osd_it_ea_next, + .key = osd_it_ea_key, + .key_size = osd_it_ea_key_size, + .rec = osd_it_ea_rec, + .rec_size = osd_it_ea_rec_size, + .store = osd_it_ea_store, + .load = osd_it_ea_load + } }; static void *osd_key_init(const struct lu_context *ctx, @@ -5302,6 +5718,7 @@ static void osd_key_fini(const struct lu_context *ctx, OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); lu_buf_free(&info->oti_iobuf.dr_pg_buf); lu_buf_free(&info->oti_iobuf.dr_bl_buf); + lu_buf_free(&info->oti_big_buf); OBD_FREE_PTR(info); } @@ -5337,6 +5754,43 @@ static int osd_device_init(const struct lu_env *env, struct lu_device *d, return osd_procfs_init(osd, name); } +static int osd_fid_init(const struct lu_env *env, struct osd_device *osd) +{ + struct seq_server_site *ss = osd_seq_site(osd); + int rc; + ENTRY; + + if (osd->od_is_ost || osd->od_cl_seq != NULL) + RETURN(0); + + if (unlikely(ss == NULL)) + RETURN(-ENODEV); + + OBD_ALLOC_PTR(osd->od_cl_seq); + if (osd->od_cl_seq == NULL) + RETURN(-ENOMEM); + + rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA, + osd->od_svname, ss->ss_server_seq); + + if (rc != 0) { + OBD_FREE_PTR(osd->od_cl_seq); + osd->od_cl_seq = NULL; + } + + RETURN(rc); +} + +static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd) +{ + if (osd->od_cl_seq == NULL) + return; + + seq_client_fini(osd->od_cl_seq); + OBD_FREE_PTR(osd->od_cl_seq); + osd->od_cl_seq = NULL; +} + static int osd_shutdown(const struct lu_env *env, struct osd_device *o) { ENTRY; @@ -5347,6 +5801,8 @@ static int osd_shutdown(const struct lu_env *env, struct osd_device *o) o->od_quota_slave = NULL; } + osd_fid_fini(env, o); + RETURN(0); } @@ -5354,11 +5810,6 @@ 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); @@ -5394,13 +5845,6 @@ 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, GFP_IOFS); if (__page == NULL) GOTO(out, rc = -ENOMEM); @@ -5431,7 +5875,7 @@ 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); @@ -5449,13 +5893,22 @@ static int osd_mount(const struct lu_env *env, #endif if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb, - LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) { + LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) { CERROR("%s: device %s is mounted w/o journal\n", name, dev); GOTO(out_mnt, rc = -EINVAL); } +#ifdef LDISKFS_MOUNT_DIRDATA + if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb, + LDISKFS_FEATURE_INCOMPAT_DIRDATA)) + LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA; + else if (!o->od_is_ost) + CWARN("%s: device %s was upgraded from Lustre-1.x without " + "enabling the dirdata feature. If you do not want to " + "downgrade to Lustre-1.x again, you can enable it via " + "'tune2fs -O dirdata device'\n", name, dev); +#endif 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) { @@ -5475,8 +5928,6 @@ out_mnt: out: if (__page) OBD_PAGE_FREE(__page); - if (rc) - fsfilt_put_ops(o->od_fsops); return rc; } @@ -5487,8 +5938,8 @@ 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); @@ -5517,7 +5968,6 @@ static int osd_device_init0(const struct lu_env *env, spin_lock_init(&o->od_osfs_lock); mutex_init(&o->od_otable_mutex); - o->od_osfs_age = cfs_time_shift_64(-1000); o->od_capa_hash = init_capa_hash(); if (o->od_capa_hash == NULL) @@ -5527,17 +5977,20 @@ static int osd_device_init0(const struct lu_env *env, o->od_writethrough_cache = 1; o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE; - rc = osd_mount(env, o, cfg); - if (rc) - GOTO(out_capa, rc); - cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4), sizeof(o->od_svname)); if (cplen >= sizeof(o->od_svname)) { rc = -E2BIG; - GOTO(out_mnt, rc); + GOTO(out_capa, rc); } + if (server_name_is_ost(o->od_svname)) + o->od_is_ost = 1; + + rc = osd_mount(env, o, cfg); + if (rc != 0) + GOTO(out_capa, rc); + rc = osd_obj_map_init(env, o); if (rc != 0) GOTO(out_mnt, rc); @@ -5551,7 +6004,10 @@ static int osd_device_init0(const struct lu_env *env, if (rc != 0) GOTO(out_site, rc); - CFS_INIT_LIST_HEAD(&o->od_ios_list); + /* self-repair LMA by default */ + o->od_lma_self_repair = 1; + + INIT_LIST_HEAD(&o->od_ios_list); /* setup scrub, including OI files initialization */ rc = osd_scrub_setup(env, o); if (rc < 0) @@ -5643,23 +6099,32 @@ static struct lu_device *osd_device_free(const struct lu_env *env, static int osd_process_config(const struct lu_env *env, struct lu_device *d, struct lustre_cfg *cfg) { - struct osd_device *o = osd_dev(d); - int err; - ENTRY; + struct osd_device *o = osd_dev(d); + int rc; + ENTRY; - switch(cfg->lcfg_command) { - case LCFG_SETUP: - err = osd_mount(env, o, cfg); - break; - case LCFG_CLEANUP: + switch (cfg->lcfg_command) { + case LCFG_SETUP: + rc = osd_mount(env, o, cfg); + break; + case LCFG_CLEANUP: lu_dev_del_linkage(d->ld_site, d); - err = osd_shutdown(env, o); + rc = osd_shutdown(env, o); break; - default: - err = -ENOSYS; - } + case LCFG_PARAM: + LASSERT(&o->od_dt_dev); + rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars, + cfg, &o->od_dt_dev); + if (rc > 0 || rc == -ENOSYS) + rc = class_process_proc_param(PARAM_OST, + lprocfs_osd_obd_vars, + cfg, &o->od_dt_dev); + break; + default: + rc = -ENOSYS; + } - RETURN(err); + RETURN(rc); } static int osd_recovery_complete(const struct lu_env *env, @@ -5737,13 +6202,26 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, int result = 0; ENTRY; - if (osd->od_quota_slave != NULL) + if (osd->od_quota_slave != NULL) { /* set up quota slave objects */ result = qsd_prepare(env, osd->od_quota_slave); + if (result != 0) + RETURN(result); + } + + result = osd_fid_init(env, osd); RETURN(result); } +int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp, + struct lu_fid *fid, struct md_op_data *op_data) +{ + struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev); + + return seq_client_alloc_fid(env, osd->od_cl_seq, fid); +} + static const struct lu_object_operations osd_lu_obj_ops = { .loo_object_init = osd_object_init, .loo_object_delete = osd_object_delete, @@ -5787,22 +6265,32 @@ struct lu_device_type osd_device_type = { static struct obd_ops osd_obd_device_ops = { .o_owner = THIS_MODULE, .o_connect = osd_obd_connect, - .o_disconnect = osd_obd_disconnect + .o_disconnect = osd_obd_disconnect, + .o_fid_alloc = osd_fid_alloc, }; static int __init osd_mod_init(void) { - struct lprocfs_static_vars lvars; + int rc; + + osd_oi_mod_init(); - 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, true, + lprocfs_osd_module_vars, + 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. ");