X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_handler.c;h=47b97cf2cee60d3a1bc0db671e25733711e08a9f;hp=68f98d00f372badde23d69e39eddd7df491b4de3;hb=6fd75f264c5f5c186bbfe559e1a98fb3769d8128;hpb=6a3599a9389da07ac435d94873376c9f86edea9d diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c index 68f98d0..47b97cf 100644 --- a/lustre/mdt/mdt_handler.c +++ b/lustre/mdt/mdt_handler.c @@ -27,7 +27,6 @@ */ /* * This file is part of Lustre, http://www.lustre.org/ - * Lustre is a trademark of Sun Microsystems, Inc. * * lustre/mdt/mdt_handler.c * @@ -164,6 +163,13 @@ void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm) lh->mlh_type = MDT_REG_LOCK; } +void mdt_lh_reg_init(struct mdt_lock_handle *lh, struct ldlm_lock *lock) +{ + mdt_lock_reg_init(lh, lock->l_req_mode); + if (lock->l_req_mode == LCK_GROUP) + lh->mlh_gid = lock->l_policy_data.l_inodebits.li_gid; +} + void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode, const struct lu_name *lname) { @@ -264,11 +270,47 @@ static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o EXIT; } +/** + * Check whether \a o is directory stripe object. + * + * \param[in] info thread environment + * \param[in] o MDT object + * + * \retval 1 is directory stripe. + * \retval 0 isn't directory stripe. + * \retval < 1 error code + */ +static int mdt_is_dir_stripe(struct mdt_thread_info *info, + struct mdt_object *o) +{ + struct md_attr *ma = &info->mti_attr; + struct lmv_mds_md_v1 *lmv; + int rc; + + rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV); + if (rc < 0) + return rc; + + if (!(ma->ma_valid & MA_LMV)) + return 0; + + lmv = &ma->ma_lmv->lmv_md_v1; + + if (!lmv_is_sane2(lmv)) + return -EBADF; + + if (le32_to_cpu(lmv->lmv_magic) == LMV_MAGIC_STRIPE) + return 1; + + return 0; +} + static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset, struct lu_fid *fid) { struct mdt_device *mdt = info->mti_mdt; struct lu_name *lname = &info->mti_name; + const char *start = fileset; char *filename = info->mti_filename; struct mdt_object *parent; u32 mode; @@ -283,8 +325,8 @@ static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset, */ *fid = mdt->mdt_md_root_fid; - while (rc == 0 && fileset != NULL && *fileset != '\0') { - const char *s1 = fileset; + while (rc == 0 && start != NULL && *start != '\0') { + const char *s1 = start; const char *s2; while (*++s1 == '/') @@ -296,7 +338,7 @@ static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset, if (s2 == s1) break; - fileset = s2; + start = s2; lname->ln_namelen = s2 - s1; if (lname->ln_namelen > NAME_MAX) { @@ -332,9 +374,21 @@ static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset, rc = PTR_ERR(parent); else { mode = lu_object_attr(&parent->mot_obj); - mdt_object_put(info->mti_env, parent); - if (!S_ISDIR(mode)) + if (!S_ISDIR(mode)) { rc = -ENOTDIR; + } else if (mdt_is_remote_object(info, parent, parent)) { + if (!mdt->mdt_enable_remote_subdir_mount) { + rc = -EREMOTE; + LCONSOLE_WARN("%s: subdir mount '%s' refused because 'enable_remote_subdir_mount=0': rc = %d\n", + mdt_obd_name(mdt), + fileset, rc); + } else { + LCONSOLE_INFO("%s: subdir mount '%s' is remote and may be slow\n", + mdt_obd_name(mdt), + fileset); + } + } + mdt_object_put(info->mti_env, parent); } } @@ -416,6 +470,7 @@ static int mdt_statfs(struct tgt_session_info *tsi) struct mdt_body *reqbody = NULL; struct mdt_statfs_cache *msf; ktime_t kstart = ktime_get(); + int current_blockbits; int rc; ENTRY; @@ -472,6 +527,15 @@ static int mdt_statfs(struct tgt_session_info *tsi) spin_unlock(&mdt->mdt_lock); } + /* tgd_blockbit is recordsize bits set during mkfs. + * This once set does not change. However, 'zfs set' + * can be used to change the MDT blocksize. Instead + * of using cached value of 'tgd_blockbit' always + * calculate the blocksize bits which may have + * changed. + */ + current_blockbits = fls64(osfs->os_bsize) - 1; + /* at least try to account for cached pages. its still racy and * might be under-reporting if clients haven't announced their * caches with brw recently */ @@ -479,12 +543,12 @@ static int mdt_statfs(struct tgt_session_info *tsi) " pending %llu free %llu avail %llu\n", tgd->tgd_tot_dirty, tgd->tgd_tot_granted, tgd->tgd_tot_pending, - osfs->os_bfree << tgd->tgd_blockbits, - osfs->os_bavail << tgd->tgd_blockbits); + osfs->os_bfree << current_blockbits, + osfs->os_bavail << current_blockbits); osfs->os_bavail -= min_t(u64, osfs->os_bavail, ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending + - osfs->os_bsize - 1) >> tgd->tgd_blockbits)); + osfs->os_bsize - 1) >> current_blockbits)); tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__); CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; " @@ -493,15 +557,15 @@ static int mdt_statfs(struct tgt_session_info *tsi) osfs->os_files, osfs->os_ffree, osfs->os_state); if (!exp_grant_param_supp(tsi->tsi_exp) && - tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) { + current_blockbits > COMPAT_BSIZE_SHIFT) { /* clients which don't support OBD_CONNECT_GRANT_PARAM * should not see a block size > page size, otherwise * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12) * block size which is the biggest block size known to work * with all client's page size. */ - osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT; - osfs->os_bfree <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT; - osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT; + osfs->os_blocks <<= current_blockbits - COMPAT_BSIZE_SHIFT; + osfs->os_bfree <<= current_blockbits - COMPAT_BSIZE_SHIFT; + osfs->os_bavail <<= current_blockbits - COMPAT_BSIZE_SHIFT; osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT; } if (rc == 0) @@ -638,6 +702,7 @@ int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody, if (buf->lb_len == 0) RETURN(0); + LASSERT(!info->mti_big_acl_used); again: rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS); if (rc < 0) { @@ -647,10 +712,9 @@ again: rc = 0; } else if (rc == -EOPNOTSUPP) { rc = 0; - } else { - if (rc == -ERANGE && - exp_connect_large_acl(info->mti_exp) && - buf->lb_buf != info->mti_big_acl) { + } else if (rc == -ERANGE) { + if (exp_connect_large_acl(info->mti_exp) && + !info->mti_big_acl_used) { if (info->mti_big_acl == NULL) { info->mti_big_aclsize = min_t(unsigned int, @@ -676,47 +740,19 @@ again: buf->lb_buf = info->mti_big_acl; buf->lb_len = info->mti_big_aclsize; - + info->mti_big_acl_used = 1; goto again; } - + /* FS has ACL bigger that our limits */ + CDEBUG(D_INODE, "%s: "DFID" ACL can't fit into %d\n", + mdt_obd_name(mdt), PFID(mdt_object_fid(o)), + info->mti_big_aclsize); + rc = -E2BIG; + } else { CERROR("%s: unable to read "DFID" ACL: rc = %d\n", mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc); } } else { - int client; - int server; - int acl_buflen; - int lmm_buflen = 0; - int lmmsize = 0; - - acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER); - if (acl_buflen >= rc) - goto map; - - /* If LOV/LMA EA is small, we can reuse part of their buffer */ - client = ptlrpc_req_get_repsize(pill->rc_req); - server = lustre_packed_msg_size(pill->rc_req->rq_repmsg); - if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) { - lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD, - RCL_SERVER); - lmmsize = repbody->mbo_eadatasize; - } - - if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) { - CDEBUG(D_INODE, "%s: client prepared buffer size %d " - "is not big enough with the ACL size %d (%d)\n", - mdt_obd_name(mdt), client, rc, - server - acl_buflen - lmm_buflen + rc + lmmsize); - repbody->mbo_aclsize = 0; - repbody->mbo_valid &= ~OBD_MD_FLACL; - RETURN(-ERANGE); - } - -map: - if (buf->lb_buf == info->mti_big_acl) - info->mti_big_acl_used = 1; - rc = nodemap_map_acl(nodemap, buf->lb_buf, rc, NODEMAP_FS_TO_CLIENT); /* if all ACLs mapped out, rc is still >= 0 */ @@ -1396,10 +1432,10 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, rc = mdt_attr_get_complex(info, o, ma); if (unlikely(rc)) { - CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR, - "%s: getattr error for "DFID": rc = %d\n", - mdt_obd_name(info->mti_mdt), - PFID(mdt_object_fid(o)), rc); + CDEBUG_LIMIT(rc == -ENOENT ? D_OTHER : D_ERROR, + "%s: getattr error for "DFID": rc = %d\n", + mdt_obd_name(info->mti_mdt), + PFID(mdt_object_fid(o)), rc); RETURN(rc); } @@ -1418,6 +1454,7 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, if (mdt_body_has_lov(la, reqbody)) { u32 stripe_count = 1; + bool fixed_layout = false; if (ma->ma_valid & MA_LOV) { LASSERT(ma->ma_lmm_size); @@ -1442,6 +1479,7 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA); stripe_count = le32_to_cpu(lmv->lmv_stripe_count); + fixed_layout = lmv_is_fixed(lmv); if (magic == LMV_MAGIC_STRIPE && lmv_is_restriping(lmv)) mdt_restripe_migrate_add(info, o); else if (magic == LMV_MAGIC_V1 && @@ -1473,7 +1511,8 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, !fid_is_root(mdt_object_fid(o)) && mdt->mdt_enable_dir_auto_split && !o->mot_restriping && - stripe_count < atomic_read(&mdt->mdt_mds_mds_conns) + 1) + stripe_count < atomic_read(&mdt->mdt_mds_mds_conns) + 1 && + !fixed_layout) mdt_auto_split_add(info, o); } else if (S_ISLNK(la->la_mode) && reqbody->mbo_valid & OBD_MD_LINKNAME) { @@ -1804,30 +1843,62 @@ out: static int mdt_raw_lookup(struct mdt_thread_info *info, struct mdt_object *parent, - const struct lu_name *lname, - struct ldlm_reply *ldlm_rep) + const struct lu_name *lname) { - struct lu_fid *child_fid = &info->mti_tmp_fid1; - int rc; + struct lu_fid *fid = &info->mti_tmp_fid1; + struct mdt_body *repbody; + bool is_dotdot = false; + bool is_old_parent_stripe = false; + bool is_new_parent_checked = false; + int rc; + ENTRY; LASSERT(!info->mti_cross_ref); + /* Always allow to lookup ".." */ + if (lname->ln_namelen == 2 && + lname->ln_name[0] == '.' && lname->ln_name[1] == '.') { + info->mti_spec.sp_permitted = 1; + is_dotdot = true; + if (mdt_is_dir_stripe(info, parent) == 1) + is_old_parent_stripe = true; + } + mdt_object_get(info->mti_env, parent); +lookup: /* Only got the fid of this obj by name */ - fid_zero(child_fid); - rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object), - lname, child_fid, &info->mti_spec); - if (rc == 0) { - struct mdt_body *repbody; + fid_zero(fid); + rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname, fid, + &info->mti_spec); + mdt_object_put(info->mti_env, parent); + if (rc) + RETURN(rc); + + /* getattr_name("..") should return master object FID for striped dir */ + if (is_dotdot && (is_old_parent_stripe || !is_new_parent_checked)) { + parent = mdt_object_find(info->mti_env, info->mti_mdt, fid); + if (IS_ERR(parent)) + RETURN(PTR_ERR(parent)); + + /* old client getattr_name("..") with stripe FID */ + if (unlikely(is_old_parent_stripe)) { + is_old_parent_stripe = false; + goto lookup; + } + + /* ".." may be a stripe */ + if (unlikely(mdt_is_dir_stripe(info, parent) == 1)) { + is_new_parent_checked = true; + goto lookup; + } - repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); - repbody->mbo_fid1 = *child_fid; - repbody->mbo_valid = OBD_MD_FLID; - mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS); - } else if (rc == -ENOENT) { - mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG); + mdt_object_put(info->mti_env, parent); } + repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); + repbody->mbo_fid1 = *fid; + repbody->mbo_valid = OBD_MD_FLID; + RETURN(rc); } @@ -1987,14 +2058,8 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, } if (lu_name_is_valid(lname)) { - /* Always allow to lookup ".." */ - if (unlikely(lname->ln_namelen == 2 && - lname->ln_name[0] == '.' && - lname->ln_name[1] == '.')) - info->mti_spec.sp_permitted = 1; - if (info->mti_body->mbo_valid == OBD_MD_FLID) { - rc = mdt_raw_lookup(info, parent, lname, ldlm_rep); + rc = mdt_raw_lookup(info, parent, lname); RETURN(rc); } @@ -2129,6 +2194,16 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, PLDLMRES(lock->l_resource), PFID(mdt_object_fid(child))); + if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_PTLRPC_ENQ_RESEND))) { + if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)) + OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_ENQ_RESEND, + req->rq_deadline - + req->rq_arrival_time.tv_sec + + cfs_fail_val ?: 3); + /* Put the lock to the waiting list and force the cancel */ + ldlm_set_ast_sent(lock); + } + if (S_ISREG(lu_object_attr(&child->mot_obj)) && !mdt_object_remote(child) && child != parent) { mdt_object_put(info->mti_env, child); @@ -2288,7 +2363,7 @@ static int mdt_rmfid_check_permission(struct mdt_thread_info *info, if (la->la_flags & LUSTRE_IMMUTABLE_FL) rc = -EACCES; - if (md_capable(uc, CFS_CAP_DAC_OVERRIDE)) + if (cap_raised(uc->uc_cap, CAP_DAC_OVERRIDE)) RETURN(0); if (uc->uc_fsuid == la->la_uid) { if ((la->la_mode & S_IWUSR) == 0) @@ -2459,7 +2534,7 @@ static int mdt_set_info(struct tgt_session_info *tsi) tgt_name(tsi->tsi_tgt), vallen); RETURN(-EINVAL); } - if (ptlrpc_req_need_swab(req)) { + if (req_capsule_req_need_swab(&req->rq_pill)) { __swab64s(&cs->cs_recno); __swab32s(&cs->cs_id); } @@ -2515,7 +2590,7 @@ static int mdt_readpage(struct tgt_session_info *tsi) exp_max_brw_size(tsi->tsi_exp)); rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >> PAGE_SHIFT; - OBD_ALLOC_PTR_ARRAY(rdpg->rp_pages, rdpg->rp_npages); + OBD_ALLOC_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages); if (rdpg->rp_pages == NULL) RETURN(-ENOMEM); @@ -2539,7 +2614,7 @@ free_rdpg: for (i = 0; i < rdpg->rp_npages; i++) if (rdpg->rp_pages[i] != NULL) __free_page(rdpg->rp_pages[i]); - OBD_FREE_PTR_ARRAY(rdpg->rp_pages, rdpg->rp_npages); + OBD_FREE_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages); if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) RETURN(0); @@ -2936,6 +3011,7 @@ static int mdt_quotactl(struct tgt_session_info *tsi) case LUSTRE_Q_SETDEFAULT: case LUSTRE_Q_SETQUOTAPOOL: case LUSTRE_Q_SETINFOPOOL: + case LUSTRE_Q_SETDEFAULT_POOL: if (!nodemap_can_setquota(nodemap)) GOTO(out_nodemap, rc = -EPERM); /* fallthrough */ @@ -2944,6 +3020,7 @@ static int mdt_quotactl(struct tgt_session_info *tsi) case LUSTRE_Q_GETDEFAULT: case LUSTRE_Q_GETQUOTAPOOL: case LUSTRE_Q_GETINFOPOOL: + case LUSTRE_Q_GETDEFAULT_POOL: if (qmt == NULL) GOTO(out_nodemap, rc = -EOPNOTSUPP); /* slave quotactl */ @@ -3002,6 +3079,8 @@ static int mdt_quotactl(struct tgt_session_info *tsi) case LUSTRE_Q_GETQUOTAPOOL: case LUSTRE_Q_SETINFOPOOL: case LUSTRE_Q_GETINFOPOOL: + case LUSTRE_Q_SETDEFAULT_POOL: + case LUSTRE_Q_GETDEFAULT_POOL: /* forward quotactl request to QMT */ rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl); break; @@ -3528,6 +3607,7 @@ int mdt_object_local_lock(struct mdt_thread_info *info, struct mdt_object *o, policy->l_inodebits.bits = *ibits; policy->l_inodebits.try_bits = trybits; + policy->l_inodebits.li_gid = lh->mlh_gid; /* * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is @@ -4675,6 +4755,23 @@ static int mdt_intent_policy(const struct lu_env *env, } else { rc = err_serious(-EFAULT); } + } else if (ldesc->l_resource.lr_type == LDLM_IBITS && + ldesc->l_policy_data.l_inodebits.bits == MDS_INODELOCK_DOM) { + struct ldlm_reply *rep; + + /* No intent was provided but INTENT flag is set along with + * DOM bit, this is considered as GLIMPSE request. + * This logic is common for MDT and OST glimpse + */ + mdt_ptlrpc_stats_update(req, IT_GLIMPSE); + rc = mdt_glimpse_enqueue(info, ns, lockp, flags); + /* Check whether the reply has been packed successfully. */ + if (req->rq_repmsg != NULL) { + rep = req_capsule_server_get(info->mti_pill, + &RMF_DLM_REP); + rep->lock_policy_res2 = + ptlrpc_status_hton(rep->lock_policy_res2); + } } else { /* No intent was provided */ req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0); @@ -5428,6 +5525,9 @@ TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_PUNCH, mdt_punch_hdl, mdt_hp_punch), TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, mdt_data_sync), +TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE, + mdt_fallocate_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek), }; static struct tgt_handler mdt_sec_ctx_ops[] = { @@ -5654,8 +5754,9 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, m->mdt_skip_lfsck = 1; } - /* DoM files get IO lock at open optionally by default */ - m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN; + /* Just try to get a DoM lock by default. Otherwise, having a group + * lock granted, it may get blocked for a long time. */ + m->mdt_opts.mo_dom_lock = TRYLOCK_DOM_ON_OPEN; /* DoM files are read at open and data is packed in the reply */ m->mdt_opts.mo_dom_read_open = 1; @@ -5673,6 +5774,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, m->mdt_enable_chprojid_gid = 0; m->mdt_enable_remote_rename = 1; m->mdt_dir_restripe_nsonly = 1; + m->mdt_enable_remote_subdir_mount = 1; atomic_set(&m->mdt_mds_mds_conns, 0); atomic_set(&m->mdt_async_commit_count, 0); @@ -5728,8 +5830,13 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, LDLM_NAMESPACE_SERVER, LDLM_NAMESPACE_GREEDY, LDLM_NS_TYPE_MDT); - if (m->mdt_namespace == NULL) - GOTO(err_fini_seq, rc = -ENOMEM); + if (IS_ERR(m->mdt_namespace)) { + rc = PTR_ERR(m->mdt_namespace); + CERROR("%s: unable to create server namespace: rc = %d\n", + obd->obd_name, rc); + m->mdt_namespace = NULL; + GOTO(err_fini_seq, rc); + } m->mdt_namespace->ns_lvbp = m; m->mdt_namespace->ns_lvbo = &mdt_lvbo; @@ -6290,11 +6397,7 @@ static int mdt_connect_internal(const struct lu_env *env, if (OCD_HAS_FLAG(data, CKSUM)) { __u32 cksum_types = data->ocd_cksum_types; - /* The client set in ocd_cksum_types the checksum types it - * supports. We have to mask off the algorithms that we don't - * support */ - data->ocd_cksum_types &= - obd_cksum_types_supported_server(obd_name); + tgt_mask_cksum_types(&mdt->mdt_lut, &data->ocd_cksum_types); if (unlikely(data->ocd_cksum_types == 0)) { CERROR("%s: Connect with checksum support but no " @@ -6320,6 +6423,9 @@ static int mdt_connect_internal(const struct lu_env *env, mdt_enable_slc(mdt); } + if (!mdt->mdt_lut.lut_dt_conf.ddp_has_lseek_data_hole) + data->ocd_connect_flags2 &= ~OBD_CONNECT2_LSEEK; + return 0; } @@ -6412,7 +6518,8 @@ static int mdt_export_cleanup(struct obd_export *exp) rc = mdt_ctxt_add_dirty_flag(&env, info, mfd); /* Don't unlink orphan on failover umount, LU-184 */ - if (exp->exp_flags & OBD_OPT_FAILOVER) { + if (exp->exp_flags & OBD_OPT_FAILOVER || + exp->exp_obd->obd_stopping) { ma->ma_valid = MA_FLAGS; ma->ma_attr_flags |= MDS_KEEP_ORPHAN; } @@ -6696,18 +6803,19 @@ static int mdt_path_current(struct mdt_thread_info *info, struct getinfo_fid2path *fp, struct lu_fid *root_fid) { - struct mdt_device *mdt = info->mti_mdt; - struct mdt_object *mdt_obj; - struct link_ea_header *leh; - struct link_ea_entry *lee; - struct lu_name *tmpname = &info->mti_name; - struct lu_fid *tmpfid = &info->mti_tmp_fid1; - struct lu_buf *buf = &info->mti_big_buf; - char *ptr; - int reclen; - struct linkea_data ldata = { NULL }; - int rc = 0; - bool first = true; + struct mdt_device *mdt = info->mti_mdt; + struct lu_name *tmpname = &info->mti_name; + struct lu_fid *tmpfid = &info->mti_tmp_fid1; + struct lu_buf *buf = &info->mti_big_buf; + struct linkea_data ldata = { NULL }; + bool first = true; + struct mdt_object *mdt_obj; + struct link_ea_header *leh; + struct link_ea_entry *lee; + char *ptr; + int reclen; + int rc = 0; + ENTRY; /* temp buffer for path element, the buffer will be finally freed @@ -6723,8 +6831,6 @@ static int mdt_path_current(struct mdt_thread_info *info, *tmpfid = fp->gf_fid = *mdt_object_fid(obj); while (!lu_fid_eq(root_fid, &fp->gf_fid)) { - struct lu_buf lmv_buf; - if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) && lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid)) GOTO(out, rc = -ENOENT); @@ -6772,22 +6878,14 @@ static int mdt_path_current(struct mdt_thread_info *info, fp->gf_linkno++; } - lmv_buf.lb_buf = info->mti_xattr_buf; - lmv_buf.lb_len = sizeof(info->mti_xattr_buf); /* Check if it is slave stripes */ - rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj), - &lmv_buf, XATTR_NAME_LMV); + rc = mdt_is_dir_stripe(info, mdt_obj); mdt_object_put(info->mti_env, mdt_obj); - if (rc > 0) { - union lmv_mds_md *lmm = lmv_buf.lb_buf; - - /* For slave stripes, get its master */ - if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) { - fp->gf_fid = *tmpfid; - continue; - } - } else if (rc < 0 && rc != -ENODATA) { + if (rc < 0) GOTO(out, rc); + if (rc == 1) { + fp->gf_fid = *tmpfid; + continue; } /* Pack the name in the end of the buffer */ @@ -6939,7 +7037,7 @@ static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen, fpin = key + cfs_size_round(sizeof(KEY_FID2PATH)); fpout = val; - if (ptlrpc_req_need_swab(info->mti_pill->rc_req)) + if (req_capsule_req_need_swab(info->mti_pill)) lustre_swab_fid2path(fpin); memcpy(fpout, fpin, sizeof(*fpin)); @@ -6950,7 +7048,7 @@ static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen, sizeof(struct lu_fid)) { /* client sent its root FID, which is normally fileset FID */ root_fid = fpin->gf_u.gf_root_fid; - if (ptlrpc_req_need_swab(info->mti_pill->rc_req)) + if (req_capsule_req_need_swab(info->mti_pill)) lustre_swab_lu_fid(root_fid); if (root_fid != NULL && !fid_is_sane(root_fid)) @@ -7082,12 +7180,21 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, if (rc == 0) rc = dt_ro(&env, dt); break; - case OBD_IOC_ABORT_RECOVERY: + case OBD_IOC_ABORT_RECOVERY: { + struct obd_ioctl_data *data = karg; + CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt)); - obd->obd_abort_recovery = 1; - target_stop_recovery_thread(obd); + if (data->ioc_type & OBD_FLG_ABORT_RECOV_MDT) { + obd->obd_abort_recov_mdt = 1; + wake_up(&obd->obd_next_transno_waitq); + } else { /* if (data->ioc_type & OBD_FLG_ABORT_RECOV_OST) */ + /* lctl didn't set OBD_FLG_ABORT_RECOV_OST < 2.13.57 */ + obd->obd_abort_recovery = 1; + target_stop_recovery_thread(obd); + } rc = 0; break; + } case OBD_IOC_CHANGELOG_REG: case OBD_IOC_CHANGELOG_DEREG: case OBD_IOC_CHANGELOG_CLEAR: @@ -7332,10 +7439,10 @@ int mdt_cos_is_enabled(struct mdt_device *mdt) return mdt->mdt_opts.mo_cos != 0; } -static struct lu_device_type_operations mdt_device_type_ops = { - .ldto_device_alloc = mdt_device_alloc, - .ldto_device_free = mdt_device_free, - .ldto_device_fini = mdt_device_fini +static const struct lu_device_type_operations mdt_device_type_ops = { + .ldto_device_alloc = mdt_device_alloc, + .ldto_device_free = mdt_device_free, + .ldto_device_fini = mdt_device_fini }; static struct lu_device_type mdt_device_type = { @@ -7361,7 +7468,7 @@ static int __init mdt_init(void) if (rc) GOTO(lu_fini, rc); - rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL, + rc = class_register_type(&mdt_obd_device_ops, NULL, true, LUSTRE_MDT_NAME, &mdt_device_type); if (rc) GOTO(mds_fini, rc);