X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_handler.c;h=23a213649e327eaa133e6641095d6e066c848b8c;hp=1a0d670fbfb056e333a412b2fd43932c75b8f3eb;hb=f0c1b06f2418d2016e23eb6e8277be85ca13fc53;hpb=deb7a5c9f42c1fb7a6f225e2eb09b64b79efc64c diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c index 1a0d670..23a2136 100644 --- a/lustre/mdt/mdt_handler.c +++ b/lustre/mdt/mdt_handler.c @@ -62,13 +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, @@ -98,7 +98,13 @@ static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags); 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[] = { { @@ -107,6 +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 } }; @@ -154,6 +170,13 @@ void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm, if (name != NULL && (name[0] != '\0')) { LASSERT(namelen > 0); lh->mlh_pdo_hash = full_name_hash(name, 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)) + lh->mlh_pdo_hash = 1; } else { LASSERT(namelen == 0); lh->mlh_pdo_hash = 0ull; @@ -235,32 +258,34 @@ 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 mdt_thread_info *info = tsi2mdt_info(tsi); + struct mdt_device *mdt = info->mti_mdt; struct mdt_body *repbody; - int rc; + int rc; + ENTRY; - 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_GETSTATUS_PACK)) - RETURN(err_serious(-ENOMEM)); + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK)) + GOTO(out, rc = err_serious(-ENOMEM)); repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); repbody->fid1 = mdt->mdt_md_root_fid; - repbody->valid |= OBD_MD_FLID; + repbody->valid |= OBD_MD_FLID; - if (mdt->mdt_opts.mo_mds_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; + 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); @@ -271,13 +296,16 @@ 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; @@ -285,23 +313,23 @@ 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, @@ -309,7 +337,7 @@ 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(); @@ -323,7 +351,8 @@ 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); } @@ -388,7 +417,18 @@ 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; @@ -400,7 +440,7 @@ 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); } @@ -719,18 +759,26 @@ 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; @@ -744,8 +792,9 @@ 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); } } @@ -770,18 +819,20 @@ 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; @@ -790,8 +841,11 @@ 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; @@ -845,9 +899,12 @@ 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; @@ -858,7 +915,7 @@ 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_mdt->mdt_lut.lut_mds_capa && exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) { struct lustre_capa *capa; @@ -890,7 +947,7 @@ 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 || + if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa || !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA)) RETURN(0); @@ -910,8 +967,9 @@ 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; @@ -932,19 +990,19 @@ 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); @@ -974,11 +1032,14 @@ 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; @@ -1001,12 +1062,14 @@ int mdt_is_subdir(struct mdt_thread_info *info) if (rc == 0 || rc == -EREMOTE) repbody->valid |= OBD_MD_FLID; + mdt_thread_info_fini(info); RETURN(rc); } -int mdt_swap_layouts(struct mdt_thread_info *info) +int mdt_swap_layouts(struct tgt_session_info *tsi) { - struct ptlrpc_request *req = mdt_info_req(info); + 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; @@ -1024,6 +1087,7 @@ int mdt_swap_layouts(struct mdt_thread_info *info) 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, @@ -1071,13 +1135,13 @@ int mdt_swap_layouts(struct mdt_thread_info *info) lh2 = &info->mti_lh[MDT_LH_OLD]; mdt_lock_reg_init(lh2, LCK_EX); - rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT, - MDT_LOCAL_LOCK); + 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, - MDT_LOCAL_LOCK); + rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT | + MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK); if (rc < 0) GOTO(unlock1, rc); @@ -1091,6 +1155,7 @@ unlock1: put: mdt_object_put(info->mti_env, o); out: + mdt_thread_info_fini(info); RETURN(rc); } @@ -1204,9 +1269,9 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, } mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD); - if (unlikely(!mdt_object_exists(parent))) { + if (unlikely(!mdt_object_exists(parent)) && 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) { @@ -1266,7 +1331,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, if (lname) { /* step 1: lock parent only if parent is a directory */ - if (S_ISDIR(lu_object_attr(&parent->mot_obj.mo_lu))) { + if (S_ISDIR(lu_object_attr(&parent->mot_obj))) { lhp = &info->mti_lh[MDT_LH_PARENT]; mdt_lock_pdo_init(lhp, LCK_PR, name, namelen); rc = mdt_object_lock(info, parent, lhp, @@ -1299,35 +1364,33 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, */ 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 { + 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: @@ -1337,7 +1400,7 @@ relock: if (!mdt_object_exists(child)) { LU_OBJECT_DEBUG(D_INODE, info->mti_env, - &child->mot_obj.mo_lu, + &child->mot_obj, "Object doesn't exist!\n"); GOTO(out_child, rc = -ENOENT); } @@ -1352,11 +1415,11 @@ 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 + @@ -1369,7 +1432,7 @@ 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)) && + S_ISREG(lu_object_attr(&child->mot_obj)) && ldlm_rep != NULL) { /* try to grant layout lock for regular file. */ try_layout = true; @@ -1409,17 +1472,15 @@ 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))); + } 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); } @@ -1436,8 +1497,9 @@ 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; @@ -1469,47 +1531,42 @@ 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) *exp_connect_flags_ptr(req->rq_export) |= @@ -1518,130 +1575,44 @@ int mdt_set_info(struct mdt_thread_info *info) *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_connect_flags_ptr(exp) = 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, 1, BULK_PUT_SOURCE, - MDS_BULK_PORTAL); - if (desc == NULL) - RETURN(-ENOMEM); - - if (!(exp_connect_flags(exp) & 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)); /* @@ -1657,53 +1628,53 @@ int mdt_readpage(struct mdt_thread_info *info) } rdpg->rp_attrs = reqbody->mode; - if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_64BITHASH) + 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(info->mti_exp)); - rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1) >> - CFS_PAGE_SHIFT; + 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) { @@ -1764,23 +1735,24 @@ out_shrink: return rc; } -static long mdt_reint_opcode(struct mdt_thread_info *info, +static long mdt_reint_opcode(struct ptlrpc_request *req, const struct req_format **fmt) { - struct mdt_rec_reint *rec; - long opc; + struct mdt_device *mdt; + struct mdt_rec_reint *rec; + long opc; - rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT); + rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT); if (rec != NULL) { opc = rec->rr_opcode; - DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc); + DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc); if (opc < REINT_MAX && fmt[opc] != NULL) - req_capsule_extend(info->mti_pill, fmt[opc]); + req_capsule_extend(&req->rq_pill, fmt[opc]); else { - CERROR("%s: Unsupported opcode '%ld' from client '%s': " - "rc = %d\n", mdt_obd_name(info->mti_mdt), opc, - info->mti_mdt->mdt_ldlm_client->cli_name, - -EFAULT); + 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 { @@ -1789,11 +1761,10 @@ static long mdt_reint_opcode(struct mdt_thread_info *info, 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, @@ -1802,24 +1773,26 @@ int mdt_reint(struct mdt_thread_info *info) [REINT_RENAME] = &RQF_MDS_REINT_RENAME, [REINT_OPEN] = &RQF_MDS_REINT_OPEN, [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR, - [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK + [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK }; - ENTRY; + ENTRY; - 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; - } + 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; + } - info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP; - RETURN(rc); + tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP; + RETURN(rc); } /* this should sync the whole device */ @@ -1851,93 +1824,60 @@ 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); @@ -1978,7 +1918,7 @@ 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)) @@ -2013,13 +1953,13 @@ 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; @@ -2035,119 +1975,6 @@ 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, - exp_max_brw_size(info->mti_exp)); - 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 @@ -2192,150 +2019,45 @@ 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) @@ -2346,7 +2068,7 @@ 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 @@ -2355,20 +2077,21 @@ 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); } /** @@ -2508,7 +2231,7 @@ int mdt_remote_object_lock(struct mdt_thread_info *mti, LASSERT(mdt_object_remote(o)); - LASSERT((ibits & MDS_INODELOCK_UPDATE)); + LASSERT(ibits & MDS_INODELOCK_UPDATE); memset(einfo, 0, sizeof(*einfo)); einfo->ei_type = LDLM_IBITS; @@ -2563,8 +2286,8 @@ static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o, 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); } } @@ -2753,40 +2476,6 @@ 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: * @@ -2881,150 +2570,13 @@ 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 && - exp_connect_flags(req->rq_export) & 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); - } - - /* 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); + struct md_device *next = m->mdt_child; - 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) @@ -3049,12 +2601,11 @@ 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 */ @@ -3067,10 +2618,9 @@ 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); + info->mti_mos = NULL; memset(&info->mti_attr, 0, sizeof(info->mti_attr)); info->mti_big_buf = LU_BUF_NULL; @@ -3087,11 +2637,10 @@ static void mdt_thread_info_init(struct ptlrpc_request *req, info->mti_spec.sp_rm_entry = 0; } -static void mdt_thread_info_fini(struct mdt_thread_info *info) +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; @@ -3100,310 +2649,62 @@ static void mdt_thread_info_fini(struct mdt_thread_info *info) 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); } -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 UPDATE_OBJ: - 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) +struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi) { - 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); - } - - /* 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); - } - } + struct mdt_thread_info *mti; + struct lustre_capa *lc; - /* 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. */ + mti = mdt_th_info(tsi->tsi_env); + LASSERT(mti != NULL); - obd = req->rq_export->exp_obd; + mdt_thread_info_init(tgt_ses_req(tsi), mti); + if (tsi->tsi_corpus != NULL) { + struct req_capsule *pill = tsi->tsi_pill; - /* 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); -} + mti->mti_object = mdt_obj(tsi->tsi_corpus); + lu_object_get(tsi->tsi_corpus); -static int mdt_msg_check_version(struct lustre_msg *msg) -{ - int 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; - 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 MDS_SWAP_LAYOUTS: - 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; + return mti; } -static int mdt_handle0(struct ptlrpc_request *req, - struct mdt_thread_info *info, - struct mdt_opc_slice *supported) +int mdt_tgt_connect(struct tgt_session_info *tsi) { - 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); -} - -/* - * 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 lu_env *env; - struct mdt_thread_info *info; - int rc; - ENTRY; - - 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); - - mdt_thread_info_init(req, info); - - rc = mdt_handle0(req, info, supported); - - mdt_thread_info_fini(info); - RETURN(rc); -} + struct ptlrpc_request *req = tgt_ses_req(tsi); + int rc; -/* - * 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 rc; - ENTRY; + ENTRY; - 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; - } + rc = tgt_connect(tsi); + if (rc != 0) + RETURN(rc); - 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 { @@ -3425,6 +2726,12 @@ 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 **, @@ -3452,7 +2759,14 @@ 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 }, @@ -3489,9 +2803,9 @@ 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, @@ -3642,6 +2956,44 @@ 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, @@ -3691,8 +3043,8 @@ 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; @@ -3720,15 +3072,38 @@ 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)); @@ -3743,7 +3118,7 @@ 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); } @@ -3764,7 +3139,7 @@ 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); @@ -3924,30 +3299,42 @@ static int mdt_intent_opc(long itopc, struct mdt_thread_info *info, 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); @@ -3979,7 +3366,8 @@ 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, @@ -4123,7 +3511,7 @@ static int mdt_fld_init(const struct lu_env *env, static void mdt_stack_pre_fini(const struct lu_env *env, struct mdt_device *m, struct lu_device *top) { - struct obd_device *obd = mdt2obd_dev(m); + struct obd_device *obd; struct lustre_cfg_bufs *bufs; struct lustre_cfg *lcfg; struct mdt_thread_info *info; @@ -4145,7 +3533,7 @@ static void mdt_stack_pre_fini(const struct lu_env *env, * 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, obd->obd_name); + 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) { @@ -4158,44 +3546,44 @@ static void mdt_stack_pre_fini(const struct lu_env *env, } 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; @@ -4219,7 +3607,7 @@ 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); } @@ -4229,7 +3617,7 @@ 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); } @@ -4345,9 +3733,9 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, 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); @@ -4364,7 +3752,7 @@ 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: @@ -4475,7 +3863,7 @@ 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); @@ -4530,10 +3918,128 @@ 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; @@ -4541,8 +4047,14 @@ 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(); @@ -4569,50 +4081,49 @@ 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); + 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)); + /* + * Finish the stack + */ + mdt_stack_fini(env, m, md2lu_dev(m->mdt_child)); - LASSERT(cfs_atomic_read(&d->ld_ref) == 0); + LASSERT(cfs_atomic_read(&d->ld_ref) == 0); - server_put_mount(mdt2obd_dev(m)->obd_name, NULL); + server_put_mount(mdt_obd_name(m), NULL); - EXIT; + 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; + 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; - } + 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", + mdt_obd_name(m), rc); + return rc; + } - sptlrpc_target_update_exp_flavor(obd, &tmp_rset); + sptlrpc_target_update_exp_flavor(obd, &tmp_rset); - 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); + write_lock(&m->mdt_lut.lut_sptlrpc_lock); + sptlrpc_rule_set_free(&m->mdt_lut.lut_sptlrpc_rset); + m->mdt_lut.lut_sptlrpc_rset = tmp_rset; + write_unlock(&m->mdt_lut.lut_sptlrpc_lock); - return 0; + return 0; } int mdt_postrecov(const struct lu_env *, struct mdt_device *); @@ -4631,11 +4142,11 @@ 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 @@ -4659,6 +4170,11 @@ 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); @@ -4669,15 +4185,9 @@ 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; @@ -4692,20 +4202,21 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, 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; @@ -4720,28 +4231,24 @@ 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, mdt_obd_name(m), 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, @@ -4758,13 +4265,26 @@ 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); + + rc = mdt_fs_setup(env, m, obd, lsi); + if (rc) + GOTO(err_tgt, rc); mdt_adapt_sptlrpc_conf(obd, 1); @@ -4772,30 +4292,35 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, 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) { @@ -4811,7 +4336,7 @@ 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 @@ -4826,34 +4351,34 @@ 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); + return(rc); } /* For interoperability, the left element is old parameter, the right one @@ -4918,9 +4443,16 @@ 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); @@ -4943,12 +4475,12 @@ 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); @@ -4956,9 +4488,10 @@ 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, @@ -4993,11 +4526,14 @@ 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, @@ -5017,6 +4553,7 @@ 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; @@ -5031,9 +4568,15 @@ static int mdt_prepare(const struct lu_env *env, if (rc) RETURN(rc); + 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, NULL); + 0, &lsp); if (rc != 0) { CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n", mdt_obd_name(mdt), rc); @@ -5046,8 +4589,9 @@ static int mdt_prepare(const struct lu_env *env, 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); @@ -5148,7 +4692,7 @@ 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); @@ -5168,12 +4712,10 @@ static int mdt_connect_internal(struct obd_export *exp, } data->ocd_version = LUSTRE_VERSION_CODE; - exp->exp_connect_data = *data; - 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; } @@ -5182,7 +4724,7 @@ 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; } @@ -5196,51 +4738,9 @@ static int mdt_connect_internal(struct obd_export *exp, } } - 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; - } - - 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 */ @@ -5250,11 +4750,9 @@ 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; @@ -5262,16 +4760,14 @@ 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); if (rc) @@ -5286,26 +4782,17 @@ 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; @@ -5322,23 +4809,12 @@ 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); @@ -5346,6 +4822,31 @@ 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_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; @@ -5390,6 +4891,24 @@ 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; @@ -5533,9 +5052,7 @@ static int mdt_links_read(struct mdt_thread_info *info, if (rc < 0) return rc; - linkea_init(ldata); - - return 0; + return linkea_init(ldata); } static int mdt_path_current(struct mdt_thread_info *info, @@ -5587,15 +5104,15 @@ static int mdt_path_current(struct mdt_thread_info *info, rc = mdt_links_read(info, mdt_obj, &ldata); mdt_object_put(info->mti_env, mdt_obj); if (rc != 0) - GOTO(out, rc = PTR_ERR(buf)); + 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)) { + 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 *) @@ -5680,7 +5197,6 @@ static int mdt_fid2path(struct mdt_thread_info *info, struct getinfo_fid2path *fp) { struct mdt_device *mdt = info->mti_mdt; - struct obd_device *obd = mdt2obd_dev(mdt); struct mdt_object *obj; int rc; ENTRY; @@ -5693,15 +5209,15 @@ static int mdt_fid2path(struct mdt_thread_info *info, 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(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); } @@ -5715,7 +5231,7 @@ static int mdt_fid2path(struct mdt_thread_info *info, 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); } @@ -5750,71 +5266,75 @@ static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, RETURN(rc); } -int mdt_get_info(struct mdt_thread_info *info) +int mdt_get_info(struct tgt_session_info *tsi) { - struct ptlrpc_request *req = mdt_info_req(info); - char *key; - int keylen; - __u32 *vallen; - void *valout; - int rc; - ENTRY; + char *key; + int keylen; + __u32 *vallen; + void *valout; + int rc; - 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); + ENTRY; - 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); - } + 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); - 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); - } + 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)); + } - if (KEY_IS(KEY_FID2PATH)) - rc = mdt_rpc_fid2path(info, key, valout, *vallen); - else - rc = -EINVAL; + 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)); - lustre_msg_set_status(req->rq_repmsg, 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)); + } - RETURN(rc); + 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_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) @@ -5887,29 +5407,35 @@ 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: { @@ -5923,11 +5449,11 @@ 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; - } + default: + rc = -EOPNOTSUPP; + CERROR("%s: Not supported cmd = %d, rc = %d\n", + mdt_obd_name(mdt), cmd, rc); + } lu_env_fini(&env); RETURN(rc); @@ -5981,12 +5507,13 @@ 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, @@ -6000,7 +5527,7 @@ 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); @@ -6095,6 +5622,10 @@ 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;