X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_handler.c;h=1b0998223b30b119344a27b795dced5eed3e5fcf;hp=2f98d85487d1b2a6c0e08a62a760c1fc27b3cba4;hb=df1d59429cbfd1ea2464e863458b6a4a268e516b;hpb=a7241f1692be2f725a4b2607801e6829c842891c diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c index 2f98d85..1b09982 100644 --- a/lustre/mdt/mdt_handler.c +++ b/lustre/mdt/mdt_handler.c @@ -62,12 +62,13 @@ /* lu2dt_dev() */ #include #include -#include #include #include "mdt_internal.h" #include #include #include +#include +#include mdl_mode_t mdt_mdl_lock_modes[] = { [LCK_MINMODE] = MDL_MINMODE, @@ -91,16 +92,19 @@ ldlm_mode_t mdt_dlm_lock_modes[] = { [MDL_GROUP] = LCK_GROUP }; - static struct mdt_device *mdt_dev(struct lu_device *d); static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags); -static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt, - struct getinfo_fid2path *fp); static const struct lu_object_operations mdt_obj_ops; /* Slab for MDT object allocation */ -static cfs_mem_cache_t *mdt_object_kmem; +static struct kmem_cache *mdt_object_kmem; + +/* For HSM restore handles */ +struct kmem_cache *mdt_hsm_cdt_kmem; + +/* For HSM request handles */ +struct kmem_cache *mdt_hsm_car_kmem; static struct lu_kmem_descr mdt_caches[] = { { @@ -109,6 +113,16 @@ static struct lu_kmem_descr mdt_caches[] = { .ckd_size = sizeof(struct mdt_object) }, { + .ckd_cache = &mdt_hsm_cdt_kmem, + .ckd_name = "mdt_cdt_restore_handle", + .ckd_size = sizeof(struct cdt_restore_handle) + }, + { + .ckd_cache = &mdt_hsm_car_kmem, + .ckd_name = "mdt_cdt_agent_req", + .ckd_size = sizeof(struct cdt_agent_req) + }, + { .ckd_cache = NULL } }; @@ -121,43 +135,57 @@ int mdt_get_disposition(struct ldlm_reply *rep, int flag) } void mdt_clear_disposition(struct mdt_thread_info *info, - struct ldlm_reply *rep, int flag) + struct ldlm_reply *rep, int flag) { - if (info) - info->mti_opdata &= ~flag; - if (rep) - rep->lock_policy_res1 &= ~flag; + if (info) { + info->mti_opdata &= ~flag; + tgt_opdata_clear(info->mti_env, flag); + } + if (rep) + rep->lock_policy_res1 &= ~flag; } void mdt_set_disposition(struct mdt_thread_info *info, - struct ldlm_reply *rep, int flag) + struct ldlm_reply *rep, int flag) { - if (info) - info->mti_opdata |= flag; - if (rep) - rep->lock_policy_res1 |= flag; + if (info) { + info->mti_opdata |= flag; + tgt_opdata_set(info->mti_env, flag); + } + if (rep) + rep->lock_policy_res1 |= flag; } void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm) { lh->mlh_pdo_hash = 0; lh->mlh_reg_mode = lm; + lh->mlh_rreg_mode = lm; lh->mlh_type = MDT_REG_LOCK; } -void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm, - const char *name, int namelen) -{ - lh->mlh_reg_mode = lm; - lh->mlh_type = MDT_PDO_LOCK; - - if (name != NULL && (name[0] != '\0')) { - LASSERT(namelen > 0); - lh->mlh_pdo_hash = full_name_hash(name, namelen); - } else { - LASSERT(namelen == 0); - lh->mlh_pdo_hash = 0ull; - } +void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode, + const struct lu_name *lname) +{ + lh->mlh_reg_mode = lock_mode; + lh->mlh_rreg_mode = lock_mode; + lh->mlh_type = MDT_PDO_LOCK; + + if (lu_name_is_valid(lname)) { + lh->mlh_pdo_hash = full_name_hash(lname->ln_name, + lname->ln_namelen); + /* XXX Workaround for LU-2856 + * + * Zero is a valid return value of full_name_hash, but + * several users of mlh_pdo_hash assume a non-zero + * hash value. We therefore map zero onto an + * arbitrary, but consistent value (1) to avoid + * problems further down the road. */ + if (unlikely(lh->mlh_pdo_hash == 0)) + lh->mlh_pdo_hash = 1; + } else { + lh->mlh_pdo_hash = 0; + } } static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o, @@ -235,37 +263,34 @@ static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o EXIT; } -int mdt_getstatus(struct mdt_thread_info *info) +int mdt_getstatus(struct tgt_session_info *tsi) { - struct mdt_device *mdt = info->mti_mdt; - struct md_device *next = mdt->mdt_child; - struct mdt_body *repbody; - int rc; - - ENTRY; + struct mdt_thread_info *info = tsi2mdt_info(tsi); + struct mdt_device *mdt = info->mti_mdt; + struct mdt_body *repbody; + int rc; - rc = mdt_check_ucred(info); - if (rc) - RETURN(err_serious(rc)); + ENTRY; - if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK)) - RETURN(err_serious(-ENOMEM)); + rc = mdt_check_ucred(info); + if (rc) + GOTO(out, rc = err_serious(rc)); - repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); - rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1); - if (rc != 0) - RETURN(rc); + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK)) + GOTO(out, rc = err_serious(-ENOMEM)); - repbody->valid |= OBD_MD_FLID; + repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); + repbody->fid1 = mdt->mdt_md_root_fid; + repbody->valid |= OBD_MD_FLID; - if (mdt->mdt_opts.mo_mds_capa && - info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) { - struct mdt_object *root; - struct lustre_capa *capa; + if (tsi->tsi_tgt->lut_mds_capa && + exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) { + struct mdt_object *root; + struct lustre_capa *capa; - root = mdt_object_find(info->mti_env, mdt, &repbody->fid1); - if (IS_ERR(root)) - RETURN(PTR_ERR(root)); + root = mdt_object_find(info->mti_env, mdt, &repbody->fid1); + if (IS_ERR(root)) + GOTO(out, rc = PTR_ERR(root)); capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1); LASSERT(capa); @@ -276,13 +301,16 @@ int mdt_getstatus(struct mdt_thread_info *info) if (rc == 0) repbody->valid |= OBD_MD_FLMDSCAPA; } - - RETURN(rc); + EXIT; +out: + mdt_thread_info_fini(info); + return rc; } -int mdt_statfs(struct mdt_thread_info *info) +int mdt_statfs(struct tgt_session_info *tsi) { - struct ptlrpc_request *req = mdt_info_req(info); + struct ptlrpc_request *req = tgt_ses_req(tsi); + struct mdt_thread_info *info = tsi2mdt_info(tsi); struct md_device *next = info->mti_mdt->mdt_child; struct ptlrpc_service_part *svcpt; struct obd_statfs *osfs; @@ -290,23 +318,23 @@ int mdt_statfs(struct mdt_thread_info *info) ENTRY; - svcpt = info->mti_pill->rc_req->rq_rqbd->rqbd_svcpt; + svcpt = req->rq_rqbd->rqbd_svcpt; /* This will trigger a watchdog timeout */ OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP, (MDT_SERVICE_WATCHDOG_FACTOR * at_get(&svcpt->scp_at_estimate)) + 1); - rc = mdt_check_ucred(info); - if (rc) - RETURN(err_serious(rc)); + rc = mdt_check_ucred(info); + if (rc) + GOTO(out, rc = err_serious(rc)); if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) - RETURN(err_serious(-ENOMEM)); + GOTO(out, rc = err_serious(-ENOMEM)); osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS); if (!osfs) - RETURN(-EPROTO); + GOTO(out, rc = -EPROTO); /** statfs information are cached in the mdt_device */ if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age, @@ -314,7 +342,7 @@ int mdt_statfs(struct mdt_thread_info *info) /** statfs data is too old, get up-to-date one */ rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs); if (rc) - RETURN(rc); + GOTO(out, rc); spin_lock(&info->mti_mdt->mdt_osfs_lock); info->mti_mdt->mdt_osfs = *osfs; info->mti_mdt->mdt_osfs_age = cfs_time_current_64(); @@ -328,7 +356,8 @@ int mdt_statfs(struct mdt_thread_info *info) if (rc == 0) mdt_counter_incr(req, LPROC_MDT_STATFS); - +out: + mdt_thread_info_fini(info); RETURN(rc); } @@ -393,17 +422,22 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, b->blocks = 0; /* if no object is allocated on osts, the size on mds is valid. b=22272 */ b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; - } + } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL && + ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) { + /* A released file stores its size on MDS. */ + /* But return 1 block for released file, unless tools like tar + * will consider it fully sparse. (LU-3864) + */ + if (unlikely(b->size == 0)) + b->blocks = 0; + else + b->blocks = 1; + b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; + } if (fid) { b->fid1 = *fid; b->valid |= OBD_MD_FLID; - - /* FIXME: these should be fixed when new igif ready.*/ - b->ino = fid_oid(fid); /* 1.6 compatibility */ - b->generation = fid_ver(fid); /* 1.6 compatibility */ - b->valid |= OBD_MD_FLGENER; /* 1.6 compatibility */ - CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n", PFID(fid), b->nlink, b->mode, b->size); } @@ -411,7 +445,7 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, if (info) mdt_body_reverse_idmap(info, b); - if (b->valid & OBD_MD_FLSIZE) + if (fid != NULL && (b->valid & OBD_MD_FLSIZE)) CDEBUG(D_VFSTRACE, DFID": returning size %llu\n", PFID(fid), (unsigned long long)b->size); } @@ -432,9 +466,9 @@ void mdt_client_compatibility(struct mdt_thread_info *info) struct lu_attr *la = &ma->ma_attr; ENTRY; - if (exp->exp_connect_flags & OBD_CONNECT_LAYOUTLOCK) - /* the client can deal with 16-bit lmm_stripe_count */ - RETURN_EXIT; + if (exp_connect_layout(exp)) + /* the client can deal with 16-bit lmm_stripe_count */ + RETURN_EXIT; body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); @@ -689,32 +723,46 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, ma->ma_valid = 0; - rc = mdt_object_exists(o); - if (rc < 0) { - /* This object is located on remote node.*/ - repbody->fid1 = *mdt_object_fid(o); - repbody->valid = OBD_MD_FLID | OBD_MD_MDS; - GOTO(out, rc = 0); - } + if (mdt_object_remote(o)) { + /* This object is located on remote node.*/ + /* Return -EIO for old client */ + if (!mdt_is_dne_client(req->rq_export)) + GOTO(out, rc = -EIO); + + repbody->fid1 = *mdt_object_fid(o); + repbody->valid = OBD_MD_FLID | OBD_MD_MDS; + GOTO(out, rc = 0); + } buffer->lb_len = reqbody->eadatasize; - if (buffer->lb_len > 0) + if (buffer->lb_len > 0) { buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD); - else + if (buffer->lb_buf == NULL) + GOTO(out, rc = -EPROTO); + } else { buffer->lb_buf = NULL; + ma_need &= ~(MA_LOV | MA_LMV); + CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n", + mdt_obd_name(info->mti_mdt), + req->rq_export->exp_client_uuid.uuid); + } - /* If it is dir object and client require MEA, then we got MEA */ - if (S_ISDIR(lu_object_attr(&next->mo_lu)) && - reqbody->valid & OBD_MD_MEA) { - /* Assumption: MDT_MD size is enough for lmv size. */ - ma->ma_lmv = buffer->lb_buf; - ma->ma_lmv_size = buffer->lb_len; - ma->ma_need = MA_LMV | MA_INODE; - } else { - ma->ma_lmm = buffer->lb_buf; - ma->ma_lmm_size = buffer->lb_len; - ma->ma_need = MA_LOV | MA_INODE | MA_HSM; - } + /* If it is dir object and client require MEA, then we got MEA */ + if (S_ISDIR(lu_object_attr(&next->mo_lu)) && + reqbody->valid & OBD_MD_MEA) { + /* Assumption: MDT_MD size is enough for lmv size. */ + ma->ma_lmv = buffer->lb_buf; + ma->ma_lmv_size = buffer->lb_len; + ma->ma_need = MA_INODE; + if (ma->ma_lmm_size > 0) + ma->ma_need |= MA_LMV; + } else { + ma->ma_lmm = buffer->lb_buf; + ma->ma_lmm_size = buffer->lb_len; + ma->ma_need = MA_INODE | MA_HSM; + if (ma->ma_lmm_size > 0) + ma->ma_need |= MA_LOV; + } if (S_ISDIR(lu_object_attr(&next->mo_lu)) && reqbody->valid & OBD_MD_FLDIREA && @@ -727,18 +775,26 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, ma->ma_som = &info->mti_u.som.data; rc = mdt_attr_get_complex(info, o, ma); - if (unlikely(rc)) { - CERROR("getattr error for "DFID": %d\n", - PFID(mdt_object_fid(o)), rc); - RETURN(rc); - } + if (unlikely(rc)) { + CERROR("%s: getattr error for "DFID": rc = %d\n", + mdt_obd_name(info->mti_mdt), + PFID(mdt_object_fid(o)), rc); + RETURN(rc); + } + + /* if file is released, check if a restore is running */ + if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) && + mdt_hsm_restore_is_running(info, mdt_object_fid(o))) { + repbody->t_state = MS_RESTORE; + repbody->valid |= OBD_MD_TSTATE; + } is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid); /* the Lustre protocol supposes to return default striping * on the user-visible root if explicitly requested */ if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) && - (ma->ma_need & MA_LOV_DEF && is_root) && (ma->ma_need & MA_LOV)) { + (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) { struct lu_fid rootfid; struct mdt_object *root; struct mdt_device *mdt = info->mti_mdt; @@ -752,8 +808,9 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, rc = mdt_attr_get_lov(info, root, ma); mdt_object_put(info->mti_env, root); if (unlikely(rc)) { - CERROR("getattr error for "DFID": %d\n", - PFID(mdt_object_fid(o)), rc); + CERROR("%s: getattr error for "DFID": rc = %d\n", + mdt_obd_name(info->mti_mdt), + PFID(mdt_object_fid(o)), rc); RETURN(rc); } } @@ -778,18 +835,20 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, repbody->eadatasize = ma->ma_lmv_size; repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA); } - } else if (S_ISLNK(la->la_mode) && - reqbody->valid & OBD_MD_LINKNAME) { - buffer->lb_buf = ma->ma_lmm; - /* eadatasize from client includes NULL-terminator, so - * there is no need to read it */ - buffer->lb_len = reqbody->eadatasize - 1; - rc = mo_readlink(env, next, buffer); - if (unlikely(rc <= 0)) { - CERROR("readlink failed: %d\n", rc); - rc = -EFAULT; - } else { - int print_limit = min_t(int, CFS_PAGE_SIZE - 128, rc); + } else if (S_ISLNK(la->la_mode) && + reqbody->valid & OBD_MD_LINKNAME) { + buffer->lb_buf = ma->ma_lmm; + /* eadatasize from client includes NULL-terminator, so + * there is no need to read it */ + buffer->lb_len = reqbody->eadatasize - 1; + rc = mo_readlink(env, next, buffer); + if (unlikely(rc <= 0)) { + CERROR("%s: readlink failed for "DFID": rc = %d\n", + mdt_obd_name(info->mti_mdt), + PFID(mdt_object_fid(o)), rc); + rc = -EFAULT; + } else { + int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc); if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO)) rc -= 2; @@ -798,8 +857,11 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, * because client expects that */ repbody->eadatasize = rc + 1; if (repbody->eadatasize != reqbody->eadatasize) - CERROR("Read shorter symlink %d, expected %d\n", - rc, reqbody->eadatasize - 1); + CDEBUG(D_INODE, "%s: Read shorter symlink %d " + "on "DFID ", expected %d\n", + mdt_obd_name(info->mti_mdt), + rc, PFID(mdt_object_fid(o)), + reqbody->eadatasize - 1); /* NULL terminate */ ((char *)ma->ma_lmm)[rc] = 0; @@ -838,8 +900,8 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, } } #ifdef CONFIG_FS_POSIX_ACL - else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) && - (reqbody->valid & OBD_MD_FLACL)) { + else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) && + (reqbody->valid & OBD_MD_FLACL)) { buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL); buffer->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER); @@ -853,9 +915,12 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, rc = 0; } else if (rc == -EOPNOTSUPP) { rc = 0; - } else { - CERROR("got acl size: %d\n", rc); - } + } else { + CERROR("%s: unable to read "DFID + " ACL: rc = %d\n", + mdt_obd_name(info->mti_mdt), + PFID(mdt_object_fid(o)), rc); + } } else { repbody->aclsize = rc; repbody->valid |= OBD_MD_FLACL; @@ -865,9 +930,9 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, } #endif - if (reqbody->valid & OBD_MD_FLMDSCAPA && - info->mti_mdt->mdt_opts.mo_mds_capa && - info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) { + if (reqbody->valid & OBD_MD_FLMDSCAPA && + info->mti_mdt->mdt_lut.lut_mds_capa && + exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) { struct lustre_capa *capa; capa = req_capsule_server_get(pill, &RMF_CAPA1); @@ -898,9 +963,9 @@ static int mdt_renew_capa(struct mdt_thread_info *info) * return directly, client will find body->valid OBD_MD_FLOSSCAPA * flag not set. */ - if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa || - !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA)) - RETURN(0); + if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa || + !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA)) + RETURN(0); body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); LASSERT(body != NULL); @@ -918,8 +983,9 @@ static int mdt_renew_capa(struct mdt_thread_info *info) RETURN(rc); } -int mdt_getattr(struct mdt_thread_info *info) +int mdt_getattr(struct tgt_session_info *tsi) { + struct mdt_thread_info *info = tsi2mdt_info(tsi); struct mdt_object *obj = info->mti_object; struct req_capsule *pill = info->mti_pill; struct mdt_body *reqbody; @@ -940,19 +1006,19 @@ int mdt_getattr(struct mdt_thread_info *info) } LASSERT(obj != NULL); - LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu)); + LASSERT(lu_object_assert_exists(&obj->mot_obj)); - mode = lu_object_attr(&obj->mot_obj.mo_lu); + mode = lu_object_attr(&obj->mot_obj); - /* old clients may not report needed easize, use max value then */ - req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, - reqbody->eadatasize == 0 ? - info->mti_mdt->mdt_max_mdsize : - reqbody->eadatasize); + /* old clients may not report needed easize, use max value then */ + req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, + reqbody->eadatasize == 0 ? + info->mti_mdt->mdt_max_mdsize : + reqbody->eadatasize); - rc = req_capsule_server_pack(pill); - if (unlikely(rc != 0)) - RETURN(err_serious(rc)); + rc = req_capsule_server_pack(pill); + if (unlikely(rc != 0)) + GOTO(out, rc = err_serious(rc)); repbody = req_capsule_server_get(pill, &RMF_MDT_BODY); LASSERT(repbody != NULL); @@ -982,11 +1048,14 @@ out_shrink: rc2 = mdt_fix_reply(info); if (rc == 0) rc = rc2; - return rc; +out: + mdt_thread_info_fini(info); + return rc; } -int mdt_is_subdir(struct mdt_thread_info *info) +int mdt_is_subdir(struct tgt_session_info *tsi) { + struct mdt_thread_info *info = tsi2mdt_info(tsi); struct mdt_object *o = info->mti_object; struct req_capsule *pill = info->mti_pill; const struct mdt_body *body = info->mti_body; @@ -998,18 +1067,112 @@ int mdt_is_subdir(struct mdt_thread_info *info) repbody = req_capsule_server_get(pill, &RMF_MDT_BODY); - /* - * We save last checked parent fid to @repbody->fid1 for remote - * directory case. - */ - LASSERT(fid_is_sane(&body->fid2)); - LASSERT(mdt_object_exists(o) > 0); - rc = mdo_is_subdir(info->mti_env, mdt_object_child(o), - &body->fid2, &repbody->fid1); - if (rc == 0 || rc == -EREMOTE) - repbody->valid |= OBD_MD_FLID; + /* + * We save last checked parent fid to @repbody->fid1 for remote + * directory case. + */ + LASSERT(fid_is_sane(&body->fid2)); + LASSERT(mdt_object_exists(o) && !mdt_object_remote(o)); + rc = mdo_is_subdir(info->mti_env, mdt_object_child(o), + &body->fid2, &repbody->fid1); + if (rc == 0 || rc == -EREMOTE) + repbody->valid |= OBD_MD_FLID; + + mdt_thread_info_fini(info); + RETURN(rc); +} - RETURN(rc); +int mdt_swap_layouts(struct tgt_session_info *tsi) +{ + struct mdt_thread_info *info; + struct ptlrpc_request *req = tgt_ses_req(tsi); + struct obd_export *exp = req->rq_export; + struct mdt_object *o1, *o2, *o; + struct mdt_lock_handle *lh1, *lh2; + struct mdc_swap_layouts *msl; + int rc; + ENTRY; + + /* client does not support layout lock, so layout swaping + * is disabled. + * FIXME: there is a problem for old clients which don't support + * layout lock yet. If those clients have already opened the file + * they won't be notified at all so that old layout may still be + * used to do IO. This can be fixed after file release is landed by + * doing exclusive open and taking full EX ibits lock. - Jinshan */ + if (!exp_connect_layout(exp)) + RETURN(-EOPNOTSUPP); + + info = tsi2mdt_info(tsi); + if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT)) + mdt_set_capainfo(info, 0, &info->mti_body->fid1, + req_capsule_client_get(info->mti_pill, + &RMF_CAPA1)); + + if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT)) + mdt_set_capainfo(info, 1, &info->mti_body->fid2, + req_capsule_client_get(info->mti_pill, + &RMF_CAPA2)); + + o1 = info->mti_object; + o = o2 = mdt_object_find(info->mti_env, info->mti_mdt, + &info->mti_body->fid2); + if (IS_ERR(o)) + GOTO(out, rc = PTR_ERR(o)); + + if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */ + GOTO(put, rc = -ENOENT); + + rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2); + if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */ + GOTO(put, rc); + + if (rc < 0) + swap(o1, o2); + + /* permission check. Make sure the calling process having permission + * to write both files. */ + rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL, + MAY_WRITE); + if (rc < 0) + GOTO(put, rc); + + rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL, + MAY_WRITE); + if (rc < 0) + GOTO(put, rc); + + msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS); + if (msl == NULL) + GOTO(put, rc = -EPROTO); + + lh1 = &info->mti_lh[MDT_LH_NEW]; + mdt_lock_reg_init(lh1, LCK_EX); + lh2 = &info->mti_lh[MDT_LH_OLD]; + mdt_lock_reg_init(lh2, LCK_EX); + + rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT | + MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK); + if (rc < 0) + GOTO(put, rc); + + rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT | + MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK); + if (rc < 0) + GOTO(unlock1, rc); + + rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1), + mdt_object_child(o2), msl->msl_flags); + GOTO(unlock2, rc); +unlock2: + mdt_object_unlock(info, o2, lh2, rc); +unlock1: + mdt_object_unlock(info, o1, lh1, rc); +put: + mdt_object_put(info->mti_env, o); +out: + mdt_thread_info_fini(info); + RETURN(rc); } static int mdt_raw_lookup(struct mdt_thread_info *info, @@ -1070,8 +1233,6 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, struct md_object *next = mdt_object_child(parent); struct lu_fid *child_fid = &info->mti_tmp_fid1; struct lu_name *lname = NULL; - const char *name = NULL; - int namelen = 0; struct mdt_lock_handle *lhp = NULL; struct ldlm_lock *lock; struct ldlm_res_id *res_id; @@ -1086,53 +1247,51 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)); LASSERT(parent != NULL); - name = req_capsule_client_get(info->mti_pill, &RMF_NAME); - if (name == NULL) - RETURN(err_serious(-EFAULT)); - namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME, - RCL_CLIENT) - 1; - if (!info->mti_cross_ref) { - /* - * XXX: Check for "namelen == 0" is for getattr by fid - * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name, - * that is the name must contain at least one character and - * the terminating '\0' - */ - if (namelen == 0) { - reqbody = req_capsule_client_get(info->mti_pill, - &RMF_MDT_BODY); - if (unlikely(reqbody == NULL)) - RETURN(err_serious(-EFAULT)); - - if (unlikely(!fid_is_sane(&reqbody->fid2))) - RETURN(err_serious(-EINVAL)); - - name = NULL; - CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", " - "ldlm_rep = %p\n", - PFID(mdt_object_fid(parent)), - PFID(&reqbody->fid2), ldlm_rep); - } else { - lname = mdt_name(info->mti_env, (char *)name, namelen); - CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, " - "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)), - name, ldlm_rep); - } - } + lname = &info->mti_name; + mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON); + + if (!info->mti_cross_ref) { + /* + * XXX: Check for anonymous name is for getattr by fid + * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name, + * that is the name must contain at least one character and + * the terminating '\0'. + */ + if (!lu_name_is_valid(lname)) { + reqbody = req_capsule_client_get(info->mti_pill, + &RMF_MDT_BODY); + if (unlikely(reqbody == NULL)) + RETURN(err_serious(-EFAULT)); + + if (unlikely(!fid_is_sane(&reqbody->fid2))) + RETURN(err_serious(-EINVAL)); + + CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", " + "ldlm_rep = %p\n", + PFID(mdt_object_fid(parent)), + PFID(&reqbody->fid2), ldlm_rep); + } else { + CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", " + "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)), + PNAME(lname), ldlm_rep); + } + } + mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD); - rc = mdt_object_exists(parent); - if (unlikely(rc == 0)) { + if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) { LU_OBJECT_DEBUG(D_INODE, info->mti_env, - &parent->mot_obj.mo_lu, + &parent->mot_obj, "Parent doesn't exist!\n"); RETURN(-ESTALE); } else if (!info->mti_cross_ref) { - LASSERTF(rc > 0, "Parent "DFID" is on remote server\n", + LASSERTF(!mdt_object_remote(parent), + "Parent "DFID" is on remote server\n", PFID(mdt_object_fid(parent))); } - if (lname) { + + if (lu_name_is_valid(lname)) { rc = mdt_raw_lookup(info, parent, lname, ldlm_rep); if (rc != 0) { if (rc > 0) @@ -1158,21 +1317,30 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, LDLM_LOCK_PUT(lock); rc = 0; } else { - mdt_lock_handle_init(lhc); - mdt_lock_reg_init(lhc, LCK_PR); + mdt_lock_handle_init(lhc); + mdt_lock_reg_init(lhc, LCK_PR); - /* - * Object's name is on another MDS, no lookup lock is - * needed here but update is. - */ - child_bits &= ~MDS_INODELOCK_LOOKUP; - child_bits |= MDS_INODELOCK_UPDATE; + /* + * Object's name is on another MDS, no lookup or layout + * lock is needed here but update lock is. + */ + child_bits &= ~(MDS_INODELOCK_LOOKUP | + MDS_INODELOCK_LAYOUT); + child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE; - rc = mdt_object_lock(info, child, lhc, child_bits, - MDT_LOCAL_LOCK); - } + rc = mdt_object_lock(info, child, lhc, child_bits, + MDT_LOCAL_LOCK); + } if (rc == 0) { /* Finally, we can get attr for child. */ + if (!mdt_object_exists(child)) { + LU_OBJECT_DEBUG(D_INFO, info->mti_env, + &child->mot_obj, + "remote object doesn't exist.\n"); + mdt_object_unlock(info, child, lhc, 1); + RETURN(-ENOENT); + } + mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA); rc = mdt_getattr_internal(info, child, 0); @@ -1182,11 +1350,11 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, RETURN(rc); } - if (lname) { - /* step 1: lock parent only if parent is a directory */ - if (S_ISDIR(lu_object_attr(&parent->mot_obj.mo_lu))) { - lhp = &info->mti_lh[MDT_LH_PARENT]; - mdt_lock_pdo_init(lhp, LCK_PR, name, namelen); + if (lu_name_is_valid(lname)) { + /* step 1: lock parent only if parent is a directory */ + if (S_ISDIR(lu_object_attr(&parent->mot_obj))) { + lhp = &info->mti_lh[MDT_LH_PARENT]; + mdt_lock_pdo_init(lhp, LCK_PR, lname); rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE, MDT_LOCAL_LOCK); @@ -1211,41 +1379,50 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS); } - /* - *step 3: find the child object by fid & lock it. - * regardless if it is local or remote. - */ - child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid); - - if (unlikely(IS_ERR(child))) - GOTO(out_parent, rc = PTR_ERR(child)); - if (is_resent) { - /* Do not take lock for resent case. */ - lock = ldlm_handle2lock(&lhc->mlh_reg_lh); - LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n", - lhc->mlh_reg_lh.cookie); - - res_id = &lock->l_resource->lr_name; - if (!fid_res_name_eq(mdt_object_fid(child), - &lock->l_resource->lr_name)) { - LASSERTF(fid_res_name_eq(mdt_object_fid(parent), - &lock->l_resource->lr_name), - "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n", - (unsigned long)res_id->name[0], - (unsigned long)res_id->name[1], - (unsigned long)res_id->name[2], - PFID(mdt_object_fid(parent))); - CWARN("Although resent, but still not get child lock" - "parent:"DFID" child:"DFID"\n", - PFID(mdt_object_fid(parent)), - PFID(mdt_object_fid(child))); - lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT); - LDLM_LOCK_PUT(lock); - GOTO(relock, 0); - } - LDLM_LOCK_PUT(lock); - rc = 0; - } else { + /* + *step 3: find the child object by fid & lock it. + * regardless if it is local or remote. + * + *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e) + * set parent dir fid the same as child fid in getattr by fid case + * we should not lu_object_find() the object again, could lead + * to hung if there is a concurrent unlink destroyed the object. + */ + if (lu_fid_eq(mdt_object_fid(parent), child_fid)) { + mdt_object_get(info->mti_env, parent); + child = parent; + } else { + child = mdt_object_find(info->mti_env, info->mti_mdt, + child_fid); + } + + if (unlikely(IS_ERR(child))) + GOTO(out_parent, rc = PTR_ERR(child)); + if (is_resent) { + /* Do not take lock for resent case. */ + lock = ldlm_handle2lock(&lhc->mlh_reg_lh); + LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n", + lhc->mlh_reg_lh.cookie); + + res_id = &lock->l_resource->lr_name; + if (!fid_res_name_eq(mdt_object_fid(child), + &lock->l_resource->lr_name)) { + LASSERTF(fid_res_name_eq(mdt_object_fid(parent), + &lock->l_resource->lr_name), + "Lock res_id: "DLDLMRES", fid: "DFID"\n", + PLDLMRES(lock->l_resource), + PFID(mdt_object_fid(parent))); + CWARN("Although resent, but still not get child lock" + "parent:"DFID" child:"DFID"\n", + PFID(mdt_object_fid(parent)), + PFID(mdt_object_fid(child))); + lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT); + LDLM_LOCK_PUT(lock); + GOTO(relock, 0); + } + LDLM_LOCK_PUT(lock); + rc = 0; + } else { bool try_layout = false; relock: @@ -1253,14 +1430,15 @@ relock: mdt_lock_handle_init(lhc); mdt_lock_reg_init(lhc, LCK_PR); - if (mdt_object_exists(child) == 0) { - LU_OBJECT_DEBUG(D_INODE, info->mti_env, - &child->mot_obj.mo_lu, - "Object doesn't exist!\n"); - GOTO(out_child, rc = -ENOENT); - } + if (!mdt_object_exists(child)) { + LU_OBJECT_DEBUG(D_INODE, info->mti_env, + &child->mot_obj, + "Object doesn't exist!\n"); + GOTO(out_child, rc = -ENOENT); + } - if (!(child_bits & MDS_INODELOCK_UPDATE)) { + if (!(child_bits & MDS_INODELOCK_UPDATE) && + mdt_object_exists(child) && !mdt_object_remote(child)) { struct md_attr *ma = &info->mti_attr; ma->ma_valid = 0; @@ -1269,11 +1447,11 @@ relock: if (unlikely(rc != 0)) GOTO(out_child, rc); - /* If the file has not been changed for some time, we - * return not only a LOOKUP lock, but also an UPDATE - * lock and this might save us RPC on later STAT. For - * directories, it also let negative dentry starts - * working for this dir. */ + /* If the file has not been changed for some time, we + * return not only a LOOKUP lock, but also an UPDATE + * lock and this might save us RPC on later STAT. For + * directories, it also let negative dentry cache start + * working for this dir. */ if (ma->ma_valid & MA_INODE && ma->ma_attr.la_valid & LA_CTIME && info->mti_mdt->mdt_namespace->ns_ctime_age_limit + @@ -1286,8 +1464,8 @@ relock: LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT)); if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) && exp_connect_layout(info->mti_exp) && - S_ISREG(lu_object_attr(&child->mot_obj.mo_lu)) && - ldlm_rep != NULL) { + S_ISREG(lu_object_attr(&child->mot_obj)) && + !mdt_object_remote(child) && ldlm_rep != NULL) { /* try to grant layout lock for regular file. */ try_layout = true; } @@ -1326,18 +1504,17 @@ relock: rc = mdt_getattr_internal(info, child, ma_need); if (unlikely(rc != 0)) { mdt_object_unlock(info, child, lhc, 1); - } else if (lock) { - /* Debugging code. */ - res_id = &lock->l_resource->lr_name; - LDLM_DEBUG(lock, "Returning lock to client"); - LASSERTF(fid_res_name_eq(mdt_object_fid(child), - &lock->l_resource->lr_name), - "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n", - (unsigned long)res_id->name[0], - (unsigned long)res_id->name[1], - (unsigned long)res_id->name[2], - PFID(mdt_object_fid(child))); - mdt_pack_size2body(info, child); + } else if (lock) { + /* Debugging code. */ + res_id = &lock->l_resource->lr_name; + LDLM_DEBUG(lock, "Returning lock to client"); + LASSERTF(fid_res_name_eq(mdt_object_fid(child), + &lock->l_resource->lr_name), + "Lock res_id: "DLDLMRES", fid: "DFID"\n", + PLDLMRES(lock->l_resource), + PFID(mdt_object_fid(child))); + if (mdt_object_exists(child) && !mdt_object_remote(child)) + mdt_pack_size2body(info, child); } if (lock) LDLM_LOCK_PUT(lock); @@ -1352,8 +1529,9 @@ out_parent: } /* normal handler: should release the child lock */ -int mdt_getattr_name(struct mdt_thread_info *info) +int mdt_getattr_name(struct tgt_session_info *tsi) { + struct mdt_thread_info *info = tsi2mdt_info(tsi); struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD]; struct mdt_body *reqbody; struct mdt_body *repbody; @@ -1385,177 +1563,88 @@ out_shrink: rc2 = mdt_fix_reply(info); if (rc == 0) rc = rc2; - return rc; + mdt_thread_info_fini(info); + return rc; } static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, void *karg, void *uarg); -int mdt_set_info(struct mdt_thread_info *info) +int mdt_set_info(struct tgt_session_info *tsi) { - struct ptlrpc_request *req = mdt_info_req(info); - char *key; - void *val; - int keylen, vallen, rc = 0; - ENTRY; - - rc = req_capsule_server_pack(info->mti_pill); - if (rc) - RETURN(rc); + struct ptlrpc_request *req = tgt_ses_req(tsi); + char *key; + void *val; + int keylen, vallen, rc = 0; - key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY); - if (key == NULL) { - DEBUG_REQ(D_HA, req, "no set_info key"); - RETURN(-EFAULT); - } + ENTRY; - keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY, - RCL_CLIENT); + key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY); + if (key == NULL) { + DEBUG_REQ(D_HA, req, "no set_info key"); + RETURN(err_serious(-EFAULT)); + } - val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL); - if (val == NULL) { - DEBUG_REQ(D_HA, req, "no set_info val"); - RETURN(-EFAULT); - } + keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY, + RCL_CLIENT); - vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL, - RCL_CLIENT); + val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL); + if (val == NULL) { + DEBUG_REQ(D_HA, req, "no set_info val"); + RETURN(err_serious(-EFAULT)); + } - /* Swab any part of val you need to here */ - if (KEY_IS(KEY_READ_ONLY)) { - req->rq_status = 0; - lustre_msg_set_status(req->rq_repmsg, 0); + vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL, + RCL_CLIENT); + /* Swab any part of val you need to here */ + if (KEY_IS(KEY_READ_ONLY)) { spin_lock(&req->rq_export->exp_lock); if (*(__u32 *)val) - req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY; + *exp_connect_flags_ptr(req->rq_export) |= + OBD_CONNECT_RDONLY; else - req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY; + *exp_connect_flags_ptr(req->rq_export) &= + ~OBD_CONNECT_RDONLY; spin_unlock(&req->rq_export->exp_lock); + } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) { + struct changelog_setinfo *cs = val; - } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) { - struct changelog_setinfo *cs = - (struct changelog_setinfo *)val; - if (vallen != sizeof(*cs)) { - CERROR("Bad changelog_clear setinfo size %d\n", vallen); - RETURN(-EINVAL); - } - if (ptlrpc_req_need_swab(req)) { - __swab64s(&cs->cs_recno); - __swab32s(&cs->cs_id); - } - - rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp, - vallen, val, NULL); - lustre_msg_set_status(req->rq_repmsg, rc); - - } else { - RETURN(-EINVAL); - } - RETURN(0); -} - -/** - * Top-level handler for MDT connection requests. - */ -int mdt_connect(struct mdt_thread_info *info) -{ - int rc; - struct obd_connect_data *reply; - struct obd_export *exp; - struct ptlrpc_request *req = mdt_info_req(info); - - rc = target_handle_connect(req); - if (rc != 0) - return err_serious(rc); + if (vallen != sizeof(*cs)) { + CERROR("%s: bad changelog_clear setinfo size %d\n", + tgt_name(tsi->tsi_tgt), vallen); + RETURN(-EINVAL); + } + if (ptlrpc_req_need_swab(req)) { + __swab64s(&cs->cs_recno); + __swab32s(&cs->cs_id); + } - LASSERT(req->rq_export != NULL); - info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev); - rc = mdt_init_sec_level(info); - if (rc != 0) { - obd_disconnect(class_export_get(req->rq_export)); - return rc; + rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export, + vallen, val, NULL); + } else { + RETURN(-EINVAL); } - - /* To avoid exposing partially initialized connection flags, changes up - * to this point have been staged in reply->ocd_connect_flags. Now that - * connection handling has completed successfully, atomically update - * the connect flags in the shared export data structure. LU-1623 */ - reply = req_capsule_server_get(info->mti_pill, &RMF_CONNECT_DATA); - exp = req->rq_export; - spin_lock(&exp->exp_lock); - exp->exp_connect_flags = reply->ocd_connect_flags; - spin_unlock(&exp->exp_lock); - - rc = mdt_init_idmap(info); - if (rc != 0) - obd_disconnect(class_export_get(req->rq_export)); - - return rc; -} - -int mdt_disconnect(struct mdt_thread_info *info) -{ - int rc; - ENTRY; - - rc = target_handle_disconnect(mdt_info_req(info)); - if (rc) - rc = err_serious(rc); - RETURN(rc); + RETURN(rc); } -static int mdt_sendpage(struct mdt_thread_info *info, - struct lu_rdpg *rdpg, int nob) +int mdt_readpage(struct tgt_session_info *tsi) { - struct ptlrpc_request *req = mdt_info_req(info); - struct obd_export *exp = req->rq_export; - struct ptlrpc_bulk_desc *desc; - struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info; - int tmpcount; - int tmpsize; - int i; - int rc; - ENTRY; - - desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE, - MDS_BULK_PORTAL); - if (desc == NULL) - RETURN(-ENOMEM); - - if (!(exp->exp_connect_flags & OBD_CONNECT_BRW_SIZE)) - /* old client requires reply size in it's PAGE_SIZE, - * which is rdpg->rp_count */ - nob = rdpg->rp_count; - - for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0; - i++, tmpcount -= tmpsize) { - tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE); - ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize); - } - - LASSERT(desc->bd_nob == nob); - rc = target_bulk_io(exp, desc, lwi); - ptlrpc_free_bulk_pin(desc); - RETURN(rc); -} + struct mdt_thread_info *info = mdt_th_info(tsi->tsi_env); + struct mdt_object *object = mdt_obj(tsi->tsi_corpus); + struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg; + const struct mdt_body *reqbody = tsi->tsi_mdt_body; + struct mdt_body *repbody; + int rc; + int i; -int mdt_readpage(struct mdt_thread_info *info) -{ - struct mdt_object *object = info->mti_object; - struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg; - struct mdt_body *reqbody; - struct mdt_body *repbody; - int rc; - int i; - ENTRY; + ENTRY; - if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) - RETURN(err_serious(-ENOMEM)); + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) + RETURN(err_serious(-ENOMEM)); - reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY); - repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); - if (reqbody == NULL || repbody == NULL) + repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY); + if (repbody == NULL || reqbody == NULL) RETURN(err_serious(-EFAULT)); /* @@ -1571,53 +1660,53 @@ int mdt_readpage(struct mdt_thread_info *info) } rdpg->rp_attrs = reqbody->mode; - if (info->mti_exp->exp_connect_flags & OBD_CONNECT_64BITHASH) - rdpg->rp_attrs |= LUDA_64BITHASH; - rdpg->rp_count = min_t(unsigned int, reqbody->nlink, - PTLRPC_MAX_BRW_SIZE); - rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1) >> - CFS_PAGE_SHIFT; + if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH) + rdpg->rp_attrs |= LUDA_64BITHASH; + rdpg->rp_count = min_t(unsigned int, reqbody->nlink, + exp_max_brw_size(tsi->tsi_exp)); + rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >> + PAGE_CACHE_SHIFT; OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]); if (rdpg->rp_pages == NULL) RETURN(-ENOMEM); for (i = 0; i < rdpg->rp_npages; ++i) { - rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD); + rdpg->rp_pages[i] = alloc_page(GFP_IOFS); if (rdpg->rp_pages[i] == NULL) GOTO(free_rdpg, rc = -ENOMEM); } /* call lower layers to fill allocated pages with directory data */ - rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg); - if (rc < 0) - GOTO(free_rdpg, rc); + rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg); + if (rc < 0) + GOTO(free_rdpg, rc); - /* send pages to client */ - rc = mdt_sendpage(info, rdpg, rc); + /* send pages to client */ + rc = tgt_sendpage(tsi, rdpg, rc); - EXIT; + EXIT; free_rdpg: - for (i = 0; i < rdpg->rp_npages; i++) - if (rdpg->rp_pages[i] != NULL) - cfs_free_page(rdpg->rp_pages[i]); - OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]); + for (i = 0; i < rdpg->rp_npages; i++) + if (rdpg->rp_pages[i] != NULL) + __free_page(rdpg->rp_pages[i]); + OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]); - if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) - RETURN(0); + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) + RETURN(0); - return rc; + return rc; } static int mdt_reint_internal(struct mdt_thread_info *info, struct mdt_lock_handle *lhc, __u32 op) { - struct req_capsule *pill = info->mti_pill; - struct mdt_body *repbody; - int rc = 0, rc2; - ENTRY; + struct req_capsule *pill = info->mti_pill; + struct mdt_body *repbody; + int rc = 0, rc2; + ENTRY; rc = mdt_reint_unpack(info, op); if (rc != 0) { @@ -1678,57 +1767,64 @@ out_shrink: return rc; } -static long mdt_reint_opcode(struct mdt_thread_info *info, - const struct req_format **fmt) -{ - struct mdt_rec_reint *rec; - long opc; - - opc = err_serious(-EFAULT); - rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT); - if (rec != NULL) { - opc = rec->rr_opcode; - DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc); - if (opc < REINT_MAX && fmt[opc] != NULL) - req_capsule_extend(info->mti_pill, fmt[opc]); - else { - CERROR("Unsupported opc: %ld\n", opc); - opc = err_serious(opc); - } - } - return opc; +static long mdt_reint_opcode(struct ptlrpc_request *req, + const struct req_format **fmt) +{ + struct mdt_device *mdt; + struct mdt_rec_reint *rec; + long opc; + + rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT); + if (rec != NULL) { + opc = rec->rr_opcode; + DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc); + if (opc < REINT_MAX && fmt[opc] != NULL) + req_capsule_extend(&req->rq_pill, fmt[opc]); + else { + mdt = mdt_exp2dev(req->rq_export); + CERROR("%s: Unsupported opcode '%ld' from client '%s':" + " rc = %d\n", req->rq_export->exp_obd->obd_name, + opc, mdt->mdt_ldlm_client->cli_name, -EFAULT); + opc = err_serious(-EFAULT); + } + } else { + opc = err_serious(-EFAULT); + } + return opc; } -int mdt_reint(struct mdt_thread_info *info) +int mdt_reint(struct tgt_session_info *tsi) { - long opc; - int rc; - - static const struct req_format *reint_fmts[REINT_MAX] = { - [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR, - [REINT_CREATE] = &RQF_MDS_REINT_CREATE, - [REINT_LINK] = &RQF_MDS_REINT_LINK, - [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK, - [REINT_RENAME] = &RQF_MDS_REINT_RENAME, - [REINT_OPEN] = &RQF_MDS_REINT_OPEN, - [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR - }; - - ENTRY; + long opc; + int rc; + static const struct req_format *reint_fmts[REINT_MAX] = { + [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR, + [REINT_CREATE] = &RQF_MDS_REINT_CREATE, + [REINT_LINK] = &RQF_MDS_REINT_LINK, + [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK, + [REINT_RENAME] = &RQF_MDS_REINT_RENAME, + [REINT_OPEN] = &RQF_MDS_REINT_OPEN, + [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR, + [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK + }; - opc = mdt_reint_opcode(info, reint_fmts); - if (opc >= 0) { - /* - * No lock possible here from client to pass it to reint code - * path. - */ - rc = mdt_reint_internal(info, NULL, opc); - } else { - rc = opc; - } + ENTRY; - info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP; - RETURN(rc); + opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts); + if (opc >= 0) { + struct mdt_thread_info *info = tsi2mdt_info(tsi); + /* + * No lock possible here from client to pass it to reint code + * path. + */ + rc = mdt_reint_internal(info, NULL, opc); + mdt_thread_info_fini(info); + } else { + rc = opc; + } + + tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP; + RETURN(rc); } /* this should sync the whole device */ @@ -1760,93 +1856,60 @@ static int mdt_object_sync(struct mdt_thread_info *info) RETURN(rc); } -int mdt_sync(struct mdt_thread_info *info) -{ - struct ptlrpc_request *req = mdt_info_req(info); - struct req_capsule *pill = info->mti_pill; - struct mdt_body *body; - int rc; - ENTRY; - - /* The fid may be zero, so we req_capsule_set manually */ - req_capsule_set(pill, &RQF_MDS_SYNC); - - body = req_capsule_client_get(pill, &RMF_MDT_BODY); - if (body == NULL) - RETURN(err_serious(-EINVAL)); - - if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK)) - RETURN(err_serious(-ENOMEM)); - - if (fid_seq(&body->fid1) == 0) { - /* sync the whole device */ - rc = req_capsule_server_pack(pill); - if (rc == 0) - rc = mdt_device_sync(info->mti_env, info->mti_mdt); - else - rc = err_serious(rc); - } else { - /* sync an object */ - rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO); - if (rc == 0) { - rc = mdt_object_sync(info); - if (rc == 0) { - const struct lu_fid *fid; - struct lu_attr *la = &info->mti_attr.ma_attr; - - info->mti_attr.ma_need = MA_INODE; - info->mti_attr.ma_valid = 0; - rc = mdt_attr_get_complex(info, info->mti_object, - &info->mti_attr); - if (rc == 0) { - body = req_capsule_server_get(pill, - &RMF_MDT_BODY); - fid = mdt_object_fid(info->mti_object); - mdt_pack_attr2body(info, body, la, fid); - } - } - } else - rc = err_serious(rc); - } - if (rc == 0) - mdt_counter_incr(req, LPROC_MDT_SYNC); - - RETURN(rc); -} - -/* - * Quotacheck handler. - * in-kernel quotacheck isn't supported any more. - */ -int mdt_quotacheck(struct mdt_thread_info *info) +int mdt_sync(struct tgt_session_info *tsi) { - struct obd_quotactl *oqctl; + struct ptlrpc_request *req = tgt_ses_req(tsi); + struct req_capsule *pill = tsi->tsi_pill; + struct mdt_body *body; int rc; + ENTRY; - oqctl = req_capsule_client_get(info->mti_pill, &RMF_OBD_QUOTACTL); - if (oqctl == NULL) - RETURN(err_serious(-EPROTO)); + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK)) + RETURN(err_serious(-ENOMEM)); - rc = req_capsule_server_pack(info->mti_pill); - if (rc) - RETURN(err_serious(rc)); + if (fid_seq(&tsi->tsi_mdt_body->fid1) == 0) { + rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp)); + } else { + struct mdt_thread_info *info = tsi2mdt_info(tsi); + + /* sync an object */ + rc = mdt_object_sync(info); + if (rc == 0) { + const struct lu_fid *fid; + struct lu_attr *la = &info->mti_attr.ma_attr; + + info->mti_attr.ma_need = MA_INODE; + info->mti_attr.ma_valid = 0; + rc = mdt_attr_get_complex(info, info->mti_object, + &info->mti_attr); + if (rc == 0) { + body = req_capsule_server_get(pill, + &RMF_MDT_BODY); + fid = mdt_object_fid(info->mti_object); + mdt_pack_attr2body(info, body, la, fid); + } + } + mdt_thread_info_fini(info); + } + if (rc == 0) + mdt_counter_incr(req, LPROC_MDT_SYNC); - /* deprecated, not used any more */ - RETURN(-EOPNOTSUPP); + RETURN(rc); } /* * Handle quota control requests to consult current usage/limit, but also * to configure quota enforcement */ -int mdt_quotactl(struct mdt_thread_info *info) +int mdt_quotactl(struct tgt_session_info *tsi) { - struct obd_export *exp = info->mti_exp; - struct req_capsule *pill = info->mti_pill; + struct obd_export *exp = tsi->tsi_exp; + struct req_capsule *pill = tsi->tsi_pill; struct obd_quotactl *oqctl, *repoqc; int id, rc; - struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev; + struct mdt_device *mdt = mdt_exp2dev(exp); + struct lu_device *qmt = mdt->mdt_qmt_dev; ENTRY; oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL); @@ -1887,7 +1950,7 @@ int mdt_quotactl(struct mdt_thread_info *info) if (exp_connect_rmtclient(exp)) { struct lustre_idmap_table *idmap; - idmap = mdt_req2med(mdt_info_req(info))->med_idmap; + idmap = exp->exp_mdt_data.med_idmap; if (unlikely(oqctl->qc_cmd != Q_GETQUOTA && oqctl->qc_cmd != Q_GETINFO)) @@ -1922,13 +1985,13 @@ int mdt_quotactl(struct mdt_thread_info *info) case Q_SETQUOTA: case Q_GETQUOTA: /* forward quotactl request to QMT */ - rc = qmt_hdls.qmth_quotactl(info->mti_env, qmt, oqctl); + rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl); break; case Q_GETOINFO: case Q_GETOQUOTA: /* slave quotactl */ - rc = lquotactl_slv(info->mti_env, info->mti_mdt->mdt_bottom, + rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, oqctl); break; @@ -1944,119 +2007,6 @@ int mdt_quotactl(struct mdt_thread_info *info) RETURN(rc); } -/* - * OBD PING and other handlers. - */ -int mdt_obd_ping(struct mdt_thread_info *info) -{ - int rc; - ENTRY; - - req_capsule_set(info->mti_pill, &RQF_OBD_PING); - - rc = target_handle_ping(mdt_info_req(info)); - if (rc < 0) - rc = err_serious(rc); - RETURN(rc); -} - -/* - * OBD_IDX_READ handler - */ -int mdt_obd_idx_read(struct mdt_thread_info *info) -{ - struct mdt_device *mdt = info->mti_mdt; - struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg; - struct idx_info *req_ii, *rep_ii; - int rc, i; - ENTRY; - - memset(rdpg, 0, sizeof(*rdpg)); - req_capsule_set(info->mti_pill, &RQF_OBD_IDX_READ); - - /* extract idx_info buffer from request & reply */ - req_ii = req_capsule_client_get(info->mti_pill, &RMF_IDX_INFO); - if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC) - RETURN(err_serious(-EPROTO)); - - rc = req_capsule_server_pack(info->mti_pill); - if (rc) - RETURN(err_serious(rc)); - - rep_ii = req_capsule_server_get(info->mti_pill, &RMF_IDX_INFO); - if (rep_ii == NULL) - RETURN(err_serious(-EFAULT)); - rep_ii->ii_magic = IDX_INFO_MAGIC; - - /* extract hash to start with */ - rdpg->rp_hash = req_ii->ii_hash_start; - - /* extract requested attributes */ - rdpg->rp_attrs = req_ii->ii_attrs; - - /* check that fid packed in request is valid and supported */ - if (!fid_is_sane(&req_ii->ii_fid)) - RETURN(-EINVAL); - rep_ii->ii_fid = req_ii->ii_fid; - - /* copy flags */ - rep_ii->ii_flags = req_ii->ii_flags; - - /* compute number of pages to allocate, ii_count is the number of 4KB - * containers */ - if (req_ii->ii_count <= 0) - GOTO(out, rc = -EFAULT); - rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT, - PTLRPC_MAX_BRW_SIZE); - rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE -1) >> CFS_PAGE_SHIFT; - - /* allocate pages to store the containers */ - OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0])); - if (rdpg->rp_pages == NULL) - GOTO(out, rc = -ENOMEM); - for (i = 0; i < rdpg->rp_npages; i++) { - rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD); - if (rdpg->rp_pages[i] == NULL) - GOTO(out, rc = -ENOMEM); - } - - /* populate pages with key/record pairs */ - rc = dt_index_read(info->mti_env, mdt->mdt_bottom, rep_ii, rdpg); - if (rc < 0) - GOTO(out, rc); - - LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than " - "asked %d > %d\n", rc, rdpg->rp_count); - - /* send pages to client */ - rc = mdt_sendpage(info, rdpg, rc); - - GOTO(out, rc); -out: - if (rdpg->rp_pages) { - for (i = 0; i < rdpg->rp_npages; i++) - if (rdpg->rp_pages[i]) - cfs_free_page(rdpg->rp_pages[i]); - OBD_FREE(rdpg->rp_pages, - rdpg->rp_npages * sizeof(rdpg->rp_pages[0])); - } - return rc; -} - -int mdt_obd_log_cancel(struct mdt_thread_info *info) -{ - return err_serious(-EOPNOTSUPP); -} - -int mdt_obd_qc_callback(struct mdt_thread_info *info) -{ - return err_serious(-EOPNOTSUPP); -} - -/* - * LLOG handlers. - */ - /** clone llog ctxt from child (mdd) * This allows remote llog (replicator) access. * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the @@ -2101,150 +2051,45 @@ static int mdt_llog_ctxt_unclone(const struct lu_env *env, return 0; } -int mdt_llog_create(struct mdt_thread_info *info) -{ - int rc; - - req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE); - rc = llog_origin_handle_open(mdt_info_req(info)); - return (rc < 0 ? err_serious(rc) : rc); -} - -int mdt_llog_destroy(struct mdt_thread_info *info) -{ - int rc; - - req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY); - rc = llog_origin_handle_destroy(mdt_info_req(info)); - return (rc < 0 ? err_serious(rc) : rc); -} - -int mdt_llog_read_header(struct mdt_thread_info *info) -{ - int rc; - - req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER); - rc = llog_origin_handle_read_header(mdt_info_req(info)); - return (rc < 0 ? err_serious(rc) : rc); -} - -int mdt_llog_next_block(struct mdt_thread_info *info) -{ - int rc; - - req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK); - rc = llog_origin_handle_next_block(mdt_info_req(info)); - return (rc < 0 ? err_serious(rc) : rc); -} - -int mdt_llog_prev_block(struct mdt_thread_info *info) -{ - int rc; - - req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK); - rc = llog_origin_handle_prev_block(mdt_info_req(info)); - return (rc < 0 ? err_serious(rc) : rc); -} - - -/* - * DLM handlers. - */ - -static struct ldlm_callback_suite cbs = { - .lcs_completion = ldlm_server_completion_ast, - .lcs_blocking = ldlm_server_blocking_ast, - .lcs_glimpse = ldlm_server_glimpse_ast -}; - -int mdt_enqueue(struct mdt_thread_info *info) -{ - struct ptlrpc_request *req; - int rc; - - /* - * info->mti_dlm_req already contains swapped and (if necessary) - * converted dlm request. - */ - LASSERT(info->mti_dlm_req != NULL); - - req = mdt_info_req(info); - rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace, - req, info->mti_dlm_req, &cbs); - info->mti_fail_id = OBD_FAIL_LDLM_REPLY; - return rc ? err_serious(rc) : req->rq_status; -} - -int mdt_convert(struct mdt_thread_info *info) -{ - int rc; - struct ptlrpc_request *req; - - LASSERT(info->mti_dlm_req); - req = mdt_info_req(info); - rc = ldlm_handle_convert0(req, info->mti_dlm_req); - return rc ? err_serious(rc) : req->rq_status; -} - -int mdt_bl_callback(struct mdt_thread_info *info) -{ - CERROR("bl callbacks should not happen on MDS\n"); - LBUG(); - return err_serious(-EOPNOTSUPP); -} - -int mdt_cp_callback(struct mdt_thread_info *info) -{ - CERROR("cp callbacks should not happen on MDS\n"); - LBUG(); - return err_serious(-EOPNOTSUPP); -} - /* * sec context handlers */ -int mdt_sec_ctx_handle(struct mdt_thread_info *info) +int mdt_sec_ctx_handle(struct tgt_session_info *tsi) { - int rc; - - rc = mdt_handle_idmap(info); + int rc; - if (unlikely(rc)) { - struct ptlrpc_request *req = mdt_info_req(info); - __u32 opc; + rc = mdt_handle_idmap(tsi); + if (unlikely(rc)) { + struct ptlrpc_request *req = tgt_ses_req(tsi); + __u32 opc; - opc = lustre_msg_get_opc(req->rq_reqmsg); - if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT) - sptlrpc_svc_ctx_invalidate(req); - } + opc = lustre_msg_get_opc(req->rq_reqmsg); + if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT) + sptlrpc_svc_ctx_invalidate(req); + } - CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val); + CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val); - return rc; + return rc; } /* * quota request handlers */ -int mdt_quota_dqacq(struct mdt_thread_info *info) +int mdt_quota_dqacq(struct tgt_session_info *tsi) { - struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev; + struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp); + struct lu_device *qmt = mdt->mdt_qmt_dev; int rc; ENTRY; if (qmt == NULL) RETURN(err_serious(-EOPNOTSUPP)); - rc = qmt_hdls.qmth_dqacq(info->mti_env, qmt, mdt_info_req(info)); + rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi)); RETURN(rc); } -static struct mdt_object *mdt_obj(struct lu_object *o) -{ - LASSERT(lu_device_is_mdt(o->lo_dev)); - return container_of0(o, struct mdt_object, mot_obj.mo_lu); -} - struct mdt_object *mdt_object_new(const struct lu_env *env, struct mdt_device *d, const struct lu_fid *f) @@ -2255,7 +2100,7 @@ struct mdt_object *mdt_object_new(const struct lu_env *env, ENTRY; CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f)); - o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, &conf); + o = lu_object_find(env, &d->mdt_lu_dev, f, &conf); if (unlikely(IS_ERR(o))) m = (struct mdt_object *)o; else @@ -2264,20 +2109,21 @@ struct mdt_object *mdt_object_new(const struct lu_env *env, } struct mdt_object *mdt_object_find(const struct lu_env *env, - struct mdt_device *d, - const struct lu_fid *f) + struct mdt_device *d, + const struct lu_fid *f) { - struct lu_object *o; - struct mdt_object *m; - ENTRY; + struct lu_object *o; + struct mdt_object *m; + ENTRY; - CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f)); - o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL); - if (unlikely(IS_ERR(o))) - m = (struct mdt_object *)o; - else - m = mdt_obj(o); - RETURN(m); + CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f)); + o = lu_object_find(env, &d->mdt_lu_dev, f, NULL); + if (unlikely(IS_ERR(o))) + m = (struct mdt_object *)o; + else + m = mdt_obj(o); + + RETURN(m); } /** @@ -2382,6 +2228,57 @@ int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, RETURN(rc); } +int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, + void *data, int flag) +{ + struct lustre_handle lockh; + int rc; + + switch (flag) { + case LDLM_CB_BLOCKING: + ldlm_lock2handle(lock, &lockh); + rc = ldlm_cli_cancel(&lockh, LCF_ASYNC); + if (rc < 0) { + CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc); + RETURN(rc); + } + break; + case LDLM_CB_CANCELING: + LDLM_DEBUG(lock, "Revoke remote lock\n"); + break; + default: + LBUG(); + } + RETURN(0); +} + +int mdt_remote_object_lock(struct mdt_thread_info *mti, + struct mdt_object *o, struct lustre_handle *lh, + ldlm_mode_t mode, __u64 ibits) +{ + struct ldlm_enqueue_info *einfo = &mti->mti_einfo; + ldlm_policy_data_t *policy = &mti->mti_policy; + int rc = 0; + ENTRY; + + LASSERT(mdt_object_remote(o)); + + LASSERT(ibits & MDS_INODELOCK_UPDATE); + + memset(einfo, 0, sizeof(*einfo)); + einfo->ei_type = LDLM_IBITS; + einfo->ei_mode = mode; + einfo->ei_cb_bl = mdt_md_blocking_ast; + einfo->ei_cb_cp = ldlm_completion_ast; + + memset(policy, 0, sizeof(*policy)); + policy->l_inodebits.bits = ibits; + + rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo, + policy); + RETURN(rc); +} + static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o, struct mdt_lock_handle *lh, __u64 ibits, bool nonblock, int locality) @@ -2398,28 +2295,33 @@ static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o, LASSERT(lh->mlh_reg_mode != LCK_MINMODE); LASSERT(lh->mlh_type != MDT_NUL_LOCK); - if (mdt_object_exists(o) < 0) { + if (mdt_object_remote(o)) { if (locality == MDT_CROSS_LOCK) { - /* cross-ref object fix */ - ibits &= ~MDS_INODELOCK_UPDATE; + ibits &= ~(MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM | + MDS_INODELOCK_LAYOUT); ibits |= MDS_INODELOCK_LOOKUP; } else { - LASSERT(!(ibits & MDS_INODELOCK_UPDATE)); + LASSERTF(!(ibits & + (MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM | + MDS_INODELOCK_LAYOUT)), + "%s: wrong bit "LPX64" for remote obj "DFID"\n", + mdt_obd_name(info->mti_mdt), ibits, + PFID(mdt_object_fid(o))); LASSERT(ibits & MDS_INODELOCK_LOOKUP); } /* No PDO lock on remote object */ LASSERT(lh->mlh_type != MDT_PDO_LOCK); } - if (lh->mlh_type == MDT_PDO_LOCK) { + if (lh->mlh_type == MDT_PDO_LOCK) { /* check for exists after object is locked */ if (mdt_object_exists(o) == 0) { /* Non-existent object shouldn't have PDO lock */ RETURN(-ESTALE); } else { /* Non-dir object shouldn't have PDO lock */ - if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu))) - RETURN(-ENOTDIR); + if (!S_ISDIR(lu_object_attr(&o->mot_obj))) + RETURN(-ENOTDIR); } } @@ -2529,15 +2431,21 @@ void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h, LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n", h->cookie); - CDEBUG(D_HA, "request = %p reply state = %p" - " transno = "LPD64"\n", - req, req->rq_reply_state, req->rq_transno); - if (mdt_cos_is_enabled(mdt)) { - no_ack = 1; - ldlm_lock_downgrade(lock, LCK_COS); - mode = LCK_COS; - } - ptlrpc_save_lock(req, h, mode, no_ack); + /* there is no request if mdt_object_unlock() is called + * from mdt_export_cleanup()->mdt_add_dirty_flag() */ + if (likely(req != NULL)) { + CDEBUG(D_HA, "request = %p reply state = %p" + " transno = "LPD64"\n", req, + req->rq_reply_state, req->rq_transno); + if (mdt_cos_is_enabled(mdt)) { + no_ack = 1; + ldlm_lock_downgrade(lock, LCK_COS); + mode = LCK_COS; + } + ptlrpc_save_lock(req, h, mode, no_ack); + } else { + ldlm_lock_decref(h, mode); + } if (mdt_is_lock_sync(lock)) { CDEBUG(D_HA, "found sync-lock," " async commit started\n"); @@ -2572,6 +2480,9 @@ void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o, mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref); mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref); + if (lustre_handle_is_used(&lh->mlh_rreg_lh)) + ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode); + EXIT; } @@ -2605,40 +2516,6 @@ void mdt_object_unlock_put(struct mdt_thread_info * info, mdt_object_put(info->mti_env, o); } -struct mdt_handler *mdt_handler_find(__u32 opc, struct mdt_opc_slice *supported) -{ - struct mdt_opc_slice *s; - struct mdt_handler *h; - - h = NULL; - for (s = supported; s->mos_hs != NULL; s++) { - if (s->mos_opc_start <= opc && opc < s->mos_opc_end) { - h = s->mos_hs + (opc - s->mos_opc_start); - if (likely(h->mh_opc != 0)) - LASSERTF(h->mh_opc == opc, - "opcode mismatch %d != %d\n", - h->mh_opc, opc); - else - h = NULL; /* unsupported opc */ - break; - } - } - return h; -} - -static int mdt_lock_resname_compat(struct mdt_device *m, - struct ldlm_request *req) -{ - /* XXX something... later. */ - return 0; -} - -static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep) -{ - /* XXX something... later. */ - return 0; -} - /* * Generic code handling requests that have struct mdt_body passed in: * @@ -2733,150 +2610,13 @@ static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags) static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m) { - struct md_device *next = m->mdt_child; - - return next->md_ops->mdo_init_capa_ctxt(env, next, - m->mdt_opts.mo_mds_capa, - m->mdt_capa_timeout, - m->mdt_capa_alg, - m->mdt_capa_keys); -} - -/* - * Invoke handler for this request opc. Also do necessary preprocessing - * (according to handler ->mh_flags), and post-processing (setting of - * ->last_{xid,committed}). - */ -static int mdt_req_handle(struct mdt_thread_info *info, - struct mdt_handler *h, struct ptlrpc_request *req) -{ - int rc, serious = 0; - __u32 flags; - - ENTRY; - - LASSERT(h->mh_act != NULL); - LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg)); - LASSERT(current->journal_info == NULL); - - /* - * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try - * to put same checks into handlers like mdt_close(), mdt_reint(), - * etc., without talking to mdt authors first. Checking same thing - * there again is useless and returning 0 error without packing reply - * is buggy! Handlers either pack reply or return error. - * - * We return 0 here and do not send any reply in order to emulate - * network failure. Do not send any reply in case any of NET related - * fail_id has occured. - */ - if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE)) - RETURN(0); - - rc = 0; - flags = h->mh_flags; - LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL)); - - if (h->mh_fmt != NULL) { - req_capsule_set(info->mti_pill, h->mh_fmt); - rc = mdt_unpack_req_pack_rep(info, flags); - } - - if (rc == 0 && flags & MUTABOR && - req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY) - /* should it be rq_status? */ - rc = -EROFS; - - if (rc == 0 && flags & HABEO_CLAVIS) { - struct ldlm_request *dlm_req; - - LASSERT(h->mh_fmt != NULL); - - dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ); - if (dlm_req != NULL) { - if (unlikely(dlm_req->lock_desc.l_resource.lr_type == - LDLM_IBITS && - dlm_req->lock_desc.l_policy_data.\ - l_inodebits.bits == 0)) { - /* - * Lock without inodebits makes no sense and - * will oops later in ldlm. If client miss to - * set such bits, do not trigger ASSERTION. - * - * For liblustre flock case, it maybe zero. - */ - rc = -EPROTO; - } else { - if (info->mti_mdt->mdt_opts.mo_compat_resname) - rc = mdt_lock_resname_compat( - info->mti_mdt, - dlm_req); - info->mti_dlm_req = dlm_req; - } - } else { - rc = -EFAULT; - } - } - - /* capability setting changed via /proc, needs reinitialize ctxt */ - if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) { - mdt_init_capa_ctxt(info->mti_env, info->mti_mdt); - info->mti_mdt->mdt_capa_conf = 0; - } - - if (likely(rc == 0)) { - /* - * Process request, there can be two types of rc: - * 1) errors with msg unpack/pack, other failures outside the - * operation itself. This is counted as serious errors; - * 2) errors during fs operation, should be placed in rq_status - * only - */ - rc = h->mh_act(info); - if (rc == 0 && - !req->rq_no_reply && req->rq_reply_state == NULL) { - DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not " - "pack reply and returned 0 error\n", - h->mh_name); - LBUG(); - } - serious = is_serious(rc); - rc = clear_serious(rc); - } else - serious = 1; - - req->rq_status = rc; - - /* - * ELDLM_* codes which > 0 should be in rq_status only as well as - * all non-serious errors. - */ - if (rc > 0 || !serious) - rc = 0; - - LASSERT(current->journal_info == NULL); - - if (rc == 0 && (flags & HABEO_CLAVIS) && - info->mti_mdt->mdt_opts.mo_compat_resname) { - struct ldlm_reply *dlmrep; - - dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP); - if (dlmrep != NULL) - rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep); - } + struct md_device *next = m->mdt_child; - /* If we're DISCONNECTing, the mdt_export_data is already freed */ - if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT)) - target_committed_to_req(req); - - if (unlikely(req_is_replay(req) && - lustre_msg_get_transno(req->rq_reqmsg) == 0)) { - DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY"); - LBUG(); - } - - target_send_reply(req, rc, info->mti_fail_id); - RETURN(0); + return next->md_ops->mdo_init_capa_ctxt(env, next, + m->mdt_lut.lut_mds_capa, + m->mdt_capa_timeout, + m->mdt_capa_alg, + m->mdt_capa_keys); } void mdt_lock_handle_init(struct mdt_lock_handle *lh) @@ -2886,6 +2626,8 @@ void mdt_lock_handle_init(struct mdt_lock_handle *lh) lh->mlh_reg_mode = LCK_MINMODE; lh->mlh_pdo_lh.cookie = 0ull; lh->mlh_pdo_mode = LCK_MINMODE; + lh->mlh_rreg_lh.cookie = 0ull; + lh->mlh_rreg_mode = LCK_MINMODE; } void mdt_lock_handle_fini(struct mdt_lock_handle *lh) @@ -2899,12 +2641,11 @@ void mdt_lock_handle_fini(struct mdt_lock_handle *lh) * uninitialized state, because it's too expensive to zero out whole * mdt_thread_info (> 1K) on each request arrival. */ -static void mdt_thread_info_init(struct ptlrpc_request *req, - struct mdt_thread_info *info) +void mdt_thread_info_init(struct ptlrpc_request *req, + struct mdt_thread_info *info) { int i; - req_capsule_init(&req->rq_pill, req, RCL_SERVER); info->mti_pill = &req->rq_pill; /* lock handle */ @@ -2917,13 +2658,12 @@ static void mdt_thread_info_init(struct ptlrpc_request *req, info->mti_exp = req->rq_export; } else info->mti_mdt = NULL; - info->mti_env = req->rq_svc_thread->t_env; - info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET; - info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg); - info->mti_mos = NULL; + info->mti_env = req->rq_svc_thread->t_env; + info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg); memset(&info->mti_attr, 0, sizeof(info->mti_attr)); - info->mti_body = NULL; + info->mti_big_buf = LU_BUF_NULL; + info->mti_body = NULL; info->mti_object = NULL; info->mti_dlm_req = NULL; info->mti_has_trans = 0; @@ -2933,319 +2673,77 @@ static void mdt_thread_info_init(struct ptlrpc_request *req, /* To not check for split by default. */ info->mti_spec.no_create = 0; + info->mti_spec.sp_rm_entry = 0; } -static void mdt_thread_info_fini(struct mdt_thread_info *info) -{ - int i; - - req_capsule_fini(info->mti_pill); - if (info->mti_object != NULL) { - mdt_object_put(info->mti_env, info->mti_object); - info->mti_object = NULL; - } - for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++) - mdt_lock_handle_fini(&info->mti_lh[i]); - info->mti_env = NULL; -} - -static int mdt_filter_recovery_request(struct ptlrpc_request *req, - struct obd_device *obd, int *process) -{ - switch (lustre_msg_get_opc(req->rq_reqmsg)) { - case MDS_CONNECT: /* This will never get here, but for completeness. */ - case OST_CONNECT: /* This will never get here, but for completeness. */ - case MDS_DISCONNECT: - case OST_DISCONNECT: - case OBD_IDX_READ: - *process = 1; - RETURN(0); - - case MDS_CLOSE: - case MDS_DONE_WRITING: - case MDS_SYNC: /* used in unmounting */ - case OBD_PING: - case MDS_REINT: - case SEQ_QUERY: - case FLD_QUERY: - case LDLM_ENQUEUE: - *process = target_queue_recovery_request(req, obd); - RETURN(0); - - default: - DEBUG_REQ(D_ERROR, req, "not permitted during recovery"); - *process = -EAGAIN; - RETURN(0); - } -} - -/* - * Handle recovery. Return: - * +1: continue request processing; - * -ve: abort immediately with the given error code; - * 0: send reply with error code in req->rq_status; - */ -static int mdt_recovery(struct mdt_thread_info *info) +void mdt_thread_info_fini(struct mdt_thread_info *info) { - struct ptlrpc_request *req = mdt_info_req(info); - struct obd_device *obd; - - ENTRY; - - switch (lustre_msg_get_opc(req->rq_reqmsg)) { - case MDS_CONNECT: - case SEC_CTX_INIT: - case SEC_CTX_INIT_CONT: - case SEC_CTX_FINI: - { -#if 0 - int rc; - - rc = mdt_handle_idmap(info); - if (rc) - RETURN(rc); - else -#endif - RETURN(+1); - } - } - - if (unlikely(!class_connected_export(req->rq_export))) { - CDEBUG(D_HA, "operation %d on unconnected MDS from %s\n", - lustre_msg_get_opc(req->rq_reqmsg), - libcfs_id2str(req->rq_peer)); - /* FIXME: For CMD cleanup, when mds_B stop, the req from - * mds_A will get -ENOTCONN(especially for ping req), - * which will cause that mds_A deactive timeout, then when - * mds_A cleanup, the cleanup process will be suspended since - * deactive timeout is not zero. - */ - req->rq_status = -ENOTCONN; - target_send_reply(req, -ENOTCONN, info->mti_fail_id); - RETURN(0); - } + int i; - /* sanity check: if the xid matches, the request must be marked as a - * resent or replayed */ - if (req_xid_is_last(req)) { - if (!(lustre_msg_get_flags(req->rq_reqmsg) & - (MSG_RESENT | MSG_REPLAY))) { - DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, " - "expected REPLAY or RESENT flag (%x)", req->rq_xid, - lustre_msg_get_flags(req->rq_reqmsg)); - LBUG(); - req->rq_status = -ENOTCONN; - RETURN(-ENOTCONN); - } - } - - /* else: note the opposite is not always true; a RESENT req after a - * failover will usually not match the last_xid, since it was likely - * never committed. A REPLAYed request will almost never match the - * last xid, however it could for a committed, but still retained, - * open. */ - - obd = req->rq_export->exp_obd; - - /* Check for aborted recovery... */ - if (unlikely(obd->obd_recovering)) { - int rc; - int should_process; - DEBUG_REQ(D_INFO, req, "Got new replay"); - rc = mdt_filter_recovery_request(req, obd, &should_process); - if (rc != 0 || !should_process) - RETURN(rc); - else if (should_process < 0) { - req->rq_status = should_process; - rc = ptlrpc_error(req); - RETURN(rc); - } - } - RETURN(+1); -} - -static int mdt_msg_check_version(struct lustre_msg *msg) -{ - int rc; - - switch (lustre_msg_get_opc(msg)) { - case MDS_CONNECT: - case MDS_DISCONNECT: - case OBD_PING: - case SEC_CTX_INIT: - case SEC_CTX_INIT_CONT: - case SEC_CTX_FINI: - case OBD_IDX_READ: - rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION); - if (rc) - CERROR("bad opc %u version %08x, expecting %08x\n", - lustre_msg_get_opc(msg), - lustre_msg_get_version(msg), - LUSTRE_OBD_VERSION); - break; - case MDS_GETSTATUS: - case MDS_GETATTR: - case MDS_GETATTR_NAME: - case MDS_STATFS: - case MDS_READPAGE: - case MDS_WRITEPAGE: - case MDS_IS_SUBDIR: - case MDS_REINT: - case MDS_CLOSE: - case MDS_DONE_WRITING: - case MDS_PIN: - case MDS_SYNC: - case MDS_GETXATTR: - case MDS_SETXATTR: - case MDS_SET_INFO: - case MDS_GET_INFO: - case MDS_HSM_PROGRESS: - case MDS_HSM_REQUEST: - case MDS_HSM_CT_REGISTER: - case MDS_HSM_CT_UNREGISTER: - case MDS_HSM_STATE_GET: - case MDS_HSM_STATE_SET: - case MDS_HSM_ACTION: - case MDS_QUOTACHECK: - case MDS_QUOTACTL: - case UPDATE_OBJ: - case QUOTA_DQACQ: - case QUOTA_DQREL: - case SEQ_QUERY: - case FLD_QUERY: - rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION); - if (rc) - CERROR("bad opc %u version %08x, expecting %08x\n", - lustre_msg_get_opc(msg), - lustre_msg_get_version(msg), - LUSTRE_MDS_VERSION); - break; - case LDLM_ENQUEUE: - case LDLM_CONVERT: - case LDLM_BL_CALLBACK: - case LDLM_CP_CALLBACK: - rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION); - if (rc) - CERROR("bad opc %u version %08x, expecting %08x\n", - lustre_msg_get_opc(msg), - lustre_msg_get_version(msg), - LUSTRE_DLM_VERSION); - break; - case OBD_LOG_CANCEL: - case LLOG_ORIGIN_HANDLE_CREATE: - case LLOG_ORIGIN_HANDLE_NEXT_BLOCK: - case LLOG_ORIGIN_HANDLE_READ_HEADER: - case LLOG_ORIGIN_HANDLE_CLOSE: - case LLOG_ORIGIN_HANDLE_DESTROY: - case LLOG_ORIGIN_HANDLE_PREV_BLOCK: - case LLOG_CATINFO: - rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION); - if (rc) - CERROR("bad opc %u version %08x, expecting %08x\n", - lustre_msg_get_opc(msg), - lustre_msg_get_version(msg), - LUSTRE_LOG_VERSION); - break; - default: - CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg)); - rc = -ENOTSUPP; - } - return rc; -} - -static int mdt_handle0(struct ptlrpc_request *req, - struct mdt_thread_info *info, - struct mdt_opc_slice *supported) -{ - struct mdt_handler *h; - struct lustre_msg *msg; - int rc; - - ENTRY; - - if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE)) - RETURN(0); - - LASSERT(current->journal_info == NULL); - - msg = req->rq_reqmsg; - rc = mdt_msg_check_version(msg); - if (likely(rc == 0)) { - rc = mdt_recovery(info); - if (likely(rc == +1)) { - h = mdt_handler_find(lustre_msg_get_opc(msg), - supported); - if (likely(h != NULL)) { - rc = mdt_req_handle(info, h, req); - } else { - CERROR("The unsupported opc: 0x%x\n", - lustre_msg_get_opc(msg) ); - req->rq_status = -ENOTSUPP; - rc = ptlrpc_error(req); - RETURN(rc); - } - } - } else - CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n"); - RETURN(rc); + if (info->mti_object != NULL) { + mdt_object_put(info->mti_env, info->mti_object); + info->mti_object = NULL; + } + + for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++) + mdt_lock_handle_fini(&info->mti_lh[i]); + info->mti_env = NULL; + info->mti_pill = NULL; + info->mti_exp = NULL; + + if (unlikely(info->mti_big_buf.lb_buf != NULL)) + lu_buf_free(&info->mti_big_buf); } -/* - * MDT handler function called by ptlrpc service thread when request comes. - * - * XXX common "target" functionality should be factored into separate module - * shared by mdt, ost and stand-alone services like fld. - */ -int mdt_handle_common(struct ptlrpc_request *req, - struct mdt_opc_slice *supported) +struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi) { - struct lu_env *env; - struct mdt_thread_info *info; - int rc; - ENTRY; + struct mdt_thread_info *mti; + struct lustre_capa *lc; - env = req->rq_svc_thread->t_env; - /* Refill(initilize) the context(mdt_thread_info), in case it is - * not initialized yet. Usually it happens during start up, after - * MDS(ptlrpc threads) is start up, it gets the first CONNECT request, - * before MDT_thread_info is initialized */ - lu_env_refill(env); - LASSERT(env != NULL); - LASSERT(env->le_ses != NULL); - LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread); - info = lu_context_key_get(&env->le_ctx, &mdt_thread_key); - LASSERT(info != NULL); + mti = mdt_th_info(tsi->tsi_env); + LASSERT(mti != NULL); - mdt_thread_info_init(req, info); + mdt_thread_info_init(tgt_ses_req(tsi), mti); + if (tsi->tsi_corpus != NULL) { + struct req_capsule *pill = tsi->tsi_pill; - rc = mdt_handle0(req, info, supported); + mti->mti_object = mdt_obj(tsi->tsi_corpus); + lu_object_get(tsi->tsi_corpus); - mdt_thread_info_fini(info); - RETURN(rc); + /* + * XXX: must be part of tgt_mdt_body_unpack but moved here + * due to mdt_set_capainfo(). + */ + if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) && + req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) { + lc = req_capsule_client_get(pill, &RMF_CAPA1); + mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->fid1, lc); + } + } + mti->mti_body = tsi->tsi_mdt_body; + mti->mti_dlm_req = tsi->tsi_dlm_req; + + return mti; } -/* - * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ - * as well. - */ -int mdt_recovery_handle(struct ptlrpc_request *req) +int mdt_tgt_connect(struct tgt_session_info *tsi) { - int rc; - ENTRY; + struct ptlrpc_request *req = tgt_ses_req(tsi); + int rc; - switch (lustre_msg_get_opc(req->rq_reqmsg)) { - case FLD_QUERY: - rc = mdt_handle_common(req, mdt_fld_handlers); - break; - case SEQ_QUERY: - rc = mdt_handle_common(req, mdt_seq_handlers); - break; - default: - rc = mdt_handle_common(req, mdt_regular_handlers); - break; - } + ENTRY; - RETURN(rc); + rc = tgt_connect(tsi); + if (rc != 0) + RETURN(rc); + + rc = mdt_init_idmap(tsi); + if (rc != 0) + GOTO(err, rc); + RETURN(0); +err: + obd_disconnect(class_export_get(req->rq_export)); + return rc; } enum mdt_it_code { @@ -3267,6 +2765,12 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, struct mdt_thread_info *info, struct ldlm_lock **, __u64); + +static int mdt_intent_getxattr(enum mdt_it_code opcode, + struct mdt_thread_info *info, + struct ldlm_lock **lockp, + __u64 flags); + static int mdt_intent_layout(enum mdt_it_code opcode, struct mdt_thread_info *info, struct ldlm_lock **, @@ -3294,7 +2798,14 @@ static struct mdt_it_flavor { }, [MDT_IT_OCREAT] = { .it_fmt = &RQF_LDLM_INTENT, - .it_flags = MUTABOR, + /* + * OCREAT is not a MUTABOR request as if the file + * already exists. + * We do the extra check of OBD_CONNECT_RDONLY in + * mdt_reint_open() when we really need to create + * the object. + */ + .it_flags = 0, .it_act = mdt_intent_reint, .it_reint = REINT_OPEN }, @@ -3331,9 +2842,9 @@ static struct mdt_it_flavor { .it_act = NULL }, [MDT_IT_GETXATTR] = { - .it_fmt = NULL, - .it_flags = 0, - .it_act = NULL + .it_fmt = &RQF_LDLM_INTENT_GETXATTR, + .it_flags = HABEO_CORPUS, + .it_act = mdt_intent_getxattr }, [MDT_IT_LAYOUT] = { .it_fmt = &RQF_LDLM_INTENT_LAYOUT, @@ -3484,6 +2995,44 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info, remote_hdl.cookie); } +static int mdt_intent_getxattr(enum mdt_it_code opcode, + struct mdt_thread_info *info, + struct ldlm_lock **lockp, + __u64 flags) +{ + struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT]; + struct ldlm_reply *ldlm_rep = NULL; + int rc, grc; + + /* + * Initialize lhc->mlh_reg_lh either from a previously granted lock + * (for the resend case) or a new lock. Below we will use it to + * replace the original lock. + */ + mdt_intent_fixup_resent(info, *lockp, NULL, lhc); + if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) { + mdt_lock_reg_init(lhc, (*lockp)->l_req_mode); + rc = mdt_object_lock(info, info->mti_object, lhc, + MDS_INODELOCK_XATTR, + MDT_LOCAL_LOCK); + if (rc) + return rc; + } + + grc = mdt_getxattr(info); + + rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags); + + if (mdt_info_req(info)->rq_repmsg != NULL) + ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP); + if (ldlm_rep == NULL) + RETURN(err_serious(-EFAULT)); + + ldlm_rep->lock_policy_res2 = grc; + + return rc; +} + static int mdt_intent_getattr(enum mdt_it_code opcode, struct mdt_thread_info *info, struct ldlm_lock **lockp, @@ -3511,10 +3060,11 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, switch (opcode) { case MDT_IT_LOOKUP: - child_bits = MDS_INODELOCK_LOOKUP; + child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM; break; case MDT_IT_GETATTR: - child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE; + child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE | + MDS_INODELOCK_PERM; break; default: CERROR("Unsupported intent (%d)\n", opcode); @@ -3532,8 +3082,8 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, /* Get lock from request for possible resent case. */ mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc); - ldlm_rep->lock_policy_res2 = - mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep); + rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep); + ldlm_rep->lock_policy_res2 = clear_serious(rc); if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG)) ldlm_rep->lock_policy_res2 = 0; @@ -3561,15 +3111,38 @@ static int mdt_intent_layout(enum mdt_it_code opcode, __u64 flags) { struct layout_intent *layout; + struct lu_fid *fid; + struct mdt_object *obj = NULL; + struct md_object *child = NULL; int rc; ENTRY; if (opcode != MDT_IT_LAYOUT) { - CERROR("%s: Unknown intent (%d)\n", - info->mti_exp->exp_obd->obd_name, opcode); + CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt), + opcode); RETURN(-EINVAL); } + fid = &info->mti_tmp_fid2; + fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name); + + obj = mdt_object_find(info->mti_env, info->mti_mdt, fid); + if (IS_ERR(obj)) + RETURN(PTR_ERR(obj)); + + if (mdt_object_exists(obj) && !mdt_object_remote(obj)) { + child = mdt_object_child(obj); + + /* get the length of lsm */ + rc = mo_xattr_get(info->mti_env, child, &LU_BUF_NULL, + XATTR_NAME_LOV); + + if (rc > info->mti_mdt->mdt_max_mdsize) + info->mti_mdt->mdt_max_mdsize = rc; + } + + mdt_object_put(info->mti_env, obj); + (*lockp)->l_lvb_type = LVB_T_LAYOUT; req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, ldlm_lvbo_size(*lockp)); @@ -3584,7 +3157,7 @@ static int mdt_intent_layout(enum mdt_it_code opcode, RETURN(0); CERROR("%s: Unsupported layout intent (%d)\n", - info->mti_exp->exp_obd->obd_name, layout->li_opc); + mdt_obd_name(info->mti_mdt), layout->li_opc); RETURN(-EINVAL); } @@ -3605,7 +3178,7 @@ static int mdt_intent_reint(enum mdt_it_code opcode, ENTRY; - opc = mdt_reint_opcode(info, intent_fmts); + opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts); if (opc < 0) RETURN(opc); @@ -3630,14 +3203,16 @@ static int mdt_intent_reint(enum mdt_it_code opcode, if (rc != 0) mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD); - /* Cross-ref case, the lock should be returned to the client */ - if (rc == -EREMOTE) { - LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh)); - rep->lock_policy_res2 = 0; - rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags); - RETURN(rc); - } - rep->lock_policy_res2 = clear_serious(rc); + /* the open lock or the lock for cross-ref object should be + * returned to the client */ + if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) { + LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh)); + rep->lock_policy_res2 = 0; + rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags); + RETURN(rc); + } + + rep->lock_policy_res2 = clear_serious(rc); if (rep->lock_policy_res2 == -ENOENT && mdt_get_disposition(rep, DISP_LOOKUP_NEG)) @@ -3758,35 +3333,47 @@ static int mdt_intent_opc(long itopc, struct mdt_thread_info *info, rc = mdt_unpack_req_pack_rep(info, flv->it_flags); if (rc == 0) { struct ptlrpc_request *req = mdt_info_req(info); - if (flv->it_flags & MUTABOR && - req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY) - RETURN(-EROFS); + if (flv->it_flags & MUTABOR && + exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY) + RETURN(-EROFS); } if (rc == 0 && flv->it_act != NULL) { - /* execute policy */ - rc = flv->it_act(opc, info, lockp, flags); - } else { - rc = -EOPNOTSUPP; - } - RETURN(rc); + struct ldlm_reply *rep; + + /* execute policy */ + rc = flv->it_act(opc, info, lockp, flags); + + /* Check whether the reply has been packed successfully. */ + if (mdt_info_req(info)->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 { + rc = -EPROTO; + } + RETURN(rc); } static int mdt_intent_policy(struct ldlm_namespace *ns, struct ldlm_lock **lockp, void *req_cookie, ldlm_mode_t mode, __u64 flags, void *data) { - struct mdt_thread_info *info; - struct ptlrpc_request *req = req_cookie; - struct ldlm_intent *it; - struct req_capsule *pill; - int rc; + struct tgt_session_info *tsi; + struct mdt_thread_info *info; + struct ptlrpc_request *req = req_cookie; + struct ldlm_intent *it; + struct req_capsule *pill; + int rc; - ENTRY; + ENTRY; + + LASSERT(req != NULL); - LASSERT(req != NULL); + tsi = tgt_ses_info(req->rq_svc_thread->t_env); - info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx, - &mdt_thread_key); + info = tsi2mdt_info(tsi); LASSERT(info != NULL); pill = info->mti_pill; LASSERT(pill->rc_req == req); @@ -3818,100 +3405,205 @@ static int mdt_intent_policy(struct ldlm_namespace *ns, if (rc) rc = err_serious(rc); } - RETURN(rc); + mdt_thread_info_fini(info); + RETURN(rc); } -static int mdt_seq_fini(const struct lu_env *env, - struct mdt_device *m) +static void mdt_deregister_seq_exp(struct mdt_device *mdt) +{ + struct seq_server_site *ss = mdt_seq_site(mdt); + + if (ss->ss_node_id == 0) + return; + + if (ss->ss_client_seq != NULL) { + lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp); + ss->ss_client_seq->lcs_exp = NULL; + } + + if (ss->ss_server_fld != NULL) { + lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp); + ss->ss_server_fld->lsf_control_exp = NULL; + } +} + +static void mdt_seq_fini_cli(struct mdt_device *mdt) { - return seq_site_fini(env, mdt_seq_site(m)); + struct seq_server_site *ss = mdt_seq_site(mdt); + + if (ss == NULL) + return; + + if (ss->ss_server_seq == NULL) + seq_server_set_cli(NULL, ss->ss_server_seq, NULL); + + return; } -static int mdt_seq_init(const struct lu_env *env, - const char *uuid, - struct mdt_device *m) +static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt) { - struct seq_server_site *ss; - char *prefix; - int rc; + mdt_seq_fini_cli(mdt); + mdt_deregister_seq_exp(mdt); + + return seq_site_fini(env, mdt_seq_site(mdt)); +} + +/** + * It will retrieve its FLDB entries from MDT0, and it only happens + * when upgrading existent FS to 2.6 or when local FLDB is corrupted, + * and it needs to refresh FLDB from the MDT0. + **/ +static int mdt_register_lwp_callback(void *data) +{ + struct lu_env env; + struct mdt_device *mdt = data; + struct lu_server_fld *fld = mdt_seq_site(mdt)->ss_server_fld; + int rc; ENTRY; - ss = mdt_seq_site(m); + LASSERT(mdt_seq_site(mdt)->ss_node_id != 0); - /* - * This is sequence-controller node. Init seq-controller server on local - * MDT. - */ - if (ss->ss_node_id == 0) { - LASSERT(ss->ss_control_seq == NULL); + if (!likely(fld->lsf_new)) + RETURN(0); - OBD_ALLOC_PTR(ss->ss_control_seq); - if (ss->ss_control_seq == NULL) - RETURN(-ENOMEM); + rc = lu_env_init(&env, LCT_MD_THREAD); + if (rc) { + CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc); + RETURN(rc); + } + + rc = fld_update_from_controller(&env, fld); + if (rc != 0) { + CERROR("%s: cannot update controller: rc = %d\n", + mdt_obd_name(mdt), rc); + GOTO(out, rc); + } +out: + lu_env_fini(&env); + RETURN(rc); +} - rc = seq_server_init(ss->ss_control_seq, - m->mdt_bottom, uuid, - LUSTRE_SEQ_CONTROLLER, - ss, - env); +static int mdt_register_seq_exp(struct mdt_device *mdt) +{ + struct seq_server_site *ss = mdt_seq_site(mdt); + char *lwp_name = NULL; + int rc; - if (rc) - GOTO(out_seq_fini, rc); + if (ss->ss_node_id == 0) + return 0; - OBD_ALLOC_PTR(ss->ss_client_seq); - if (ss->ss_client_seq == NULL) - GOTO(out_seq_fini, rc = -ENOMEM); + OBD_ALLOC(lwp_name, MAX_OBD_NAME); + if (lwp_name == NULL) + GOTO(out_free, rc = -ENOMEM); - OBD_ALLOC(prefix, MAX_OBD_NAME + 5); - if (prefix == NULL) { - OBD_FREE_PTR(ss->ss_client_seq); - GOTO(out_seq_fini, rc = -ENOMEM); - } + rc = tgt_name2lwpname(mdt_obd_name(mdt), lwp_name); + if (rc != 0) + GOTO(out_free, rc); - snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", - uuid); + rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp, + NULL, NULL); + if (rc != 0) + GOTO(out_free, rc); - /* - * Init seq-controller client after seq-controller server is - * ready. Pass ss->ss_control_seq to it for direct talking. - */ - rc = seq_client_init(ss->ss_client_seq, NULL, - LUSTRE_SEQ_METADATA, prefix, - ss->ss_control_seq); - OBD_FREE(prefix, MAX_OBD_NAME + 5); + rc = lustre_register_lwp_item(lwp_name, + &ss->ss_server_fld->lsf_control_exp, + mdt_register_lwp_callback, mdt); + if (rc != 0) { + lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp); + ss->ss_client_seq->lcs_exp = NULL; + GOTO(out_free, rc); + } +out_free: + if (lwp_name != NULL) + OBD_FREE(lwp_name, MAX_OBD_NAME); + + return rc; +} + +/* + * Init client sequence manager which is used by local MDS to talk to sequence + * controller on remote node. + */ +static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt) +{ + struct seq_server_site *ss = mdt_seq_site(mdt); + int rc; + char *prefix; + ENTRY; + + /* check if this is adding the first MDC and controller is not yet + * initialized. */ + OBD_ALLOC_PTR(ss->ss_client_seq); + if (ss->ss_client_seq == NULL) + RETURN(-ENOMEM); + + OBD_ALLOC(prefix, MAX_OBD_NAME + 5); + if (prefix == NULL) { + OBD_FREE_PTR(ss->ss_client_seq); + ss->ss_client_seq = NULL; + RETURN(-ENOMEM); + } + + /* Note: seq_client_fini will be called in seq_site_fini */ + snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt)); + rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA, + prefix, ss->ss_node_id == 0 ? ss->ss_control_seq : + NULL); + OBD_FREE(prefix, MAX_OBD_NAME + 5); + if (rc != 0) { + OBD_FREE_PTR(ss->ss_client_seq); + ss->ss_client_seq = NULL; + RETURN(rc); + } + + rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq); + + RETURN(rc); +} +static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt) +{ + struct seq_server_site *ss; + int rc; + ENTRY; + + ss = mdt_seq_site(mdt); + /* init sequence controller server(MDT0) */ + if (ss->ss_node_id == 0) { + OBD_ALLOC_PTR(ss->ss_control_seq); + if (ss->ss_control_seq == NULL) + RETURN(-ENOMEM); + + rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom, + mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER, + ss); if (rc) GOTO(out_seq_fini, rc); } - /* Init seq-server on local MDT */ - LASSERT(ss->ss_server_seq == NULL); - + /* Init normal sequence server */ OBD_ALLOC_PTR(ss->ss_server_seq); if (ss->ss_server_seq == NULL) GOTO(out_seq_fini, rc = -ENOMEM); - rc = seq_server_init(ss->ss_server_seq, - m->mdt_bottom, uuid, - LUSTRE_SEQ_SERVER, - ss, - env); + rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom, + mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss); if (rc) - GOTO(out_seq_fini, rc = -ENOMEM); + GOTO(out_seq_fini, rc); - /* Assign seq-controller client to local seq-server. */ - if (ss->ss_node_id == 0) { - LASSERT(ss->ss_client_seq != NULL); + /* init seq client for seq server to talk to seq controller(MDT0) */ + rc = mdt_seq_init_cli(env, mdt); + if (rc != 0) + GOTO(out_seq_fini, rc); - rc = seq_server_set_cli(ss->ss_server_seq, - ss->ss_client_seq, - env); - } + if (ss->ss_node_id != 0) + /* register controler export through lwp */ + rc = mdt_register_seq_exp(mdt); EXIT; out_seq_fini: if (rc) - mdt_seq_fini(env, m); + mdt_seq_fini(env, mdt); return rc; } @@ -3949,7 +3641,7 @@ static int mdt_fld_init(const struct lu_env *env, RETURN(rc = -ENOMEM); rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid, - ss->ss_node_id, LU_SEQ_RANGE_MDT); + LU_SEQ_RANGE_MDT); if (rc) { OBD_FREE_PTR(ss->ss_server_fld); ss->ss_server_fld = NULL; @@ -3959,45 +3651,82 @@ static int mdt_fld_init(const struct lu_env *env, RETURN(0); } +static void mdt_stack_pre_fini(const struct lu_env *env, + struct mdt_device *m, struct lu_device *top) +{ + struct obd_device *obd; + struct lustre_cfg_bufs *bufs; + struct lustre_cfg *lcfg; + struct mdt_thread_info *info; + ENTRY; + + LASSERT(top); + + info = lu_context_key_get(&env->le_ctx, &mdt_thread_key); + LASSERT(info != NULL); + + bufs = &info->mti_u.bufs; + + LASSERT(m->mdt_child_exp); + LASSERT(m->mdt_child_exp->exp_obd); + obd = m->mdt_child_exp->exp_obd; + + /* process cleanup, pass mdt obd name to get obd umount flags */ + /* XXX: this is needed because all layers are referenced by + * objects (some of them are pinned by osd, for example * + * the proper solution should be a model where object used + * by osd only doesn't have mdt/mdd slices -bzzz */ + lustre_cfg_bufs_reset(bufs, mdt_obd_name(m)); + lustre_cfg_bufs_set_string(bufs, 1, NULL); + lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs); + if (!lcfg) { + CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m)); + return; + } + top->ld_ops->ldo_process_config(env, top, lcfg); + lustre_cfg_free(lcfg); + EXIT; +} + static void mdt_stack_fini(const struct lu_env *env, - struct mdt_device *m, struct lu_device *top) + struct mdt_device *m, struct lu_device *top) { - struct obd_device *obd = mdt2obd_dev(m); - struct lustre_cfg_bufs *bufs; - struct lustre_cfg *lcfg; - struct mdt_thread_info *info; - char flags[3]=""; - ENTRY; + struct obd_device *obd = mdt2obd_dev(m); + struct lustre_cfg_bufs *bufs; + struct lustre_cfg *lcfg; + struct mdt_thread_info *info; + char flags[3] = ""; + ENTRY; - info = lu_context_key_get(&env->le_ctx, &mdt_thread_key); - LASSERT(info != NULL); + info = lu_context_key_get(&env->le_ctx, &mdt_thread_key); + LASSERT(info != NULL); lu_dev_del_linkage(top->ld_site, top); lu_site_purge(env, top->ld_site, -1); - bufs = &info->mti_u.bufs; - /* process cleanup, pass mdt obd name to get obd umount flags */ + bufs = &info->mti_u.bufs; + /* process cleanup, pass mdt obd name to get obd umount flags */ /* another purpose is to let all layers to release their objects */ - lustre_cfg_bufs_reset(bufs, obd->obd_name); - if (obd->obd_force) - strcat(flags, "F"); - if (obd->obd_fail) - strcat(flags, "A"); - lustre_cfg_bufs_set_string(bufs, 1, flags); - lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs); - if (!lcfg) { - CERROR("Cannot alloc lcfg!\n"); - return; - } - LASSERT(top); - top->ld_ops->ldo_process_config(env, top, lcfg); - lustre_cfg_free(lcfg); + lustre_cfg_bufs_reset(bufs, mdt_obd_name(m)); + if (obd->obd_force) + strcat(flags, "F"); + if (obd->obd_fail) + strcat(flags, "A"); + lustre_cfg_bufs_set_string(bufs, 1, flags); + lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs); + if (!lcfg) { + CERROR("Cannot alloc lcfg!\n"); + return; + } + LASSERT(top); + top->ld_ops->ldo_process_config(env, top, lcfg); + lustre_cfg_free(lcfg); lu_site_purge(env, top->ld_site, -1); - m->mdt_child = NULL; - m->mdt_bottom = NULL; + m->mdt_child = NULL; + m->mdt_bottom = NULL; obd_disconnect(m->mdt_child_exp); m->mdt_child_exp = NULL; @@ -4021,7 +3750,7 @@ static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m, obd = class_name2obd(next); if (obd == NULL) { CERROR("%s: can't locate next device: %s\n", - m->mdt_md_dev.md_lu_dev.ld_obd->obd_name, next); + mdt_obd_name(m), next); GOTO(out, rc = -ENOTCONN); } @@ -4031,7 +3760,7 @@ static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m, rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL); if (rc) { CERROR("%s: cannot connect to next dev %s (%d)\n", - m->mdt_md_dev.md_lu_dev.ld_obd->obd_name, next, rc); + mdt_obd_name(m), next, rc); GOTO(out, rc); } @@ -4095,7 +3824,7 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, strcpy(name, dev); p = strstr(name, "-MDT"); if (p == NULL) - GOTO(cleanup_mem, rc = -ENOMEM); + GOTO(free_bufs, rc = -ENOMEM); p[3] = 'D'; snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name); @@ -4104,7 +3833,7 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, if (lprof == NULL || lprof->lp_dt == NULL) { CERROR("can't find the profile: %s\n", lustre_cfg_string(cfg, 0)); - GOTO(cleanup_mem, rc = -EINVAL); + GOTO(free_bufs, rc = -EINVAL); } lustre_cfg_bufs_reset(bufs, name); @@ -4143,13 +3872,13 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, /* connect to MDD we just setup */ rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp); if (rc) - RETURN(rc); + GOTO(class_detach, rc); site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site; LASSERT(site); - LASSERT(mdt->mdt_md_dev.md_lu_dev.ld_site == NULL); - mdt->mdt_md_dev.md_lu_dev.ld_site = site; - site->ls_top_dev = &mdt->mdt_md_dev.md_lu_dev; + LASSERT(mdt_lu_site(mdt) == NULL); + mdt->mdt_lu_dev.ld_site = site; + site->ls_top_dev = &mdt->mdt_lu_dev; mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev); @@ -4157,7 +3886,7 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, snprintf(name, MAX_OBD_NAME, "%s-osd", dev); rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp); if (rc) - RETURN(rc); + GOTO(class_detach, rc); mdt->mdt_bottom = lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev); @@ -4166,7 +3895,7 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, if (rc != 0) CERROR("Failure to refill session: '%d'\n", rc); - lu_dev_add_linkage(site, &mdt->mdt_md_dev.md_lu_dev); + lu_dev_add_linkage(site, &mdt->mdt_lu_dev); EXIT; class_detach: @@ -4277,7 +4006,7 @@ static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt, /* configure local quota objects */ rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env, - &mdt->mdt_md_dev.md_lu_dev, + &mdt->mdt_lu_dev, mdt->mdt_qmt_dev); if (rc) GOTO(class_cleanup, rc); @@ -4332,10 +4061,128 @@ static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt) EXIT; } +/* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper + * for now. This will be removed along with converting rest of MDT code + * to use tgt_session_info */ +int mdt_tgt_getxattr(struct tgt_session_info *tsi) +{ + struct mdt_thread_info *info = tsi2mdt_info(tsi); + int rc; + + rc = mdt_getxattr(info); + + mdt_thread_info_fini(info); + return rc; +} + +static struct tgt_handler mdt_tgt_handlers[] = { +TGT_RPC_HANDLER(MDS_FIRST_OPC, + 0, MDS_CONNECT, mdt_tgt_connect, + &RQF_CONNECT, LUSTRE_OBD_VERSION), +TGT_RPC_HANDLER(MDS_FIRST_OPC, + 0, MDS_DISCONNECT, tgt_disconnect, + &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION), +TGT_RPC_HANDLER(MDS_FIRST_OPC, + HABEO_REFERO, MDS_SET_INFO, mdt_set_info, + &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION), +TGT_MDT_HDL(0, MDS_GET_INFO, mdt_get_info), +TGT_MDT_HDL(0 | HABEO_REFERO, MDS_GETSTATUS, mdt_getstatus), +TGT_MDT_HDL(HABEO_CORPUS, MDS_GETATTR, mdt_getattr), +TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME, + mdt_getattr_name), +TGT_MDT_HDL(HABEO_CORPUS, MDS_GETXATTR, mdt_tgt_getxattr), +TGT_MDT_HDL(0 | HABEO_REFERO, MDS_STATFS, mdt_statfs), +TGT_MDT_HDL(0 | MUTABOR, MDS_REINT, mdt_reint), +TGT_MDT_HDL(HABEO_CORPUS, MDS_CLOSE, mdt_close), +TGT_MDT_HDL(HABEO_CORPUS, MDS_DONE_WRITING, + mdt_done_writing), +TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE, mdt_readpage), +TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC, mdt_sync), +TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_IS_SUBDIR, mdt_is_subdir), +TGT_MDT_HDL(0, MDS_QUOTACTL, mdt_quotactl), +TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS, + mdt_hsm_progress), +TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER, + mdt_hsm_ct_register), +TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER, + mdt_hsm_ct_unregister), +TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET, + mdt_hsm_state_get), +TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET, + mdt_hsm_state_set), +TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action), +TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST, + mdt_hsm_request), +TGT_MDT_HDL(HABEO_CORPUS|HABEO_REFERO | MUTABOR, MDS_SWAP_LAYOUTS, + mdt_swap_layouts) +}; + +static struct tgt_handler mdt_sec_ctx_ops[] = { +TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, mdt_sec_ctx_handle), +TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT,mdt_sec_ctx_handle), +TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, mdt_sec_ctx_handle) +}; + +static struct tgt_handler mdt_quota_ops[] = { +TGT_QUOTA_HDL(HABEO_REFERO, QUOTA_DQACQ, mdt_quota_dqacq), +}; + +static struct tgt_opc_slice mdt_common_slice[] = { + { + .tos_opc_start = MDS_FIRST_OPC, + .tos_opc_end = MDS_LAST_OPC, + .tos_hs = mdt_tgt_handlers + }, + { + .tos_opc_start = OBD_FIRST_OPC, + .tos_opc_end = OBD_LAST_OPC, + .tos_hs = tgt_obd_handlers + }, + { + .tos_opc_start = LDLM_FIRST_OPC, + .tos_opc_end = LDLM_LAST_OPC, + .tos_hs = tgt_dlm_handlers + }, + { + .tos_opc_start = SEC_FIRST_OPC, + .tos_opc_end = SEC_LAST_OPC, + .tos_hs = mdt_sec_ctx_ops + }, + { + .tos_opc_start = UPDATE_OBJ, + .tos_opc_end = UPDATE_LAST_OPC, + .tos_hs = tgt_out_handlers + }, + { + .tos_opc_start = FLD_FIRST_OPC, + .tos_opc_end = FLD_LAST_OPC, + .tos_hs = fld_handlers + }, + { + .tos_opc_start = SEQ_FIRST_OPC, + .tos_opc_end = SEQ_LAST_OPC, + .tos_hs = seq_handlers + }, + { + .tos_opc_start = QUOTA_DQACQ, + .tos_opc_end = QUOTA_LAST_OPC, + .tos_hs = mdt_quota_ops + }, + { + .tos_opc_start = LLOG_FIRST_OPC, + .tos_opc_end = LLOG_LAST_OPC, + .tos_hs = tgt_llog_handlers + }, + + { + .tos_hs = NULL + } +}; + static void mdt_fini(const struct lu_env *env, struct mdt_device *m) { - struct md_device *next = m->mdt_child; - struct lu_device *d = &m->mdt_md_dev.md_lu_dev; + struct md_device *next = m->mdt_child; + struct lu_device *d = &m->mdt_lu_dev; struct obd_device *obd = mdt2obd_dev(m); ENTRY; @@ -4343,6 +4190,14 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) ping_evictor_stop(); + mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child)); + + if (m->mdt_opts.mo_coordinator) + mdt_hsm_cdt_stop(m); + + mdt_hsm_cdt_fini(m); + + mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT); mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT); obd_exports_barrier(obd); obd_zombie_barrier(); @@ -4369,48 +4224,25 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) m->mdt_nosquash_strlen = 0; } + next->md_ops->mdo_iocontrol(env, next, OBD_IOC_PAUSE_LFSCK, 0, NULL); + mdt_seq_fini(env, m); mdt_fld_fini(env, m); - sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset); next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL); cfs_timer_disarm(&m->mdt_ck_timer); mdt_ck_thread_stop(m); - /* - * Finish the stack - */ - mdt_stack_fini(env, m, md2lu_dev(m->mdt_child)); - - LASSERT(cfs_atomic_read(&d->ld_ref) == 0); - - server_put_mount(mdt2obd_dev(m)->obd_name, NULL); - - EXIT; -} - -static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial) -{ - struct mdt_device *m = mdt_dev(obd->obd_lu_dev); - struct sptlrpc_rule_set tmp_rset; - int rc; - - sptlrpc_rule_set_init(&tmp_rset); - rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial); - if (rc) { - CERROR("mdt %s: failed get sptlrpc rules: %d\n", - obd->obd_name, rc); - return rc; - } + /* + * Finish the stack + */ + mdt_stack_fini(env, m, md2lu_dev(m->mdt_child)); - sptlrpc_target_update_exp_flavor(obd, &tmp_rset); + LASSERT(atomic_read(&d->ld_ref) == 0); - write_lock(&m->mdt_sptlrpc_lock); - sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset); - m->mdt_sptlrpc_rset = tmp_rset; - write_unlock(&m->mdt_sptlrpc_lock); + server_put_mount(mdt_obd_name(m)); - return 0; + EXIT; } int mdt_postrecov(const struct lu_env *, struct mdt_device *); @@ -4429,11 +4261,11 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, const char *identity_upcall = "NONE"; struct md_device *next; int rc; - int node_id; + long node_id; mntopt_t mntopts; ENTRY; - md_device_init(&m->mdt_md_dev, ldt); + lu_device_init(&m->mdt_lu_dev, ldt); /* * Environment (env) might be missing mdt_thread_key values at that * point, if device is allocated when mdt_thread_key is in QUIESCENT @@ -4457,6 +4289,11 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, m->mdt_som_conf = 0; m->mdt_opts.mo_cos = MDT_COS_DEFAULT; + + /* default is coordinator off, it is started through conf_param + * or /proc */ + m->mdt_opts.mo_coordinator = 0; + lmi = server_get_mount(dev); if (lmi == NULL) { CERROR("Cannot get mount info for %s!\n", dev); @@ -4467,15 +4304,9 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, LASSERT(num); node_id = simple_strtol(num, NULL, 10); obd->u.obt.obt_magic = OBT_MAGIC; - } - - rwlock_init(&m->mdt_sptlrpc_lock); - sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset); + } spin_lock_init(&m->mdt_ioepoch_lock); - m->mdt_opts.mo_compat_resname = 0; - m->mdt_opts.mo_mds_capa = 1; - m->mdt_opts.mo_oss_capa = 1; m->mdt_capa_timeout = CAPA_TIMEOUT; m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1; m->mdt_ck_timeout = CAPA_KEY_TIMEOUT; @@ -4487,21 +4318,24 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, init_rwsem(&m->mdt_squash_sem); spin_lock_init(&m->mdt_osfs_lock); m->mdt_osfs_age = cfs_time_shift_64(-1000); + m->mdt_enable_remote_dir = 0; + m->mdt_enable_remote_dir_gid = 0; - m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops; - m->mdt_md_dev.md_lu_dev.ld_obd = obd; - /* set this lu_device to obd, because error handling need it */ - obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev; + m->mdt_lu_dev.ld_ops = &mdt_lu_ops; + m->mdt_lu_dev.ld_obd = obd; + /* Set this lu_device to obd for error handling purposes. */ + obd->obd_lu_dev = &m->mdt_lu_dev; /* init the stack */ rc = mdt_stack_init((struct lu_env *)env, m, cfg); if (rc) { - CERROR("Can't init device stack, rc %d\n", rc); + CERROR("%s: Can't init device stack, rc %d\n", + mdt_obd_name(m), rc); GOTO(err_lmi, rc); } - s = m->mdt_md_dev.md_lu_dev.ld_site; - ss_site = &m->mdt_seq_site; + s = mdt_lu_site(m); + ss_site = mdt_seq_site(m); s->ld_seq_site = ss_site; ss_site->ss_lu = s; @@ -4516,28 +4350,24 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, /* No connection accepted until configurations will finish */ obd->obd_no_conn = 1; - if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) { - char *str = lustre_cfg_string(cfg, 4); - if (strchr(str, 'n')) { - CWARN("%s: recovery disabled\n", obd->obd_name); - obd->obd_replayable = 0; - } - } - - rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom); - if (rc) - GOTO(err_fini_stack, rc); + if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) { + char *str = lustre_cfg_string(cfg, 4); + if (strchr(str, 'n')) { + CWARN("%s: recovery disabled\n", mdt_obd_name(m)); + obd->obd_replayable = 0; + } + } - rc = mdt_fld_init(env, obd->obd_name, m); + rc = mdt_fld_init(env, mdt_obd_name(m), m); if (rc) - GOTO(err_lut, rc); + GOTO(err_fini_stack, rc); - rc = mdt_seq_init(env, obd->obd_name, m); + rc = mdt_seq_init(env, m); if (rc) GOTO(err_fini_fld, rc); - snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name, - LUSTRE_MDT_NAME"-%p", m); + snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s", + LUSTRE_MDT_NAME, obd->obd_uuid.uuid); m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name, LDLM_NAMESPACE_SERVER, LDLM_NAMESPACE_GREEDY, @@ -4554,44 +4384,62 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m); - rc = mdt_ck_thread_start(m); - if (rc) + rc = mdt_hsm_cdt_init(m); + if (rc != 0) { + CERROR("%s: error initializing coordinator, rc %d\n", + mdt_obd_name(m), rc); GOTO(err_free_ns, rc); + } - rc = mdt_fs_setup(env, m, obd, lsi); + rc = mdt_ck_thread_start(m); if (rc) - GOTO(err_capa, rc); + GOTO(err_free_hsm, rc); + + rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice, + OBD_FAIL_MDS_ALL_REQUEST_NET, + OBD_FAIL_MDS_ALL_REPLY_NET); + if (rc) + GOTO(err_capa, rc); - mdt_adapt_sptlrpc_conf(obd, 1); + rc = mdt_fs_setup(env, m, obd, lsi); + if (rc) + GOTO(err_tgt, rc); + + tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1); next = m->mdt_child; rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0, &mntopts); if (rc) - GOTO(err_llog_cleanup, rc); + GOTO(err_fs_cleanup, rc); if (mntopts & MNTOPT_USERXATTR) m->mdt_opts.mo_user_xattr = 1; else m->mdt_opts.mo_user_xattr = 0; + rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size); + if (rc) + GOTO(err_fs_cleanup, rc); + if (mntopts & MNTOPT_ACL) m->mdt_opts.mo_acl = 1; else m->mdt_opts.mo_acl = 0; - /* XXX: to support suppgid for ACL, we enable identity_upcall - * by default, otherwise, maybe got unexpected -EACCESS. */ - if (m->mdt_opts.mo_acl) - identity_upcall = MDT_IDENTITY_UPCALL_PATH; - - m->mdt_identity_cache = upcall_cache_init(obd->obd_name,identity_upcall, - &mdt_identity_upcall_cache_ops); - if (IS_ERR(m->mdt_identity_cache)) { - rc = PTR_ERR(m->mdt_identity_cache); - m->mdt_identity_cache = NULL; - GOTO(err_llog_cleanup, rc); - } + /* XXX: to support suppgid for ACL, we enable identity_upcall + * by default, otherwise, maybe got unexpected -EACCESS. */ + if (m->mdt_opts.mo_acl) + identity_upcall = MDT_IDENTITY_UPCALL_PATH; + + m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m), + identity_upcall, + &mdt_identity_upcall_cache_ops); + if (IS_ERR(m->mdt_identity_cache)) { + rc = PTR_ERR(m->mdt_identity_cache); + m->mdt_identity_cache = NULL; + GOTO(err_fs_cleanup, rc); + } rc = mdt_procfs_init(m, dev); if (rc) { @@ -4607,7 +4455,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL, "mdt_ldlm_client", m->mdt_ldlm_client); - ping_evictor_start(); + ping_evictor_start(); /* recovery will be started upon mdt_prepare() * when the whole stack is complete and ready @@ -4622,34 +4470,34 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT; RETURN(0); - err_procfs: - mdt_procfs_fini(m); + mdt_procfs_fini(m); err_recovery: - target_recovery_fini(obd); - upcall_cache_cleanup(m->mdt_identity_cache); - m->mdt_identity_cache = NULL; -err_llog_cleanup: - mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT); - mdt_fs_cleanup(env, m); + target_recovery_fini(obd); + upcall_cache_cleanup(m->mdt_identity_cache); + m->mdt_identity_cache = NULL; +err_fs_cleanup: + mdt_fs_cleanup(env, m); +err_tgt: + tgt_fini(env, &m->mdt_lut); err_capa: - cfs_timer_disarm(&m->mdt_ck_timer); - mdt_ck_thread_stop(m); + cfs_timer_disarm(&m->mdt_ck_timer); + mdt_ck_thread_stop(m); +err_free_hsm: + mdt_hsm_cdt_fini(m); err_free_ns: - ldlm_namespace_free(m->mdt_namespace, NULL, 0); - obd->obd_namespace = m->mdt_namespace = NULL; + ldlm_namespace_free(m->mdt_namespace, NULL, 0); + obd->obd_namespace = m->mdt_namespace = NULL; err_fini_seq: mdt_seq_fini(env, m); err_fini_fld: mdt_fld_fini(env, m); -err_lut: - tgt_fini(env, &m->mdt_lut); err_fini_stack: - mdt_stack_fini(env, m, md2lu_dev(m->mdt_child)); + mdt_stack_fini(env, m, md2lu_dev(m->mdt_child)); err_lmi: if (lmi) - server_put_mount(dev, lmi->lmi_mnt); - return (rc); + server_put_mount(dev); + return(rc); } /* For interoperability, the left element is old parameter, the right one @@ -4714,9 +4562,16 @@ static int mdt_process_config(const struct lu_env *env, lprocfs_mdt_init_vars(&lvars); rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, cfg, obd); - if (rc > 0 || rc == -ENOSYS) - /* we don't understand; pass it on */ - rc = next->ld_ops->ldo_process_config(env, next, cfg); + if (rc > 0 || rc == -ENOSYS) { + /* is it an HSM var ? */ + rc = class_process_proc_param(PARAM_HSM, + hsm_cdt_get_proc_vars(), + cfg, obd); + if (rc > 0 || rc == -ENOSYS) + /* we don't understand; pass it on */ + rc = next->ld_ops->ldo_process_config(env, next, + cfg); + } if (old_cfg != NULL) lustre_cfg_free(cfg); @@ -4739,12 +4594,12 @@ static struct lu_object *mdt_object_alloc(const struct lu_env *env, ENTRY; - OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, CFS_ALLOC_IO); + OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, __GFP_IO); if (mo != NULL) { struct lu_object *o; struct lu_object_header *h; - o = &mo->mot_obj.mo_lu; + o = &mo->mot_obj; h = &mo->mot_header; lu_object_header_init(h); lu_object_init(o, h, d); @@ -4752,9 +4607,10 @@ static struct lu_object *mdt_object_alloc(const struct lu_env *env, o->lo_ops = &mdt_obj_ops; mutex_init(&mo->mot_ioepoch_mutex); mutex_init(&mo->mot_lov_mutex); - RETURN(o); - } else - RETURN(NULL); + init_rwsem(&mo->mot_open_sem); + RETURN(o); + } + RETURN(NULL); } static int mdt_object_init(const struct lu_env *env, struct lu_object *o, @@ -4789,11 +4645,14 @@ static void mdt_object_free(const struct lu_env *env, struct lu_object *o) CDEBUG(D_INFO, "object free, fid = "DFID"\n", PFID(lu_object_fid(o))); - lu_object_fini(o); - lu_object_header_fini(h); + LASSERT(atomic_read(&mo->mot_open_count) == 0); + LASSERT(atomic_read(&mo->mot_lease_count) == 0); + + lu_object_fini(o); + lu_object_header_fini(h); OBD_SLAB_FREE_PTR(mo, mdt_object_kmem); - EXIT; + EXIT; } static int mdt_object_print(const struct lu_env *env, void *cookie, @@ -4813,6 +4672,7 @@ static int mdt_prepare(const struct lu_env *env, struct mdt_device *mdt = mdt_dev(cdev); struct lu_device *next = &mdt->mdt_child->md_lu_dev; struct obd_device *obd = cdev->ld_obd; + struct lfsck_start_param lsp; int rc; ENTRY; @@ -4827,13 +4687,30 @@ static int mdt_prepare(const struct lu_env *env, if (rc) RETURN(rc); - rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child, - &mdt->mdt_md_root_fid); + rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT); if (rc) RETURN(rc); + lsp.lsp_start = NULL; + lsp.lsp_namespace = mdt->mdt_namespace; + rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child, + OBD_IOC_START_LFSCK, + 0, &lsp); + if (rc != 0) { + CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n", + mdt_obd_name(mdt), rc); + rc = 0; + } + + if (mdt->mdt_seq_site.ss_node_id == 0) { + rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child, + &mdt->mdt_md_root_fid); + if (rc) + RETURN(rc); + } + LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state)); - target_recovery_init(&mdt->mdt_lut, mdt_recovery_handle); + target_recovery_init(&mdt->mdt_lut, tgt_request_handle); set_bit(MDT_FL_CFGLOG, &mdt->mdt_state); LASSERT(obd->obd_no_conn); spin_lock(&obd->obd_dev_lock); @@ -4864,18 +4741,16 @@ static int mdt_obd_set_info_async(const struct lu_env *env, __u32 vallen, void *val, struct ptlrpc_request_set *set) { - struct obd_device *obd = exp->exp_obd; - int rc; - ENTRY; + int rc; - LASSERT(obd); + ENTRY; - if (KEY_IS(KEY_SPTLRPC_CONF)) { - rc = mdt_adapt_sptlrpc_conf(obd, 0); - RETURN(rc); - } + if (KEY_IS(KEY_SPTLRPC_CONF)) { + rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0); + RETURN(rc); + } - RETURN(0); + RETURN(0); } /** @@ -4884,12 +4759,12 @@ static int mdt_obd_set_info_async(const struct lu_env *env, * Compute the compatibility flags for a connection request based on * features mutually supported by client and server. * - * The obd_export::exp_connect_flags field in \a exp must not be updated - * here, otherwise a partially initialized value may be exposed. After - * the connection request is successfully processed, the top-level MDT - * connect request handler atomically updates the export connect flags - * from the obd_connect_data::ocd_connect_flags field of the reply. - * \see mdt_connect(). + * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp + * must not be updated here, otherwise a partially initialized value may + * be exposed. After the connection request is successfully processed, + * the top-level MDT connect request handler atomically updates the export + * connect flags from the obd_connect_data::ocd_connect_flags field of the + * reply. \see mdt_connect(). * * \param exp the obd_export associated with this client/target pair * \param mdt the target device for the connection @@ -4926,7 +4801,7 @@ static int mdt_connect_internal(struct obd_export *exp, if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) { data->ocd_brw_size = min(data->ocd_brw_size, - (__u32)(PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT)); + (__u32)MD_MAX_BRW_SIZE); if (data->ocd_brw_size == 0) { CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64 " ocd_version: %x ocd_grant: %d " @@ -4934,7 +4809,7 @@ static int mdt_connect_internal(struct obd_export *exp, "unexpectedly zero, network data " "corruption? Refusing connection of this" " client\n", - exp->exp_obd->obd_name, + mdt_obd_name(mdt), exp->exp_client_uuid.uuid, exp, data->ocd_connect_flags, data->ocd_version, data->ocd_grant, data->ocd_index); @@ -4942,22 +4817,22 @@ static int mdt_connect_internal(struct obd_export *exp, } } - /* NB: Disregard the rule against updating exp_connect_flags in this - * case, since tgt_client_new() needs to know if this is a lightweight - * connection, and it is safe to expose this flag before connection - * processing completes. */ + /* NB: Disregard the rule against updating + * exp_connect_data.ocd_connect_flags in this case, since + * tgt_client_new() needs to know if this is a lightweight + * connection, and it is safe to expose this flag before + * connection processing completes. */ if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) { spin_lock(&exp->exp_lock); - exp->exp_connect_flags |= OBD_CONNECT_LIGHTWEIGHT; + *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT; spin_unlock(&exp->exp_lock); } data->ocd_version = LUSTRE_VERSION_CODE; - exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known; if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) { CWARN("%s: MDS requires FID support, but client not\n", - mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name); + mdt_obd_name(mdt)); return -EBADE; } @@ -4966,55 +4841,23 @@ static int mdt_connect_internal(struct obd_export *exp, OBD_CONNECT_MDS_MDS | OBD_CONNECT_SOM))) { CWARN("%s: MDS has SOM enabled, but client does not support " - "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name); + "it\n", mdt_obd_name(mdt)); return -EBADE; } - return 0; -} - -static int mdt_connect_check_sptlrpc(struct mdt_device *mdt, - struct obd_export *exp, - struct ptlrpc_request *req) -{ - struct sptlrpc_flavor flvr; - int rc = 0; - - if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) { - read_lock(&mdt->mdt_sptlrpc_lock); - sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset, - req->rq_sp_from, - req->rq_peer.nid, - &flvr); - read_unlock(&mdt->mdt_sptlrpc_lock); - - spin_lock(&exp->exp_lock); - - exp->exp_sp_peer = req->rq_sp_from; - exp->exp_flvr = flvr; - - if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY && - exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) { - CERROR("unauthorized rpc flavor %x from %s, " - "expect %x\n", req->rq_flvr.sf_rpc, - libcfs_nid2str(req->rq_peer.nid), - exp->exp_flvr.sf_rpc); - rc = -EACCES; - } + if (OCD_HAS_FLAG(data, PINGLESS)) { + if (ptlrpc_pinger_suppress_pings()) { + spin_lock(&exp->exp_obd->obd_dev_lock); + list_del_init(&exp->exp_obd_chain_timed); + spin_unlock(&exp->exp_obd->obd_dev_lock); + } else { + data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS; + } + } - spin_unlock(&exp->exp_lock); - } else { - if (exp->exp_sp_peer != req->rq_sp_from) { - CERROR("RPC source %s doesn't match %s\n", - sptlrpc_part2name(req->rq_sp_from), - sptlrpc_part2name(exp->exp_sp_peer)); - rc = -EACCES; - } else { - rc = sptlrpc_target_export_check(exp, req); - } - } + data->ocd_max_easize = mdt->mdt_max_ea_size; - return rc; + return 0; } /* mds_connect copy */ @@ -5024,11 +4867,9 @@ static int mdt_obd_connect(const struct lu_env *env, struct obd_connect_data *data, void *localdata) { - struct mdt_thread_info *info; struct obd_export *lexp; struct lustre_handle conn = { 0 }; struct mdt_device *mdt; - struct ptlrpc_request *req; int rc; ENTRY; @@ -5036,18 +4877,18 @@ static int mdt_obd_connect(const struct lu_env *env, if (!exp || !obd || !cluuid) RETURN(-EINVAL); - info = lu_context_key_get(&env->le_ctx, &mdt_thread_key); - req = info->mti_pill->rc_req; - mdt = mdt_dev(obd->obd_lu_dev); + mdt = mdt_dev(obd->obd_lu_dev); /* * first, check whether the stack is ready to handle requests * XXX: probably not very appropriate method is used now * at some point we should find a better one */ - if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && + if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL && !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) { - rc = obd_health_check(env, mdt->mdt_child_exp->exp_obd); + rc = obd_get_info(env, mdt->mdt_child_exp, + sizeof(KEY_OSP_CONNECTED), + KEY_OSP_CONNECTED, NULL, NULL, NULL); if (rc) RETURN(-EAGAIN); set_bit(MDT_FL_SYNCED, &mdt->mdt_state); @@ -5060,26 +4901,17 @@ static int mdt_obd_connect(const struct lu_env *env, lexp = class_conn2export(&conn); LASSERT(lexp != NULL); - rc = mdt_connect_check_sptlrpc(mdt, lexp, req); - if (rc) - GOTO(out, rc); - - if (OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG)) - lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING); - rc = mdt_connect_internal(lexp, mdt, data); if (rc == 0) { struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd; LASSERT(lcd); - info->mti_exp = lexp; memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid); rc = tgt_client_new(env, lexp); if (rc == 0) mdt_export_stats_init(obd, lexp, localdata); } -out: if (rc != 0) { class_disconnect(lexp); *exp = NULL; @@ -5096,23 +4928,12 @@ static int mdt_obd_reconnect(const struct lu_env *env, struct obd_connect_data *data, void *localdata) { - struct mdt_thread_info *info; - struct mdt_device *mdt; - struct ptlrpc_request *req; int rc; ENTRY; if (exp == NULL || obd == NULL || cluuid == NULL) RETURN(-EINVAL); - info = lu_context_key_get(&env->le_ctx, &mdt_thread_key); - req = info->mti_pill->rc_req; - mdt = mdt_dev(obd->obd_lu_dev); - - rc = mdt_connect_check_sptlrpc(mdt, exp, req); - if (rc) - RETURN(rc); - rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data); if (rc == 0) mdt_export_stats_init(obd, exp, localdata); @@ -5120,6 +4941,31 @@ static int mdt_obd_reconnect(const struct lu_env *env, RETURN(rc); } +static int mdt_ctxt_add_dirty_flag(struct lu_env *env, + struct mdt_thread_info *info, + struct mdt_file_data *mfd) +{ + struct lu_context ses; + int rc; + ENTRY; + + rc = lu_context_init(&ses, LCT_SERVER_SESSION); + if (rc) + RETURN(rc); + + env->le_ses = &ses; + lu_context_enter(&ses); + + mdt_ucred(info)->uc_valid = UCRED_OLD; + rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr); + + lu_context_exit(&ses); + lu_context_fini(&ses); + env->le_ses = NULL; + + RETURN(rc); +} + static int mdt_export_cleanup(struct obd_export *exp) { struct mdt_export_data *med = &exp->exp_mdt_data; @@ -5164,6 +5010,24 @@ static int mdt_export_cleanup(struct obd_export *exp) cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) { cfs_list_del_init(&mfd->mfd_list); ma->ma_need = ma->ma_valid = 0; + + /* This file is being closed due to an eviction, it + * could have been modified and now dirty regarding to + * HSM archive, check this! + * The logic here is to mark a file dirty if there's a + * chance it was dirtied before the client was evicted, + * so that we don't have to wait for a release attempt + * before finding out the file was actually dirty and + * fail the release. Aggressively marking it dirty here + * will cause the policy engine to attempt to + * re-archive it; when rearchiving, we can compare the + * current version to the HSM data_version and make the + * archive request into a noop if it's not actually + * dirty. + */ + if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC)) + 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) { ma->ma_valid = MA_FLAGS; @@ -5260,32 +5124,199 @@ static int mdt_destroy_export(struct obd_export *exp) RETURN(0); } -static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, - void *val, int vallen) +/** The maximum depth that fid2path() will search. + * This is limited only because we want to store the fids for + * historical path lookup purposes. + */ +#define MAX_PATH_DEPTH 100 + +/** mdt_path() lookup structure. */ +struct path_lookup_info { + __u64 pli_recno; /**< history point */ + __u64 pli_currec; /**< current record */ + struct lu_fid pli_fid; + struct lu_fid pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */ + struct mdt_object *pli_mdt_obj; + char *pli_path; /**< full path */ + int pli_pathlen; + int pli_linkno; /**< which hardlink to follow */ + int pli_fidcount; /**< number of \a pli_fids */ +}; + +static int mdt_links_read(struct mdt_thread_info *info, + struct mdt_object *mdt_obj, struct linkea_data *ldata) { - struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev); - struct getinfo_fid2path *fpout, *fpin; - int rc = 0; + int rc; - fpin = key + cfs_size_round(sizeof(KEY_FID2PATH)); - fpout = val; + LASSERT(ldata->ld_buf->lb_buf != NULL); + + if (!mdt_object_exists(mdt_obj)) + return -ENODATA; + + rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj), + ldata->ld_buf, XATTR_NAME_LINK); + if (rc == -ERANGE) { + /* Buf was too small, figure out what we need. */ + lu_buf_free(ldata->ld_buf); + rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj), + ldata->ld_buf, XATTR_NAME_LINK); + if (rc < 0) + return rc; + ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc); + if (ldata->ld_buf->lb_buf == NULL) + return -ENOMEM; + rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj), + ldata->ld_buf, XATTR_NAME_LINK); + } + if (rc < 0) + return rc; - if (ptlrpc_req_need_swab(info->mti_pill->rc_req)) - lustre_swab_fid2path(fpin); + return linkea_init(ldata); +} - memcpy(fpout, fpin, sizeof(*fpin)); - if (fpout->gf_pathlen != vallen - sizeof(*fpin)) - RETURN(-EINVAL); +static int mdt_path_current(struct mdt_thread_info *info, + struct path_lookup_info *pli) +{ + 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 = { 0 }; + int rc = 0; + ENTRY; - rc = mdt_fid2path(info->mti_env, mdt, fpout); - RETURN(rc); + /* temp buffer for path element, the buffer will be finally freed + * in mdt_thread_info_fini */ + buf = lu_buf_check_and_alloc(buf, PATH_MAX); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + + ldata.ld_buf = buf; + ptr = pli->pli_path + pli->pli_pathlen - 1; + *ptr = 0; + --ptr; + pli->pli_fidcount = 0; + pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj); + + /* root FID only exists on MDT0, and fid2path should also ends at MDT0, + * so checking root_fid can only happen on MDT0. */ + while (!lu_fid_eq(&mdt->mdt_md_root_fid, + &pli->pli_fids[pli->pli_fidcount])) { + mdt_obj = mdt_object_find(info->mti_env, mdt, + &pli->pli_fids[pli->pli_fidcount]); + if (IS_ERR(mdt_obj)) + GOTO(out, rc = PTR_ERR(mdt_obj)); + if (mdt_object_remote(mdt_obj)) { + mdt_object_put(info->mti_env, mdt_obj); + GOTO(remote_out, rc = -EREMOTE); + } + if (!mdt_object_exists(mdt_obj)) { + mdt_object_put(info->mti_env, mdt_obj); + GOTO(out, rc = -ENOENT); + } + + rc = mdt_links_read(info, mdt_obj, &ldata); + mdt_object_put(info->mti_env, mdt_obj); + if (rc != 0) + GOTO(out, rc); + + leh = buf->lb_buf; + lee = (struct link_ea_entry *)(leh + 1); /* link #0 */ + linkea_entry_unpack(lee, &reclen, tmpname, tmpfid); + /* If set, use link #linkno for path lookup, otherwise use + link #0. Only do this for the final path element. */ + if (pli->pli_fidcount == 0 && + pli->pli_linkno < leh->leh_reccount) { + int count; + for (count = 0; count < pli->pli_linkno; count++) { + lee = (struct link_ea_entry *) + ((char *)lee + reclen); + linkea_entry_unpack(lee, &reclen, tmpname, + tmpfid); + } + if (pli->pli_linkno < leh->leh_reccount - 1) + /* indicate to user there are more links */ + pli->pli_linkno++; + } + + /* Pack the name in the end of the buffer */ + ptr -= tmpname->ln_namelen; + if (ptr - 1 <= pli->pli_path) + GOTO(out, rc = -EOVERFLOW); + strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen); + *(--ptr) = '/'; + + /* Store the parent fid for historic lookup */ + if (++pli->pli_fidcount >= MAX_PATH_DEPTH) + GOTO(out, rc = -EOVERFLOW); + pli->pli_fids[pli->pli_fidcount] = *tmpfid; + } + +remote_out: + ptr++; /* skip leading / */ + memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr); + + EXIT; +out: + return rc; +} + +/* Returns the full path to this fid, as of changelog record recno. */ +static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj, + char *path, int pathlen, __u64 *recno, int *linkno, + struct lu_fid *fid) +{ + struct mdt_device *mdt = info->mti_mdt; + struct path_lookup_info *pli; + int tries = 3; + int rc = -EAGAIN; + ENTRY; + + if (pathlen < 3) + RETURN(-EOVERFLOW); + + if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) { + path[0] = '\0'; + RETURN(0); + } + + OBD_ALLOC_PTR(pli); + if (pli == NULL) + RETURN(-ENOMEM); + + pli->pli_mdt_obj = obj; + pli->pli_recno = *recno; + pli->pli_path = path; + pli->pli_pathlen = pathlen; + pli->pli_linkno = *linkno; + + /* Retry multiple times in case file is being moved */ + while (tries-- && rc == -EAGAIN) + rc = mdt_path_current(info, pli); + + /* return the last resolved fids to the client, so the client will + * build the left path on another MDT for remote object */ + *fid = pli->pli_fids[pli->pli_fidcount]; + + *recno = pli->pli_currec; + /* Return next link index to caller */ + *linkno = pli->pli_linkno; + + OBD_FREE_PTR(pli); + + RETURN(rc); } -static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt, - struct getinfo_fid2path *fp) +static int mdt_fid2path(struct mdt_thread_info *info, + struct getinfo_fid2path *fp) { + struct mdt_device *mdt = info->mti_mdt; struct mdt_object *obj; - struct obd_device *obd = mdt2obd_dev(mdt); int rc; ENTRY; @@ -5295,119 +5326,156 @@ static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt, if (!fid_is_sane(&fp->gf_fid)) RETURN(-EINVAL); - if (!fid_is_norm(&fp->gf_fid) && !fid_is_igif(&fp->gf_fid)) { + if (!fid_is_namespace_visible(&fp->gf_fid)) { CWARN("%s: "DFID" is invalid, sequence should be " - ">= "LPX64"\n", obd->obd_name, - PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL); + ">= "LPX64"\n", mdt_obd_name(mdt), + PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL); RETURN(-EINVAL); } - obj = mdt_object_find(env, mdt, &fp->gf_fid); + obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid); if (obj == NULL || IS_ERR(obj)) { CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid), - PTR_ERR(obj)); + PTR_ERR(obj)); RETURN(-EINVAL); } - rc = lu_object_exists(&obj->mot_obj.mo_lu); - if (rc <= 0) { - if (rc == -1) - rc = -EREMOTE; - else - rc = -ENOENT; - mdt_object_put(env, obj); + if (mdt_object_remote(obj)) + rc = -EREMOTE; + else if (!mdt_object_exists(obj)) + rc = -ENOENT; + else + rc = 0; + + if (rc < 0) { + mdt_object_put(info->mti_env, obj); CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n", - PFID(&fp->gf_fid), rc); + PFID(&fp->gf_fid), rc); RETURN(rc); } - rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path, - fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno); - mdt_object_put(env, obj); + rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno, + &fp->gf_linkno, &fp->gf_fid); + + CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n", + PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno); + + mdt_object_put(info->mti_env, obj); RETURN(rc); } -int mdt_get_info(struct mdt_thread_info *info) +static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, + void *val, int vallen) { - struct ptlrpc_request *req = mdt_info_req(info); - char *key; - int keylen; - __u32 *vallen; - void *valout; - int rc; - ENTRY; + struct getinfo_fid2path *fpout, *fpin; + int rc = 0; - key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY); - if (key == NULL) { - CDEBUG(D_IOCTL, "No GETINFO key"); - RETURN(-EFAULT); - } - keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY, - RCL_CLIENT); + fpin = key + cfs_size_round(sizeof(KEY_FID2PATH)); + fpout = val; - vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN); - if (vallen == NULL) { - CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer"); - RETURN(-EFAULT); - } + if (ptlrpc_req_need_swab(info->mti_pill->rc_req)) + lustre_swab_fid2path(fpin); - req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER, - *vallen); - rc = req_capsule_server_pack(info->mti_pill); - valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL); - if (valout == NULL) { - CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer"); - RETURN(-EFAULT); - } + memcpy(fpout, fpin, sizeof(*fpin)); + if (fpout->gf_pathlen != vallen - sizeof(*fpin)) + RETURN(-EINVAL); - if (KEY_IS(KEY_FID2PATH)) - rc = mdt_rpc_fid2path(info, key, valout, *vallen); - else - rc = -EINVAL; + rc = mdt_fid2path(info, fpout); + RETURN(rc); +} - lustre_msg_set_status(req->rq_repmsg, rc); +int mdt_get_info(struct tgt_session_info *tsi) +{ + char *key; + int keylen; + __u32 *vallen; + void *valout; + int rc; - RETURN(rc); + ENTRY; + + key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY); + if (key == NULL) { + CDEBUG(D_IOCTL, "No GETINFO key"); + RETURN(err_serious(-EFAULT)); + } + keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY, + RCL_CLIENT); + + vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN); + if (vallen == NULL) { + CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer"); + RETURN(err_serious(-EFAULT)); + } + + req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER, + *vallen); + rc = req_capsule_server_pack(tsi->tsi_pill); + if (rc) + RETURN(err_serious(rc)); + + valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL); + if (valout == NULL) { + CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer"); + RETURN(err_serious(-EFAULT)); + } + + if (KEY_IS(KEY_FID2PATH)) { + struct mdt_thread_info *info = tsi2mdt_info(tsi); + + rc = mdt_rpc_fid2path(info, key, valout, *vallen); + mdt_thread_info_fini(info); + } else { + rc = -EINVAL; + } + RETURN(rc); } /* Pass the ioc down */ static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt, - unsigned int cmd, int len, void *data) + unsigned int cmd, int len, void *data) { - struct lu_context ioctl_session; - struct md_device *next = mdt->mdt_child; - int rc; - ENTRY; + struct lu_context ioctl_session; + struct md_device *next = mdt->mdt_child; + int rc; + ENTRY; - rc = lu_context_init(&ioctl_session, LCT_SESSION); - if (rc) - RETURN(rc); - ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current(); - lu_context_enter(&ioctl_session); - env->le_ses = &ioctl_session; + rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION); + if (rc) + RETURN(rc); + ioctl_session.lc_thread = (struct ptlrpc_thread *)current; + lu_context_enter(&ioctl_session); + env->le_ses = &ioctl_session; - LASSERT(next->md_ops->mdo_iocontrol); - rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data); + LASSERT(next->md_ops->mdo_iocontrol); + rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data); - lu_context_exit(&ioctl_session); - lu_context_fini(&ioctl_session); - RETURN(rc); + lu_context_exit(&ioctl_session); + lu_context_fini(&ioctl_session); + RETURN(rc); } static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg) { - struct obd_ioctl_data *data = karg; - struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1; - __u64 version; - struct mdt_object *obj; - struct mdt_lock_handle *lh; - int rc; - ENTRY; + struct obd_ioctl_data *data = karg; + struct lu_fid *fid; + __u64 version; + struct mdt_object *obj; + struct mdt_lock_handle *lh; + int rc; + ENTRY; - CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid)); - if (!fid_is_sane(fid)) - RETURN(-EINVAL); + if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) || + data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version)) + RETURN(-EINVAL); + + fid = (struct lu_fid *)data->ioc_inlbuf1; + + if (!fid_is_sane(fid)) + RETURN(-EINVAL); + + CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid)); lh = &mti->mti_lh[MDT_LH_PARENT]; mdt_lock_reg_init(lh, LCK_CR); @@ -5416,24 +5484,23 @@ static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg) if (IS_ERR(obj)) RETURN(PTR_ERR(obj)); - rc = mdt_object_exists(obj); - if (rc < 0) { - rc = -EREMOTE; - /** - * before calling version get the correct MDS should be - * fid, this is error to find remote object here - */ - CERROR("nonlocal object "DFID"\n", PFID(fid)); - } else if (rc == 0) { - *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION; - rc = -ENOENT; - } else { - version = dt_version_get(mti->mti_env, mdt_obj2dt(obj)); - *(__u64 *)data->ioc_inlbuf2 = version; - rc = 0; - } - mdt_object_unlock_put(mti, obj, lh, 1); - RETURN(rc); + if (mdt_object_remote(obj)) { + rc = -EREMOTE; + /** + * before calling version get the correct MDS should be + * fid, this is error to find remote object here + */ + CERROR("nonlocal object "DFID"\n", PFID(fid)); + } else if (!mdt_object_exists(obj)) { + *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION; + rc = -ENOENT; + } else { + version = dt_version_get(mti->mti_env, mdt_obj2dt(obj)); + *(__u64 *)data->ioc_inlbuf2 = version; + rc = 0; + } + mdt_object_unlock_put(mti, obj, lh, 1); + RETURN(rc); } /* ioctls on obd dev */ @@ -5459,29 +5526,35 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, case OBD_IOC_SET_READONLY: rc = dt->dd_ops->dt_ro(&env, dt); break; - case OBD_IOC_ABORT_RECOVERY: - CERROR("Aborting recovery for device %s\n", obd->obd_name); - target_stop_recovery_thread(obd); - rc = 0; - break; + case OBD_IOC_ABORT_RECOVERY: + CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt)); + target_stop_recovery_thread(obd); + rc = 0; + break; case OBD_IOC_CHANGELOG_REG: case OBD_IOC_CHANGELOG_DEREG: case OBD_IOC_CHANGELOG_CLEAR: rc = mdt_ioc_child(&env, mdt, cmd, len, karg); break; - case OBD_IOC_START_LFSCK: - case OBD_IOC_STOP_LFSCK: { + case OBD_IOC_START_LFSCK: { struct md_device *next = mdt->mdt_child; struct obd_ioctl_data *data = karg; + struct lfsck_start_param lsp; if (unlikely(data == NULL)) { rc = -EINVAL; break; } - rc = next->md_ops->mdo_iocontrol(&env, next, cmd, - data->ioc_inllen1, - data->ioc_inlbuf1); + lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1); + lsp.lsp_namespace = mdt->mdt_namespace; + rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp); + break; + } + case OBD_IOC_STOP_LFSCK: { + struct md_device *next = mdt->mdt_child; + + rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, NULL); break; } case OBD_IOC_GET_OBJ_VERSION: { @@ -5495,11 +5568,14 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, rc = mdt_ioc_version_get(mti, karg); break; } - default: - CERROR("Not supported cmd = %d for device %s\n", - cmd, obd->obd_name); - rc = -EOPNOTSUPP; - } + case OBD_IOC_CATLOGLIST: + rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg); + break; + default: + rc = -EOPNOTSUPP; + CERROR("%s: Not supported cmd = %d, rc = %d\n", + mdt_obd_name(mdt), cmd, rc); + } lu_env_fini(&env); RETURN(rc); @@ -5553,12 +5629,13 @@ static struct lu_device* mdt_device_fini(const struct lu_env *env, static struct lu_device *mdt_device_free(const struct lu_env *env, struct lu_device *d) { - struct mdt_device *m = mdt_dev(d); - ENTRY; + struct mdt_device *m = mdt_dev(d); + ENTRY; - md_device_fini(&m->mdt_md_dev); - OBD_FREE_PTR(m); - RETURN(NULL); + lu_device_fini(&m->mdt_lu_dev); + OBD_FREE_PTR(m); + + RETURN(NULL); } static struct lu_device *mdt_device_alloc(const struct lu_env *env, @@ -5572,7 +5649,7 @@ static struct lu_device *mdt_device_alloc(const struct lu_env *env, if (m != NULL) { int rc; - l = &m->mdt_md_dev.md_lu_dev; + l = &m->mdt_lu_dev; rc = mdt_init0(env, m, t, cfg); if (rc != 0) { mdt_device_free(env, l); @@ -5667,6 +5744,10 @@ static int __init mdt_mod_init(void) struct lprocfs_static_vars lvars; int rc; + CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") == + FID_NOBRACE_LEN + 1); + CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") == + FID_LEN + 1); rc = lu_kmem_init(mdt_caches); if (rc) return rc; @@ -5676,9 +5757,11 @@ static int __init mdt_mod_init(void) GOTO(lu_fini, rc); lprocfs_mdt_init_vars(&lvars); - rc = class_register_type(&mdt_obd_device_ops, NULL, - lvars.module_vars, LUSTRE_MDT_NAME, - &mdt_device_type); + rc = class_register_type(&mdt_obd_device_ops, NULL, NULL, +#ifndef HAVE_ONLY_PROCFS_SEQ + lvars.module_vars, +#endif + LUSTRE_MDT_NAME, &mdt_device_type); if (rc) GOTO(mds_fini, rc); lu_fini: