X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_handler.c;h=d4624966aed604e6b1a3901c5e383c4846c43a22;hp=4f5ec9eb1c84278d84e6a454ecd76e1f66df13e9;hb=b4d4a448b2ed93a8bc73db81bbf9c8ba08250e3a;hpb=825570b0b03e0fd70aba5400dedca4e6d94d5141 diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c index 4f5ec9e..d462496 100644 --- a/lustre/mdt/mdt_handler.c +++ b/lustre/mdt/mdt_handler.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2010, 2013, Intel Corporation. + * Copyright (c) 2010, 2015, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -71,6 +71,10 @@ #include #include +static unsigned int max_mod_rpcs_per_client = 8; +CFS_MODULE_PARM(max_mod_rpcs_per_client, "i", uint, 0644, + "maximum number of modify RPCs in flight allowed per client"); + mdl_mode_t mdt_mdl_lock_modes[] = { [LCK_MINMODE] = MDL_MINMODE, [LCK_EX] = MDL_EX, @@ -82,15 +86,15 @@ mdl_mode_t mdt_mdl_lock_modes[] = { [LCK_GROUP] = MDL_GROUP }; -ldlm_mode_t mdt_dlm_lock_modes[] = { - [MDL_MINMODE] = LCK_MINMODE, - [MDL_EX] = LCK_EX, - [MDL_PW] = LCK_PW, - [MDL_PR] = LCK_PR, - [MDL_CW] = LCK_CW, - [MDL_CR] = LCK_CR, - [MDL_NL] = LCK_NL, - [MDL_GROUP] = LCK_GROUP +enum ldlm_mode mdt_dlm_lock_modes[] = { + [MDL_MINMODE] = LCK_MINMODE, + [MDL_EX] = LCK_EX, + [MDL_PW] = LCK_PW, + [MDL_PR] = LCK_PR, + [MDL_CW] = LCK_CW, + [MDL_CR] = LCK_CR, + [MDL_NL] = LCK_NL, + [MDL_GROUP] = LCK_GROUP }; static struct mdt_device *mdt_dev(struct lu_device *d); @@ -128,44 +132,44 @@ static struct lu_kmem_descr mdt_caches[] = { } }; -int mdt_get_disposition(struct ldlm_reply *rep, int flag) +__u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag) { - if (!rep) - return 0; - return (rep->lock_policy_res1 & flag); + if (!rep) + return 0; + return rep->lock_policy_res1 & op_flag; } void mdt_clear_disposition(struct mdt_thread_info *info, - struct ldlm_reply *rep, int flag) + struct ldlm_reply *rep, __u64 op_flag) { if (info) { - info->mti_opdata &= ~flag; - tgt_opdata_clear(info->mti_env, flag); + info->mti_opdata &= ~op_flag; + tgt_opdata_clear(info->mti_env, op_flag); } if (rep) - rep->lock_policy_res1 &= ~flag; + rep->lock_policy_res1 &= ~op_flag; } void mdt_set_disposition(struct mdt_thread_info *info, - struct ldlm_reply *rep, int flag) + struct ldlm_reply *rep, __u64 op_flag) { if (info) { - info->mti_opdata |= flag; - tgt_opdata_set(info->mti_env, flag); + info->mti_opdata |= op_flag; + tgt_opdata_set(info->mti_env, op_flag); } if (rep) - rep->lock_policy_res1 |= flag; + rep->lock_policy_res1 |= op_flag; } -void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm) +void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm) { - lh->mlh_pdo_hash = 0; - lh->mlh_reg_mode = lm; + lh->mlh_pdo_hash = 0; + lh->mlh_reg_mode = lm; lh->mlh_rreg_mode = lm; - lh->mlh_type = MDT_REG_LOCK; + lh->mlh_type = MDT_REG_LOCK; } -void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode, +void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode, const struct lu_name *lname) { lh->mlh_reg_mode = lock_mode; @@ -284,24 +288,6 @@ static int mdt_getstatus(struct tgt_session_info *tsi) repbody->mbo_fid1 = mdt->mdt_md_root_fid; repbody->mbo_valid |= OBD_MD_FLID; - if (tsi->tsi_tgt->lut_mds_capa && - exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) { - struct mdt_object *root; - struct lustre_capa *capa; - - root = mdt_object_find(info->mti_env, mdt, &repbody->mbo_fid1); - if (IS_ERR(root)) - GOTO(out, rc = PTR_ERR(root)); - - capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1); - LASSERT(capa); - capa->lc_opc = CAPA_OPC_MDS_DEFAULT; - rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa, - 0); - mdt_object_put(info->mti_env, root); - if (rc == 0) - repbody->mbo_valid |= OBD_MD_FLMDSCAPA; - } EXIT; out: mdt_thread_info_fini(info); @@ -362,33 +348,6 @@ out: RETURN(rc); } -/** - * Pack SOM attributes into the reply. - * Call under a DLM UPDATE lock. - */ -static void mdt_pack_size2body(struct mdt_thread_info *info, - struct mdt_object *mo) -{ - struct mdt_body *b; - struct md_attr *ma = &info->mti_attr; - - LASSERT(ma->ma_attr.la_valid & LA_MODE); - b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); - - /* Check if Size-on-MDS is supported, if this is a regular file, - * if SOM is enabled on the object and if SOM cache exists and valid. - * Otherwise do not pack Size-on-MDS attributes to the reply. */ - if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) || - !S_ISREG(ma->ma_attr.la_mode) || - !mdt_object_is_som_enabled(mo) || - !(ma->ma_valid & MA_SOM)) - return; - - b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; - b->mbo_size = ma->ma_som->msd_size; - b->mbo_blocks = ma->ma_som->msd_blocks; -} - #ifdef CONFIG_FS_POSIX_ACL /* * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap. @@ -454,27 +413,60 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, LASSERT(ma->ma_valid & MA_INODE); - b->mbo_atime = attr->la_atime; - b->mbo_mtime = attr->la_mtime; - b->mbo_ctime = attr->la_ctime; - b->mbo_mode = attr->la_mode; - b->mbo_size = attr->la_size; - b->mbo_blocks = attr->la_blocks; - b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID, - NODEMAP_FS_TO_CLIENT, - attr->la_uid); - b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID, - NODEMAP_FS_TO_CLIENT, - attr->la_gid); - b->mbo_flags = attr->la_flags; - b->mbo_nlink = attr->la_nlink; - b->mbo_rdev = attr->la_rdev; - - /* XXX should pack the reply body according to lu_valid */ - b->mbo_valid |= OBD_MD_FLCTIME | OBD_MD_FLUID | - OBD_MD_FLGID | OBD_MD_FLTYPE | - OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS | - OBD_MD_FLATIME | OBD_MD_FLMTIME ; + if (attr->la_valid & LA_ATIME) { + b->mbo_atime = attr->la_atime; + b->mbo_valid |= OBD_MD_FLATIME; + } + if (attr->la_valid & LA_MTIME) { + b->mbo_mtime = attr->la_mtime; + b->mbo_valid |= OBD_MD_FLMTIME; + } + if (attr->la_valid & LA_CTIME) { + b->mbo_ctime = attr->la_ctime; + b->mbo_valid |= OBD_MD_FLCTIME; + } + if (attr->la_valid & LA_FLAGS) { + b->mbo_flags = attr->la_flags; + b->mbo_valid |= OBD_MD_FLFLAGS; + } + if (attr->la_valid & LA_NLINK) { + b->mbo_nlink = attr->la_nlink; + b->mbo_valid |= OBD_MD_FLNLINK; + } + if (attr->la_valid & LA_UID) { + b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID, + NODEMAP_FS_TO_CLIENT, + attr->la_uid); + b->mbo_valid |= OBD_MD_FLUID; + } + if (attr->la_valid & LA_GID) { + b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID, + NODEMAP_FS_TO_CLIENT, + attr->la_gid); + b->mbo_valid |= OBD_MD_FLGID; + } + b->mbo_mode = attr->la_mode; + if (attr->la_valid & LA_MODE) + b->mbo_valid |= OBD_MD_FLMODE; + if (attr->la_valid & LA_TYPE) + b->mbo_valid |= OBD_MD_FLTYPE; + + if (fid != NULL) { + b->mbo_fid1 = *fid; + b->mbo_valid |= OBD_MD_FLID; + CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n", + PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid); + } + + if (info != NULL) + mdt_body_reverse_idmap(info, b); + + if (!(attr->la_valid & LA_TYPE)) + return; + + b->mbo_rdev = attr->la_rdev; + b->mbo_size = attr->la_size; + b->mbo_blocks = attr->la_blocks; if (!S_ISREG(attr->la_mode)) { b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV; @@ -499,17 +491,6 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; } - if (fid) { - b->mbo_fid1 = *fid; - b->mbo_valid |= OBD_MD_FLID; - CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n", - PFID(fid), b->mbo_nlink, - b->mbo_mode, b->mbo_size); - } - - if (info) - mdt_body_reverse_idmap(info, b); - if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE)) CDEBUG(D_VFSTRACE, DFID": returning size %llu\n", PFID(fid), (unsigned long long)b->mbo_size); @@ -547,7 +528,8 @@ void mdt_client_compatibility(struct mdt_thread_info *info) EXIT; } -int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o) +static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, + struct mdt_object *o) { const struct lu_env *env = info->mti_env; int rc, rc2; @@ -791,19 +773,6 @@ int mdt_attr_get_complex(struct mdt_thread_info *info, GOTO(out, rc); } - if (need & MA_SOM && S_ISREG(mode)) { - buf->lb_buf = info->mti_xattr_buf; - buf->lb_len = sizeof(info->mti_xattr_buf); - CLASSERT(sizeof(struct som_attrs) <= - sizeof(info->mti_xattr_buf)); - rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM); - rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som); - if (rc2 == 0) - ma->ma_valid |= MA_SOM; - else if (rc2 < 0 && rc2 != -ENODATA) - GOTO(out, rc = rc2); - } - if (need & MA_HSM && S_ISREG(mode)) { buf->lb_buf = info->mti_xattr_buf; buf->lb_len = sizeof(info->mti_xattr_buf); @@ -918,8 +887,6 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, ma->ma_need |= MA_LOV_DEF; } ma->ma_need |= ma_need; - if (ma->ma_need & MA_SOM) - ma->ma_som = &info->mti_u.som.data; rc = mdt_attr_get_complex(info, o, ma); if (unlikely(rc)) { @@ -930,10 +897,11 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, } /* 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->mbo_t_state = MS_RESTORE; + if (ma->ma_valid & MA_HSM) { repbody->mbo_valid |= OBD_MD_TSTATE; + if ((ma->ma_hsm.mh_flags & HS_RELEASED) && + mdt_hsm_restore_is_running(info, mdt_object_fid(o))) + repbody->mbo_t_state = MS_RESTORE; } is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid); @@ -1037,12 +1005,10 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, } if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) { - repbody->mbo_max_cookiesize = 0; repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize; repbody->mbo_valid |= OBD_MD_FLMODEASIZE; - CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & " - "MAX_COOKIE to : %d:%d\n", repbody->mbo_max_mdsize, - repbody->mbo_max_cookiesize); + CDEBUG(D_INODE, "changing the max MD size to %u\n", + repbody->mbo_max_mdsize); } if (exp_connect_rmtclient(info->mti_exp) && @@ -1066,20 +1032,6 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, rc = mdt_pack_acl2body(info, repbody, o, nodemap); #endif - if (reqbody->mbo_valid & OBD_MD_FLMDSCAPA && - info->mti_mdt->mdt_lut.lut_mds_capa && - exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) { - struct lustre_capa *capa; - - capa = req_capsule_server_get(pill, &RMF_CAPA1); - LASSERT(capa); - capa->lc_opc = CAPA_OPC_MDS_DEFAULT; - rc = mo_capa_get(env, next, capa, 0); - if (rc) - RETURN(rc); - repbody->mbo_valid |= OBD_MD_FLMDSCAPA; - } - out: if (rc == 0) mdt_counter_incr(req, LPROC_MDT_GETATTR); @@ -1087,38 +1039,6 @@ out: RETURN(rc); } -static int mdt_renew_capa(struct mdt_thread_info *info) -{ - struct mdt_object *obj = info->mti_object; - struct mdt_body *body; - struct lustre_capa *capa, *c; - int rc; - ENTRY; - - /* if object doesn't exist, or server has disabled capability, - * return directly, client will find body->valid OBD_MD_FLOSSCAPA - * flag not set. - */ - if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa || - !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA)) - RETURN(0); - - body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); - LASSERT(body != NULL); - - c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1); - LASSERT(c); - - capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2); - LASSERT(capa); - - *capa = *c; - rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1); - if (rc == 0) - body->mbo_valid |= OBD_MD_FLOSSCAPA; - RETURN(rc); -} - static int mdt_getattr(struct tgt_session_info *tsi) { struct mdt_thread_info *info = tsi2mdt_info(tsi); @@ -1131,15 +1051,6 @@ static int mdt_getattr(struct tgt_session_info *tsi) reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY); LASSERT(reqbody); - - if (reqbody->mbo_valid & OBD_MD_FLOSSCAPA) { - rc = req_capsule_server_pack(pill); - if (unlikely(rc)) - RETURN(err_serious(rc)); - rc = mdt_renew_capa(info); - GOTO(out_shrink, rc); - } - LASSERT(obj != NULL); LASSERT(lu_object_assert_exists(&obj->mot_obj)); @@ -1192,11 +1103,6 @@ static int mdt_getattr(struct tgt_session_info *tsi) info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF); - /* - * Don't check capability at all, because rename might getattr for - * remote obj, and at that time no capability is available. - */ - mdt_set_capainfo(info, 1, &reqbody->mbo_fid1, BYPASS_CAPA); rc = mdt_getattr_internal(info, obj, 0); if (reqbody->mbo_valid & OBD_MD_FLRMTPERM) mdt_exit_ucred(info); @@ -1211,6 +1117,32 @@ out: return rc; } +/** + * Exchange MOF_LOV_CREATED flags between two objects after a + * layout swap. No assumption is made on whether o1 or o2 have + * created objects or not. + * + * \param[in,out] o1 First swap layout object + * \param[in,out] o2 Second swap layout object + */ +static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2) +{ + __u64 o1_flags; + + mutex_lock(&o1->mot_lov_mutex); + mutex_lock(&o2->mot_lov_mutex); + + o1_flags = o1->mot_flags; + o1->mot_flags = (o1->mot_flags & ~MOF_LOV_CREATED) | + (o2->mot_flags & MOF_LOV_CREATED); + + o2->mot_flags = (o2->mot_flags & ~MOF_LOV_CREATED) | + (o1_flags & MOF_LOV_CREATED); + + mutex_unlock(&o2->mot_lov_mutex); + mutex_unlock(&o1->mot_lov_mutex); +} + static int mdt_swap_layouts(struct tgt_session_info *tsi) { struct mdt_thread_info *info; @@ -1237,16 +1169,6 @@ static int mdt_swap_layouts(struct tgt_session_info *tsi) if (info->mti_dlm_req != NULL) ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP); - if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT)) - mdt_set_capainfo(info, 0, &info->mti_body->mbo_fid1, - req_capsule_client_get(info->mti_pill, - &RMF_CAPA1)); - - if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT)) - mdt_set_capainfo(info, 1, &info->mti_body->mbo_fid2, - req_capsule_client_get(info->mti_pill, - &RMF_CAPA2)); - o1 = info->mti_object; o = o2 = mdt_object_find(info->mti_env, info->mti_mdt, &info->mti_body->mbo_fid2); @@ -1285,18 +1207,22 @@ static int mdt_swap_layouts(struct tgt_session_info *tsi) mdt_lock_reg_init(lh2, LCK_EX); rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT | - MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK); + MDS_INODELOCK_XATTR); if (rc < 0) GOTO(put, rc); rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT | - MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK); + MDS_INODELOCK_XATTR); if (rc < 0) GOTO(unlock1, rc); rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1), mdt_object_child(o2), msl->msl_flags); - GOTO(unlock2, rc); + if (rc < 0) + GOTO(unlock2, rc); + + mdt_swap_lov_flag(o1, o2); + unlock2: mdt_object_unlock(info, o2, lh2, rc); unlock1: @@ -1338,7 +1264,7 @@ static int mdt_raw_lookup(struct mdt_thread_info *info, } /* - * UPDATE lock should be taken against parent, and be release before exit; + * UPDATE lock should be taken against parent, and be released before exit; * child_bits lock should be taken against child, and be returned back: * (1)normal request should release the child lock; * (2)intent request will grant the lock to client. @@ -1393,8 +1319,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, MDS_INODELOCK_LAYOUT); child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE; - rc = mdt_object_lock(info, child, lhc, child_bits, - MDT_LOCAL_LOCK); + rc = mdt_object_lock(info, child, lhc, child_bits); if (rc < 0) RETURN(rc); } @@ -1408,7 +1333,6 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, RETURN(-ENOENT); } - mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA); rc = mdt_getattr_internal(info, child, 0); if (unlikely(rc != 0)) mdt_object_unlock(info, child, lhc, 1); @@ -1472,12 +1396,11 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, if (S_ISDIR(lu_object_attr(&parent->mot_obj))) { lhp = &info->mti_lh[MDT_LH_PARENT]; mdt_lock_pdo_init(lhp, LCK_PR, lname); - rc = mdt_object_lock(info, parent, lhp, - MDS_INODELOCK_UPDATE, - MDT_LOCAL_LOCK); - if (unlikely(rc != 0)) - RETURN(rc); - } + rc = mdt_object_lock(info, parent, lhp, + MDS_INODELOCK_UPDATE); + if (unlikely(rc != 0)) + RETURN(rc); + } /* step 2: lookup child's fid by name */ fid_zero(child_fid); @@ -1512,22 +1435,22 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, if (unlikely(IS_ERR(child))) GOTO(out_parent, rc = PTR_ERR(child)); + OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2); + if (!mdt_object_exists(child)) { + LU_OBJECT_DEBUG(D_INODE, info->mti_env, + &child->mot_obj, + "Object doesn't exist!\n"); + GOTO(out_child, rc = -ENOENT); + } + rc = mdt_check_resent_lock(info, child, lhc); if (rc < 0) { GOTO(out_child, rc); } else if (rc > 0) { - OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2); mdt_lock_handle_init(lhc); mdt_lock_reg_init(lhc, LCK_PR); try_layout = false; - if (!mdt_object_exists(child)) { - LU_OBJECT_DEBUG(D_INODE, info->mti_env, - &child->mot_obj, - "Object doesn't exist!\n"); - GOTO(out_child, rc = -ENOENT); - } - if (!(child_bits & MDS_INODELOCK_UPDATE) && mdt_object_exists(child) && !mdt_object_remote(child)) { struct md_attr *ma = &info->mti_attr; @@ -1566,12 +1489,12 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, child_bits |= MDS_INODELOCK_LAYOUT; /* try layout lock, it may fail to be granted due to * contention at LOOKUP or UPDATE */ - if (!mdt_object_lock_try(info, child, lhc, child_bits, - MDT_CROSS_LOCK)) { + if (!mdt_object_lock_try(info, child, lhc, + child_bits)) { child_bits &= ~MDS_INODELOCK_LAYOUT; LASSERT(child_bits != 0); rc = mdt_object_lock(info, child, lhc, - child_bits, MDT_CROSS_LOCK); + child_bits); } else { ma_need |= MA_LOV; } @@ -1580,22 +1503,15 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, * client will enqueue the lock to the remote MDT */ if (mdt_object_remote(child)) child_bits &= ~MDS_INODELOCK_UPDATE; - rc = mdt_object_lock(info, child, lhc, child_bits, - MDT_CROSS_LOCK); + rc = mdt_object_lock(info, child, lhc, child_bits); } if (unlikely(rc != 0)) GOTO(out_child, rc); } lock = ldlm_handle2lock(&lhc->mlh_reg_lh); - /* Get MA_SOM attributes if update lock is given. */ - if (lock && - lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE && - S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu))) - ma_need |= MA_SOM; /* finally, we can get attr for child. */ - mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA); rc = mdt_getattr_internal(info, child, ma_need); if (unlikely(rc != 0)) { mdt_object_unlock(info, child, lhc, 1); @@ -1607,8 +1523,6 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, "Lock res_id: "DLDLMRES", fid: "DFID"\n", PLDLMRES(lock->l_resource), PFID(mdt_object_fid(child))); - if (mdt_object_exists(child) && !mdt_object_remote(child)) - mdt_pack_size2body(info, child); } if (lock) LDLM_LOCK_PUT(lock); @@ -1716,6 +1630,9 @@ static int mdt_set_info(struct tgt_session_info *tsi) rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export, vallen, val, NULL); + } else if (KEY_IS(KEY_EVICT_BY_NID)) { + if (vallen > 0) + obd_export_evict_by_nid(req->rq_export->exp_obd, val); } else { RETURN(-EINVAL); } @@ -1845,7 +1762,10 @@ static int mdt_reint_internal(struct mdt_thread_info *info, if (rc != 0) GOTO(out_ucred, rc = err_serious(rc)); - if (mdt_check_resent(info, mdt_reconstruct, lhc)) { + rc = mdt_check_resent(info, mdt_reconstruct, lhc); + if (rc < 0) { + GOTO(out_ucred, rc); + } else if (rc == 1) { DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt."); rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg); GOTO(out_ucred, rc); @@ -2019,14 +1939,6 @@ static int mdt_quotactl(struct tgt_session_info *tsi) RETURN(err_serious(rc)); switch (oqctl->qc_cmd) { - case Q_QUOTACHECK: - case LUSTRE_Q_INVALIDATE: - case LUSTRE_Q_FINVALIDATE: - case Q_QUOTAON: - case Q_QUOTAOFF: - case Q_INITQUOTA: - /* deprecated, not used any more */ - RETURN(-EOPNOTSUPP); /* master quotactl */ case Q_SETINFO: case Q_SETQUOTA: @@ -2367,7 +2279,7 @@ int mdt_check_resent_lock(struct mdt_thread_info *info, struct mdt_lock_handle *lhc) { /* the lock might already be gotten in ldlm_handle_enqueue() */ - if (lustre_handle_is_used(&lhc->mlh_reg_lh)) { + if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) { struct ptlrpc_request *req = mdt_info_req(info); struct ldlm_lock *lock; @@ -2397,21 +2309,18 @@ int mdt_check_resent_lock(struct mdt_thread_info *info, return 1; } -int mdt_remote_object_lock(struct mdt_thread_info *mti, - struct mdt_object *o, const struct lu_fid *fid, - struct lustre_handle *lh, ldlm_mode_t mode, - __u64 ibits) +int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o, + const struct lu_fid *fid, struct lustre_handle *lh, + enum ldlm_mode mode, __u64 ibits, bool nonblock) { struct ldlm_enqueue_info *einfo = &mti->mti_einfo; - ldlm_policy_data_t *policy = &mti->mti_policy; + union ldlm_policy_data *policy = &mti->mti_policy; struct ldlm_res_id *res_id = &mti->mti_res_id; int rc = 0; ENTRY; LASSERT(mdt_object_remote(o)); - LASSERT(ibits == MDS_INODELOCK_UPDATE); - fid_build_reg_res_name(fid, res_id); memset(einfo, 0, sizeof(*einfo)); @@ -2421,6 +2330,8 @@ int mdt_remote_object_lock(struct mdt_thread_info *mti, einfo->ei_cb_cp = ldlm_completion_ast; einfo->ei_enq_slave = 0; einfo->ei_res_id = res_id; + if (nonblock) + einfo->ei_nonblock = 1; memset(policy, 0, sizeof(*policy)); policy->l_inodebits.bits = ibits; @@ -2433,14 +2344,14 @@ int mdt_remote_object_lock(struct mdt_thread_info *mti, static int mdt_object_local_lock(struct mdt_thread_info *info, struct mdt_object *o, struct mdt_lock_handle *lh, __u64 ibits, - bool nonblock, int locality) + bool nonblock) { - struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace; - ldlm_policy_data_t *policy = &info->mti_policy; - struct ldlm_res_id *res_id = &info->mti_res_id; + struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace; + union ldlm_policy_data *policy = &info->mti_policy; + struct ldlm_res_id *res_id = &info->mti_res_id; __u64 dlmflags; - int rc; - ENTRY; + int rc; + ENTRY; LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh)); LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh)); @@ -2524,25 +2435,30 @@ out_unlock: static int mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o, struct mdt_lock_handle *lh, __u64 ibits, - bool nonblock, int locality) + bool nonblock) { + struct mdt_lock_handle *local_lh = NULL; int rc; ENTRY; if (!mdt_object_remote(o)) - return mdt_object_local_lock(info, o, lh, ibits, nonblock, - locality); - - if (locality == MDT_LOCAL_LOCK) { - CERROR("%s: try to get local lock for remote object" - DFID".\n", mdt_obd_name(info->mti_mdt), - PFID(mdt_object_fid(o))); - RETURN(-EPROTO); - } + return mdt_object_local_lock(info, o, lh, ibits, nonblock); /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */ ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR); + + /* Only enqueue LOOKUP lock for remote object */ + if (ibits & MDS_INODELOCK_LOOKUP) { + rc = mdt_object_local_lock(info, o, lh, + MDS_INODELOCK_LOOKUP, + nonblock); + if (rc != ELDLM_OK) + RETURN(rc); + + local_lh = lh; + } + if (ibits & MDS_INODELOCK_UPDATE) { /* Sigh, PDO needs to enqueue 2 locks right now, but * enqueue RPC can only request 1 lock, to avoid extra @@ -2560,36 +2476,30 @@ mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o, rc = mdt_remote_object_lock(info, o, mdt_object_fid(o), &lh->mlh_rreg_lh, lh->mlh_rreg_mode, - MDS_INODELOCK_UPDATE); - if (rc != ELDLM_OK) - RETURN(rc); - } - - /* Only enqueue LOOKUP lock for remote object */ - if (ibits & MDS_INODELOCK_LOOKUP) { - rc = mdt_object_local_lock(info, o, lh, - MDS_INODELOCK_LOOKUP, - nonblock, locality); - if (rc != ELDLM_OK) + MDS_INODELOCK_UPDATE, nonblock); + if (rc != ELDLM_OK) { + if (local_lh != NULL) + mdt_object_unlock(info, o, local_lh, rc); RETURN(rc); + } } RETURN(0); } int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o, - struct mdt_lock_handle *lh, __u64 ibits, int locality) + struct mdt_lock_handle *lh, __u64 ibits) { - return mdt_object_lock_internal(info, o, lh, ibits, false, locality); + return mdt_object_lock_internal(info, o, lh, ibits, false); } int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o, - struct mdt_lock_handle *lh, __u64 ibits, int locality) + struct mdt_lock_handle *lh, __u64 ibits) { struct mdt_lock_handle tmp = *lh; int rc; - rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality); + rc = mdt_object_lock_internal(info, o, &tmp, ibits, true); if (rc == 0) *lh = tmp; @@ -2609,24 +2519,23 @@ int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o, * \param mode lock mode * \param decref force immediate lock releasing */ -static -void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h, - ldlm_mode_t mode, int decref) +static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h, + enum ldlm_mode mode, int decref) { - ENTRY; + ENTRY; - if (lustre_handle_is_used(h)) { - if (decref || !info->mti_has_trans || - !(mode & (LCK_PW | LCK_EX))){ - mdt_fid_unlock(h, mode); - } else { - struct mdt_device *mdt = info->mti_mdt; - struct ldlm_lock *lock = ldlm_handle2lock(h); - struct ptlrpc_request *req = mdt_info_req(info); - int no_ack = 0; + if (lustre_handle_is_used(h)) { + if (decref || !info->mti_has_trans || + !(mode & (LCK_PW | LCK_EX))) { + mdt_fid_unlock(h, mode); + } else { + struct mdt_device *mdt = info->mti_mdt; + struct ldlm_lock *lock = ldlm_handle2lock(h); + struct ptlrpc_request *req = mdt_info_req(info); + int no_ack = 0; - LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n", - h->cookie); + LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n", + h->cookie); /* there is no request if mdt_object_unlock() is called * from mdt_export_cleanup()->mdt_add_dirty_flag() */ if (likely(req != NULL)) { @@ -2660,7 +2569,7 @@ void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h, * Unlock mdt object. * * Immeditely release the regular lock and the PDO lock or save the - * lock in reqeuest and keep them referenced until client ACK or + * lock in request and keep them referenced until client ACK or * transaction commit. * * \param info thread info object @@ -2693,8 +2602,7 @@ struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info, if (!IS_ERR(o)) { int rc; - rc = mdt_object_lock(info, o, lh, ibits, - MDT_LOCAL_LOCK); + rc = mdt_object_lock(info, o, lh, ibits); if (rc != 0) { mdt_object_put(info->mti_env, o); o = ERR_PTR(rc); @@ -2748,27 +2656,11 @@ static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags) RETURN(-EINVAL); } - /* - * Do not get size or any capa fields before we check that request - * contains capa actually. There are some requests which do not, for - * instance MDS_IS_SUBDIR. - */ - if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) && - req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT)) - mdt_set_capainfo(info, 0, &body->mbo_fid1, - req_capsule_client_get(pill, &RMF_CAPA1)); - obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1); if (!IS_ERR(obj)) { - if ((flags & HABEO_CORPUS) && - !mdt_object_exists(obj)) { + if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) { mdt_object_put(env, obj); - /* for capability renew ENOENT will be handled in - * mdt_renew_capa */ - if (body->mbo_valid & OBD_MD_FLOSSCAPA) - rc = 0; - else - rc = -ENOENT; + rc = -ENOENT; } else { info->mti_object = obj; rc = 0; @@ -2804,17 +2696,6 @@ static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags) RETURN(rc); } -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_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) { lh->mlh_type = MDT_NUL_LOCK; @@ -2870,6 +2751,7 @@ void mdt_thread_info_init(struct ptlrpc_request *req, info->mti_spec.no_create = 0; info->mti_spec.sp_rm_entry = 0; info->mti_spec.sp_permitted = 0; + info->mti_spec.sp_migrate_close = 0; info->mti_spec.u.sp_ea.eadata = NULL; info->mti_spec.u.sp_ea.eadatalen = 0; @@ -2897,28 +2779,14 @@ void mdt_thread_info_fini(struct mdt_thread_info *info) struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi) { struct mdt_thread_info *mti; - struct lustre_capa *lc; mti = mdt_th_info(tsi->tsi_env); LASSERT(mti != NULL); mdt_thread_info_init(tgt_ses_req(tsi), mti); if (tsi->tsi_corpus != NULL) { - struct req_capsule *pill = tsi->tsi_pill; - mti->mti_object = mdt_obj(tsi->tsi_corpus); lu_object_get(tsi->tsi_corpus); - - /* - * 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->mbo_fid1, - lc); - } } mti->mti_body = tsi->tsi_mdt_body; mti->mti_dlm_req = tsi->tsi_dlm_req; @@ -2935,9 +2803,10 @@ static int mdt_tgt_connect(struct tgt_session_info *tsi) if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) && cfs_fail_val == - tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) - schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE, - msecs_to_jiffies(3 * MSEC_PER_SEC)); + tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) { + set_current_state(TASK_UNINTERRUPTIBLE); + schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC)); + } rc = tgt_connect(tsi); if (rc != 0) @@ -3169,12 +3038,12 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info, return; } - /* - * If the xid matches, then we know this is a resent request, and allow - * it. (It's probably an OPEN, for which we don't send a lock. - */ - if (req_xid_is_last(req)) - return; + /* + * If the xid matches, then we know this is a resent request, and allow + * it. (It's probably an OPEN, for which we don't send a lock. + */ + if (req_can_reconstruct(req, NULL)) + return; /* * This remote handle isn't enqueued, so we never received or processed @@ -3205,8 +3074,7 @@ static int mdt_intent_getxattr(enum mdt_it_code opcode, 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); + MDS_INODELOCK_XATTR); if (rc) return rc; } @@ -3261,9 +3129,9 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, GOTO(out_shrink, rc = -EINVAL); } - rc = mdt_init_ucred(info, reqbody); - if (rc) - GOTO(out_shrink, rc); + rc = mdt_init_ucred_intent_getattr(info, reqbody); + if (rc) + GOTO(out_shrink, rc); ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP); mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD); @@ -3400,8 +3268,8 @@ static int mdt_intent_reint(enum mdt_it_code opcode, /* the open lock or the lock for cross-ref object should be * returned to the client */ - if (rc == -MDT_EREMOTE_OPEN || mdt_get_disposition(rep, DISP_OPEN_LOCK)) { - LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh)); + if (lustre_handle_is_used(&lhc->mlh_reg_lh) && + (rc == 0 || rc == -MDT_EREMOTE_OPEN)) { rep->lock_policy_res2 = 0; rc = mdt_intent_lock_replace(info, lockp, lhc, flags); RETURN(rc); @@ -3413,6 +3281,7 @@ static int mdt_intent_reint(enum mdt_it_code opcode, mdt_get_disposition(rep, DISP_LOOKUP_NEG)) rep->lock_policy_res2 = 0; + lhc->mlh_reg_lh.cookie = 0ull; if (rc == -ENOTCONN || rc == -ENODEV || rc == -EOVERFLOW) { /**< if VBR failure then return error */ /* @@ -3421,78 +3290,64 @@ static int mdt_intent_reint(enum mdt_it_code opcode, * will detect this, then disconnect, reconnect the import * immediately, instead of impacting the following the rpc. */ - lhc->mlh_reg_lh.cookie = 0ull; RETURN(rc); - } else { - /* - * For other cases, the error will be returned by intent. - * and client will retrieve the result from intent. - */ - /* - * FIXME: when open lock is finished, that should be - * checked here. - */ - if (lustre_handle_is_used(&lhc->mlh_reg_lh)) { - LASSERTF(rc == 0, "Error occurred but lock handle " - "is still in use, rc = %d\n", rc); - rep->lock_policy_res2 = 0; - rc = mdt_intent_lock_replace(info, lockp, lhc, flags); - RETURN(rc); - } else { - lhc->mlh_reg_lh.cookie = 0ull; - RETURN(ELDLM_LOCK_ABORTED); - } } + /* + * For other cases, the error will be returned by intent, and client + * will retrieve the result from intent. + */ + RETURN(ELDLM_LOCK_ABORTED); } -static int mdt_intent_code(long itcode) +static int mdt_intent_code(enum ldlm_intent_flags itcode) { - int rc; + int rc; - switch(itcode) { - case IT_OPEN: - rc = MDT_IT_OPEN; - break; - case IT_OPEN|IT_CREAT: - rc = MDT_IT_OCREAT; - break; - case IT_CREAT: - rc = MDT_IT_CREATE; - break; - case IT_READDIR: - rc = MDT_IT_READDIR; - break; - case IT_GETATTR: - rc = MDT_IT_GETATTR; - break; - case IT_LOOKUP: - rc = MDT_IT_LOOKUP; - break; - case IT_UNLINK: - rc = MDT_IT_UNLINK; - break; - case IT_TRUNC: - rc = MDT_IT_TRUNC; - break; - case IT_GETXATTR: - rc = MDT_IT_GETXATTR; - break; - case IT_LAYOUT: - rc = MDT_IT_LAYOUT; - break; + switch (itcode) { + case IT_OPEN: + rc = MDT_IT_OPEN; + break; + case IT_OPEN|IT_CREAT: + rc = MDT_IT_OCREAT; + break; + case IT_CREAT: + rc = MDT_IT_CREATE; + break; + case IT_READDIR: + rc = MDT_IT_READDIR; + break; + case IT_GETATTR: + rc = MDT_IT_GETATTR; + break; + case IT_LOOKUP: + rc = MDT_IT_LOOKUP; + break; + case IT_UNLINK: + rc = MDT_IT_UNLINK; + break; + case IT_TRUNC: + rc = MDT_IT_TRUNC; + break; + case IT_GETXATTR: + rc = MDT_IT_GETXATTR; + break; + case IT_LAYOUT: + rc = MDT_IT_LAYOUT; + break; case IT_QUOTA_DQACQ: case IT_QUOTA_CONN: rc = MDT_IT_QUOTA; break; - default: - CERROR("Unknown intent opcode: %ld\n", itcode); - rc = -EINVAL; - break; - } - return rc; + default: + CERROR("Unknown intent opcode: 0x%08x\n", itcode); + rc = -EINVAL; + break; + } + return rc; } -static int mdt_intent_opc(long itopc, struct mdt_thread_info *info, +static int mdt_intent_opc(enum ldlm_intent_flags itopc, + struct mdt_thread_info *info, struct ldlm_lock **lockp, __u64 flags) { struct req_capsule *pill = info->mti_pill; @@ -3535,6 +3390,8 @@ static int mdt_intent_opc(long itopc, struct mdt_thread_info *info, if (flv->it_act != NULL) { struct ldlm_reply *rep; + OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10); + /* execute policy */ rc = flv->it_act(opc, info, lockp, flags); @@ -3551,8 +3408,8 @@ static int mdt_intent_opc(long itopc, struct mdt_thread_info *info, } 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 ldlm_lock **lockp, void *req_cookie, + enum ldlm_mode mode, __u64 flags, void *data) { struct tgt_session_info *tsi; struct mdt_thread_info *info; @@ -3655,20 +3512,26 @@ static int mdt_register_lwp_callback(void *data) LASSERT(mdt_seq_site(mdt)->ss_node_id != 0); - if (!likely(fld->lsf_new)) - RETURN(0); - rc = lu_env_init(&env, LCT_MD_THREAD); - if (rc) { + if (rc < 0) { CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc); RETURN(rc); } - rc = fld_update_from_controller(&env, fld); - if (rc != 0) { - CERROR("%s: cannot update controller: rc = %d\n", - mdt_obd_name(mdt), rc); + /* Allocate new sequence now to avoid creating local transaction + * in the normal transaction process */ + rc = seq_server_check_and_alloc_super(&env, + mdt_seq_site(mdt)->ss_server_seq); + if (rc < 0) GOTO(out, rc); + + if (fld->lsf_new) { + rc = fld_update_from_controller(&env, fld); + if (rc != 0) { + CERROR("%s: cannot update controller: rc = %d\n", + mdt_obd_name(mdt), rc); + GOTO(out, rc); + } } out: lu_env_fini(&env); @@ -3789,7 +3652,7 @@ static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt) GOTO(out_seq_fini, rc); if (ss->ss_node_id != 0) - /* register controler export through lwp */ + /* register controller export through lwp */ rc = mdt_register_seq_exp(mdt); EXIT; @@ -3869,10 +3732,9 @@ static void mdt_stack_pre_fini(const struct lu_env *env, lustre_cfg_bufs_reset(bufs, mdt_obd_name(m)); lustre_cfg_bufs_set_string(bufs, 1, NULL); lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs); - if (!lcfg) { - CERROR("%s: cannot alloc lcfg\n", mdt_obd_name(m)); - return; - } + if (lcfg == NULL) + RETURN_EXIT; + top->ld_ops->ldo_process_config(env, top, lcfg); lustre_cfg_free(lcfg); EXIT; @@ -3905,10 +3767,9 @@ static void mdt_stack_fini(const struct lu_env *env, 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; - } + if (lcfg == NULL) + RETURN_EXIT; + LASSERT(top); top->ld_ops->ldo_process_config(env, top, lcfg); lustre_cfg_free(lcfg); @@ -4032,8 +3893,8 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt); lcfg = lustre_cfg_new(LCFG_ATTACH, bufs); - if (!lcfg) - GOTO(free_bufs, rc = -ENOMEM); + if (lcfg == NULL) + GOTO(put_profile, rc = -ENOMEM); rc = class_attach(lcfg); if (rc) @@ -4054,6 +3915,8 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt); lcfg = lustre_cfg_new(LCFG_SETUP, bufs); + if (lcfg == NULL) + GOTO(class_detach, rc = -ENOMEM); rc = class_setup(obd, lcfg); if (rc) @@ -4091,6 +3954,8 @@ class_detach: class_detach(obd, lcfg); lcfg_cleanup: lustre_cfg_free(lcfg); +put_profile: + class_put_profile(lprof); free_bufs: OBD_FREE_PTR(bufs); cleanup_mem: @@ -4160,8 +4025,8 @@ static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt, lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt); lcfg = lustre_cfg_new(LCFG_ATTACH, bufs); - if (!lcfg) - GOTO(cleanup_mem, rc = -ENOMEM); + if (lcfg == NULL) + GOTO(put_profile, rc = -ENOMEM); rc = class_attach(lcfg); if (rc) @@ -4185,6 +4050,8 @@ static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt, mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name); lcfg = lustre_cfg_new(LCFG_SETUP, bufs); + if (lcfg == NULL) + GOTO(class_detach, rc = -ENOMEM); rc = class_setup(obd, lcfg); if (rc) @@ -4221,6 +4088,8 @@ class_detach: class_detach(obd, lcfg); lcfg_cleanup: lustre_cfg_free(lcfg); +put_profile: + class_put_profile(lprof); cleanup_mem: if (bufs) OBD_FREE_PTR(bufs); @@ -4252,7 +4121,7 @@ static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt) /* 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) +static int mdt_tgt_getxattr(struct tgt_session_info *tsi) { struct mdt_thread_info *info = tsi2mdt_info(tsi); int rc; @@ -4282,8 +4151,6 @@ 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(0, MDS_QUOTACTL, mdt_quotactl), @@ -4384,9 +4251,9 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) stop.ls_flags = 0; next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop); + mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child)); target_recovery_fini(obd); 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); @@ -4395,6 +4262,11 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT); mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT); + + if (m->mdt_namespace != NULL) + ldlm_namespace_free_prior(m->mdt_namespace, NULL, + d->ld_obd->obd_force); + obd_exports_barrier(obd); obd_zombie_barrier(); @@ -4405,11 +4277,10 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) upcall_cache_cleanup(m->mdt_identity_cache); m->mdt_identity_cache = NULL; - if (m->mdt_namespace != NULL) { - ldlm_namespace_free(m->mdt_namespace, NULL, - d->ld_obd->obd_force); - d->ld_obd->obd_namespace = m->mdt_namespace = NULL; - } + if (m->mdt_namespace != NULL) { + ldlm_namespace_free_post(m->mdt_namespace); + d->ld_obd->obd_namespace = m->mdt_namespace = NULL; + } mdt_quota_fini(env, m); @@ -4418,10 +4289,6 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) mdt_seq_fini(env, m); mdt_fld_fini(env, m); - 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 */ @@ -4474,9 +4341,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, LASSERT(obd != NULL); m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */ - - m->mdt_som_conf = 0; - + m->mdt_opts.mo_evict_tgt_nids = 1; m->mdt_opts.mo_cos = MDT_COS_DEFAULT; /* default is coordinator off, it is started through conf_param @@ -4493,12 +4358,11 @@ 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; + if (lsi->lsi_lmd != NULL && + lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK) + m->mdt_skip_lfsck = 1; } - spin_lock_init(&m->mdt_ioepoch_lock); - m->mdt_capa_timeout = CAPA_TIMEOUT; - m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1; - m->mdt_ck_timeout = CAPA_KEY_TIMEOUT; m->mdt_squash.rsi_uid = 0; m->mdt_squash.rsi_gid = 0; INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids); @@ -4569,8 +4433,6 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, /* set obd_namespace for compatibility with old code */ obd->obd_namespace = m->mdt_namespace; - cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m); - rc = mdt_hsm_cdt_init(m); if (rc != 0) { CERROR("%s: error initializing coordinator, rc %d\n", @@ -4578,15 +4440,11 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, GOTO(err_free_ns, rc); } - rc = mdt_ck_thread_start(m); - if (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); + GOTO(err_free_hsm, rc); rc = mdt_fs_setup(env, m, obd, lsi); if (rc) @@ -4648,8 +4506,6 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, * when the whole stack is complete and ready * to serve the requests */ - mdt_init_capa_ctxt(env, m); - /* Reduce the initial timeout on an MDS because it doesn't need such * a long timeout as an OST does. Adaptive timeouts will adjust this * value appropriately. */ @@ -4667,9 +4523,6 @@ 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); err_free_hsm: mdt_hsm_cdt_fini(m); err_free_ns: @@ -4790,7 +4643,7 @@ static struct lu_object *mdt_object_alloc(const struct lu_env *env, lu_object_init(o, h, d); lu_object_add_top(h, o); o->lo_ops = &mdt_obj_ops; - mutex_init(&mo->mot_ioepoch_mutex); + spin_lock_init(&mo->mot_write_lock); mutex_init(&mo->mot_lov_mutex); init_rwsem(&mo->mot_open_sem); RETURN(o); @@ -4841,13 +4694,13 @@ static void mdt_object_free(const struct lu_env *env, struct lu_object *o) } static int mdt_object_print(const struct lu_env *env, void *cookie, - lu_printer_t p, const struct lu_object *o) + lu_printer_t p, const struct lu_object *o) { - struct mdt_object *mdto = mdt_obj((struct lu_object *)o); - return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" " - "flags="LPX64", epochcount=%d, writecount=%d)", - mdto, mdto->mot_ioepoch, mdto->mot_flags, - mdto->mot_ioepoch_count, mdto->mot_writecount); + struct mdt_object *mdto = mdt_obj((struct lu_object *)o); + + return (*p)(env, cookie, + LUSTRE_MDT_NAME"-object@%p(flags=%d, writecount=%d)", + mdto, mdto->mot_flags, mdto->mot_write_count); } static int mdt_prepare(const struct lu_env *env, @@ -4888,6 +4741,7 @@ static int mdt_prepare(const struct lu_env *env, } LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state)); + target_recovery_init(&mdt->mdt_lut, tgt_request_handle); set_bit(MDT_FL_CFGLOG, &mdt->mdt_state); LASSERT(obd->obd_no_conn); @@ -4976,9 +4830,6 @@ static int mdt_connect_internal(struct obd_export *exp, if (!mdt->mdt_opts.mo_user_xattr) data->ocd_connect_flags &= ~OBD_CONNECT_XATTR; - if (!mdt->mdt_som_conf) - data->ocd_connect_flags &= ~OBD_CONNECT_SOM; - if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) { data->ocd_brw_size = min(data->ocd_brw_size, (__u32)MD_MAX_BRW_SIZE); @@ -5016,15 +4867,6 @@ static int mdt_connect_internal(struct obd_export *exp, return -EBADE; } - if (mdt->mdt_som_conf && - !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT | - OBD_CONNECT_MDS_MDS | - OBD_CONNECT_SOM))) { - CWARN("%s: MDS has SOM enabled, but client does not support " - "it\n", mdt_obd_name(mdt)); - return -EBADE; - } - if (OCD_HAS_FLAG(data, PINGLESS)) { if (ptlrpc_pinger_suppress_pings()) { spin_lock(&exp->exp_obd->obd_dev_lock); @@ -5037,6 +4879,18 @@ static int mdt_connect_internal(struct obd_export *exp, data->ocd_max_easize = mdt->mdt_max_ea_size; + /* NB: Disregard the rule against updating + * exp_connect_data.ocd_connect_flags in this case, since + * tgt_client_new() needs to know if this is client supports + * multiple modify RPCs, and it is safe to expose this flag before + * connection processing completes. */ + if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) { + data->ocd_maxmodrpcs = max_mod_rpcs_per_client; + spin_lock(&exp->exp_lock); + *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS; + spin_unlock(&exp->exp_lock); + } + return 0; } @@ -5126,7 +4980,7 @@ static int mdt_export_cleanup(struct obd_export *exp) * archive request into a noop if it's not actually * dirty. */ - if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC)) + if (mfd->mfd_mode & FMODE_WRITE) rc = mdt_ctxt_add_dirty_flag(&env, info, mfd); /* Don't unlink orphan on failover umount, LU-184 */ @@ -5155,12 +5009,12 @@ static int mdt_obd_disconnect(struct obd_export *exp) LASSERT(exp); class_export_get(exp); - nodemap_del_member(exp); rc = server_disconnect_export(exp); if (rc != 0) CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc); rc = mdt_export_cleanup(exp); + nodemap_del_member(exp); class_export_put(exp); RETURN(rc); } @@ -5191,10 +5045,11 @@ static int mdt_obd_connect(const struct lu_env *env, * at some point we should find a better one */ if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL && - !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) { + !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) && + !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) { rc = obd_get_info(env, mdt->mdt_child_exp, sizeof(KEY_OSP_CONNECTED), - KEY_OSP_CONNECTED, NULL, NULL, NULL); + KEY_OSP_CONNECTED, NULL, NULL); if (rc) RETURN(-EAGAIN); set_bit(MDT_FL_SYNCED, &mdt->mdt_state); @@ -5207,6 +5062,10 @@ static int mdt_obd_connect(const struct lu_env *env, lexp = class_conn2export(&conn); LASSERT(lexp != NULL); + rc = nodemap_add_member(*client_nid, lexp); + if (rc != 0 && rc != -EEXIST) + GOTO(out, rc); + rc = mdt_connect_internal(lexp, mdt, data); if (rc == 0) { struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd; @@ -5214,22 +5073,13 @@ static int mdt_obd_connect(const struct lu_env *env, LASSERT(lcd); memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid); rc = tgt_client_new(env, lexp); - if (rc == 0) { - rc = nodemap_add_member(*client_nid, lexp); - if (rc != 0 && rc != -EEXIST) - goto out; - + if (rc == 0) mdt_export_stats_init(obd, lexp, localdata); - } - - /* For phase I, sync for cross-ref operation. */ - spin_lock(&lexp->exp_lock); - lexp->exp_keep_sync = 1; - spin_unlock(&lexp->exp_lock); } out: if (rc != 0) { class_disconnect(lexp); + nodemap_del_member(lexp); *exp = NULL; } else { *exp = lexp; @@ -5251,12 +5101,15 @@ static int mdt_obd_reconnect(const struct lu_env *env, if (exp == NULL || obd == NULL || cluuid == NULL) RETURN(-EINVAL); + rc = nodemap_add_member(*client_nid, exp); + if (rc != 0 && rc != -EEXIST) + RETURN(rc); + rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data); - if (rc == 0) { - rc = nodemap_add_member(*client_nid, exp); - if (rc == 0 || rc == -EEXIST) - mdt_export_stats_init(obd, exp, localdata); - } + if (rc == 0) + mdt_export_stats_init(obd, exp, localdata); + else + nodemap_del_member(exp); RETURN(rc); } @@ -5378,7 +5231,7 @@ static int mdt_path_current(struct mdt_thread_info *info, struct lu_buf *buf = &info->mti_big_buf; char *ptr; int reclen; - struct linkea_data ldata = { 0 }; + struct linkea_data ldata = { NULL }; int rc = 0; bool first = true; ENTRY; @@ -5557,10 +5410,11 @@ static int mdt_fid2path(struct mdt_thread_info *info, } 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)); - RETURN(-EINVAL); + if (IS_ERR(obj)) { + rc = PTR_ERR(obj); + CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n", + PFID(&fp->gf_fid), rc); + RETURN(rc); } if (mdt_object_remote(obj)) @@ -5752,6 +5606,7 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, break; case OBD_IOC_ABORT_RECOVERY: CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt)); + obd->obd_force_abort_recovery = 1; target_stop_recovery_thread(obd); rc = 0; break; @@ -5824,18 +5679,21 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt) { struct lu_device *ld = md2lu_dev(mdt->mdt_child); - struct lfsck_start_param lsp; int rc; ENTRY; - lsp.lsp_start = NULL; - lsp.lsp_index_valid = 0; - rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child, - OBD_IOC_START_LFSCK, - 0, &lsp); - if (rc != 0 && rc != -EALREADY) - CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n", - mdt_obd_name(mdt), rc); + if (!mdt->mdt_skip_lfsck) { + struct lfsck_start_param lsp; + + lsp.lsp_start = NULL; + lsp.lsp_index_valid = 0; + rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child, + OBD_IOC_START_LFSCK, + 0, &lsp); + if (rc != 0 && rc != -EALREADY) + CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n", + mdt_obd_name(mdt), rc); + } rc = ld->ld_ops->ldo_recovery_complete(env, ld); RETURN(rc); @@ -5989,7 +5847,7 @@ static struct lu_device_type mdt_device_type = { .ldt_ctx_tags = LCT_MD_THREAD }; -static int __init mdt_mod_init(void) +static int __init mdt_init(void) { int rc; @@ -6018,15 +5876,17 @@ mds_fini: return rc; } -static void __exit mdt_mod_exit(void) +static void __exit mdt_exit(void) { class_unregister_type(LUSTRE_MDT_NAME); mds_mod_exit(); lu_kmem_fini(mdt_caches); } -MODULE_AUTHOR("Sun Microsystems, Inc. "); +MODULE_AUTHOR("OpenSFS, Inc. "); MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")"); +MODULE_VERSION(LUSTRE_VERSION_STRING); MODULE_LICENSE("GPL"); -cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit); +module_init(mdt_init); +module_exit(mdt_exit);