X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Flmv%2Flmv_obd.c;h=20fb83473144f6cf22977a2663e99d999c3a4e93;hp=3e765baf69b399b4dbfa704a3cb67c06cda6492e;hb=caa70178a413c84c400e875866cdbbd368df4758;hpb=ad538dba7f5259df31fc45e90835edc12756db40 diff --git a/lustre/lmv/lmv_obd.c b/lustre/lmv/lmv_obd.c index 3e765ba..20fb834 100644 --- a/lustre/lmv/lmv_obd.c +++ b/lustre/lmv/lmv_obd.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -28,18 +26,14 @@ /* * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. - */ -/* - * Copyright (c) 2011 Whamcloud, Inc. + * + * Copyright (c) 2011, 2013, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ * Lustre is a trademark of Sun Microsystems, Inc. */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_LMV #ifdef __KERNEL__ #include @@ -55,7 +49,7 @@ #include #endif -#include +#include #include #include #include @@ -65,10 +59,6 @@ #include #include "lmv_internal.h" -/* object cache. */ -cfs_mem_cache_t *lmv_object_cache; -cfs_atomic_t lmv_object_count = CFS_ATOMIC_INIT(0); - static void lmv_activate_target(struct lmv_obd *lmv, struct lmv_tgt_desc *tgt, int activate) @@ -87,29 +77,31 @@ static void lmv_activate_target(struct lmv_obd *lmv, * -ENOTCONN: The UUID is found, but the target connection is bad (!) * -EBADF : The UUID is found, but the OBD of the wrong type (!) */ -static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid, - int activate) +static int lmv_set_mdc_active(struct lmv_obd *lmv, + const struct obd_uuid *uuid, + int activate) { - struct lmv_tgt_desc *tgt; - struct obd_device *obd; - int i; - int rc = 0; - ENTRY; + struct lmv_tgt_desc *tgt = NULL; + struct obd_device *obd; + __u32 i; + int rc = 0; + ENTRY; - CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n", - lmv, uuid->uuid, activate); + CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n", + lmv, uuid->uuid, activate); - cfs_spin_lock(&lmv->lmv_lock); - for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) { - if (tgt->ltd_exp == NULL) - continue; + spin_lock(&lmv->lmv_lock); + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + tgt = lmv->tgts[i]; + if (tgt == NULL || tgt->ltd_exp == NULL) + continue; - CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", - i, tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie); + CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i, + tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie); - if (obd_uuid_equals(uuid, &tgt->ltd_uuid)) - break; - } + if (obd_uuid_equals(uuid, &tgt->ltd_uuid)) + break; + } if (i == lmv->desc.ld_tgt_count) GOTO(out_lmv_lock, rc = -EINVAL); @@ -135,37 +127,16 @@ static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid, EXIT; out_lmv_lock: - cfs_spin_unlock(&lmv->lmv_lock); - return rc; + spin_unlock(&lmv->lmv_lock); + return rc; } -static int lmv_set_mdc_data(struct lmv_obd *lmv, struct obd_uuid *uuid, - struct obd_connect_data *data) +struct obd_uuid *lmv_get_uuid(struct obd_export *exp) { - struct lmv_tgt_desc *tgt; - int i; - ENTRY; + struct lmv_obd *lmv = &exp->exp_obd->u.lmv; + struct lmv_tgt_desc *tgt = lmv->tgts[0]; - LASSERT(data != NULL); - - cfs_spin_lock(&lmv->lmv_lock); - for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) { - if (tgt->ltd_exp == NULL) - continue; - - if (obd_uuid_equals(uuid, &tgt->ltd_uuid)) { - lmv->datas[tgt->ltd_idx] = *data; - break; - } - } - cfs_spin_unlock(&lmv->lmv_lock); - RETURN(0); -} - -struct obd_uuid *lmv_get_uuid(struct obd_export *exp) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - return obd_get_uuid(lmv->tgts[0].ltd_exp); + return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp); } static int lmv_notify(struct obd_device *obd, struct obd_device *watched, @@ -198,27 +169,15 @@ static int lmv_notify(struct obd_device *obd, struct obd_device *watched, uuid->uuid, rc); RETURN(rc); } - } else if (ev == OBD_NOTIFY_OCD) { - conn_data = &watched->u.cli.cl_import->imp_connect_data; - - /* - * Set connect data to desired target, update exp_connect_flags. - */ - rc = lmv_set_mdc_data(lmv, uuid, conn_data); - if (rc) { - CERROR("can't set connect data to target %s, rc %d\n", - uuid->uuid, rc); - RETURN(rc); - } - - /* - * XXX: Make sure that ocd_connect_flags from all targets are - * the same. Otherwise one of MDTs runs wrong version or - * something like this. --umka - */ - obd->obd_self_export->exp_connect_flags = - conn_data->ocd_connect_flags; - } + } else if (ev == OBD_NOTIFY_OCD) { + conn_data = &watched->u.cli.cl_import->imp_connect_data; + /* + * XXX: Make sure that ocd_connect_flags from all targets are + * the same. Otherwise one of MDTs runs wrong version or + * something like this. --umka + */ + obd->obd_self_export->exp_connect_data = *conn_data; + } #if 0 else if (ev == OBD_NOTIFY_DISCON) { /* @@ -295,7 +254,7 @@ static int lmv_connect(const struct lu_env *env, * and MDC stuff will be called directly, for instance while reading * ../mdc/../kbytesfree procfs file, etc. */ - if (data->ocd_connect_flags & OBD_CONNECT_REAL) + if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL)) rc = lmv_check_connect(obd); #ifdef __KERNEL__ @@ -310,9 +269,8 @@ static int lmv_connect(const struct lu_env *env, static void lmv_set_timeouts(struct obd_device *obd) { - struct lmv_tgt_desc *tgts; - struct lmv_obd *lmv; - int i; + struct lmv_obd *lmv; + __u32 i; lmv = &obd->u.lmv; if (lmv->server_timeout == 0) @@ -321,24 +279,26 @@ static void lmv_set_timeouts(struct obd_device *obd) if (lmv->connected == 0) return; - for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) { - if (tgts->ltd_exp == NULL) - continue; + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + struct lmv_tgt_desc *tgt = lmv->tgts[i]; - obd_set_info_async(tgts->ltd_exp, sizeof(KEY_INTERMDS), - KEY_INTERMDS, 0, NULL, NULL); - } + if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) + continue; + + obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS), + KEY_INTERMDS, 0, NULL, NULL); + } } static int lmv_init_ea_size(struct obd_export *exp, int easize, int def_easize, int cookiesize) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - int i; - int rc = 0; - int change = 0; - ENTRY; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + __u32 i; + int rc = 0; + int change = 0; + ENTRY; if (lmv->max_easize < easize) { lmv->max_easize = easize; @@ -358,21 +318,23 @@ static int lmv_init_ea_size(struct obd_export *exp, int easize, if (lmv->connected == 0) RETURN(0); - for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - if (lmv->tgts[i].ltd_exp == NULL) { - CWARN("%s: NULL export for %d\n", obd->obd_name, i); - continue; - } + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + struct lmv_tgt_desc *tgt = lmv->tgts[i]; - rc = md_init_ea_size(lmv->tgts[i].ltd_exp, easize, def_easize, - cookiesize); - if (rc) { - CERROR("obd_init_ea_size() failed on MDT target %d, " - "error %d.\n", i, rc); - break; - } - } - RETURN(rc); + if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) { + CWARN("%s: NULL export for %d\n", obd->obd_name, i); + continue; + } + + rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize, + cookiesize); + if (rc) { + CERROR("%s: obd_init_ea_size() failed on MDT target %d:" + " rc = %d.\n", obd->obd_name, i, rc); + break; + } + } + RETURN(rc); } #define MAX_STRING_SIZE 128 @@ -384,7 +346,6 @@ int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt) #endif struct lmv_obd *lmv = &obd->u.lmv; struct obd_uuid *cluuid = &lmv->cluuid; - struct obd_connect_data *mdc_data = NULL; struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" }; struct obd_device *mdc_obd; struct obd_export *mdc_exp; @@ -416,12 +377,12 @@ int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt) RETURN(rc); } - /* - * Init fid sequence client for this mdc and add new fld target. - */ - rc = obd_fid_init(mdc_exp); - if (rc) - RETURN(rc); + /* + * Init fid sequence client for this mdc and add new fld target. + */ + rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA); + if (rc) + RETURN(rc); target.ft_srv = NULL; target.ft_exp = mdc_exp; @@ -429,8 +390,6 @@ int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt) fld_client_add_target(&lmv->lmv_fld, &target); - mdc_data = &class_exp2cliimp(mdc_exp)->imp_connect_data; - rc = obd_register_observer(mdc_obd, obd); if (rc) { obd_disconnect(mdc_exp); @@ -443,23 +402,19 @@ int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt) /* * Tell the observer about the new target. */ - rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd, - OBD_NOTIFY_ACTIVE, (void *)(tgt - lmv->tgts)); - if (rc) { - obd_disconnect(mdc_exp); - RETURN(rc); - } + rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd, + OBD_NOTIFY_ACTIVE, + (void *)(tgt - lmv->tgts[0])); + if (rc) { + obd_disconnect(mdc_exp); + RETURN(rc); + } } tgt->ltd_active = 1; tgt->ltd_exp = mdc_exp; lmv->desc.ld_active_tgt_count++; - /* - * Copy connect data, it may be used later. - */ - lmv->datas[tgt->ltd_idx] = *mdc_data; - md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize, lmv->max_cookiesize); @@ -492,53 +447,102 @@ int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt) RETURN(0); } -int lmv_add_target(struct obd_device *obd, struct obd_uuid *tgt_uuid) +static void lmv_del_target(struct lmv_obd *lmv, int index) +{ + if (lmv->tgts[index] == NULL) + return; + + OBD_FREE_PTR(lmv->tgts[index]); + lmv->tgts[index] = NULL; + return; +} + +static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp, + __u32 index, int gen) { struct lmv_obd *lmv = &obd->u.lmv; struct lmv_tgt_desc *tgt; int rc = 0; ENTRY; - CDEBUG(D_CONFIG, "Target uuid: %s.\n", tgt_uuid->uuid); + CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index); lmv_init_lock(lmv); - if (lmv->desc.ld_active_tgt_count >= LMV_MAX_TGT_COUNT) { - lmv_init_unlock(lmv); - CERROR("Can't add %s, LMV module compiled for %d MDCs. " - "That many MDCs already configured.\n", - tgt_uuid->uuid, LMV_MAX_TGT_COUNT); - RETURN(-EINVAL); - } - if (lmv->desc.ld_tgt_count == 0) { - struct obd_device *mdc_obd; - - mdc_obd = class_find_client_obd(tgt_uuid, LUSTRE_MDC_NAME, - &obd->obd_uuid); - if (!mdc_obd) { - lmv_init_unlock(lmv); - CERROR("Target %s not attached\n", tgt_uuid->uuid); - RETURN(-EINVAL); - } - - rc = obd_llog_init(obd, &obd->obd_olg, mdc_obd, NULL); - if (rc) { - lmv_init_unlock(lmv); - CERROR("lmv failed to setup llogging subsystems\n"); - } - } - cfs_spin_lock(&lmv->lmv_lock); - tgt = lmv->tgts + lmv->desc.ld_tgt_count++; - tgt->ltd_uuid = *tgt_uuid; - cfs_spin_unlock(&lmv->lmv_lock); - - if (lmv->connected) { - rc = lmv_connect_mdc(obd, tgt); - if (rc) { - cfs_spin_lock(&lmv->lmv_lock); - lmv->desc.ld_tgt_count--; - memset(tgt, 0, sizeof(*tgt)); - cfs_spin_unlock(&lmv->lmv_lock); + if (lmv->desc.ld_tgt_count == 0) { + struct obd_device *mdc_obd; + + mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, + &obd->obd_uuid); + if (!mdc_obd) { + lmv_init_unlock(lmv); + CERROR("%s: Target %s not attached: rc = %d\n", + obd->obd_name, uuidp->uuid, -EINVAL); + RETURN(-EINVAL); + } + } + + if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) { + tgt = lmv->tgts[index]; + CERROR("%s: UUID %s already assigned at LOV target index %d:" + " rc = %d\n", obd->obd_name, + obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST); + lmv_init_unlock(lmv); + RETURN(-EEXIST); + } + + if (index >= lmv->tgts_size) { + /* We need to reallocate the lmv target array. */ + struct lmv_tgt_desc **newtgts, **old = NULL; + __u32 newsize = 1; + __u32 oldsize = 0; + + while (newsize < index + 1) + newsize = newsize << 1; + OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize); + if (newtgts == NULL) { + lmv_init_unlock(lmv); + RETURN(-ENOMEM); + } + + if (lmv->tgts_size) { + memcpy(newtgts, lmv->tgts, + sizeof(*newtgts) * lmv->tgts_size); + old = lmv->tgts; + oldsize = lmv->tgts_size; + } + + lmv->tgts = newtgts; + lmv->tgts_size = newsize; + smp_rmb(); + if (old) + OBD_FREE(old, sizeof(*old) * oldsize); + + CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts, + lmv->tgts_size); + } + + OBD_ALLOC_PTR(tgt); + if (!tgt) { + lmv_init_unlock(lmv); + RETURN(-ENOMEM); + } + + mutex_init(&tgt->ltd_fid_mutex); + tgt->ltd_idx = index; + tgt->ltd_uuid = *uuidp; + tgt->ltd_active = 0; + lmv->tgts[index] = tgt; + if (index >= lmv->desc.ld_tgt_count) + lmv->desc.ld_tgt_count = index + 1; + + if (lmv->connected) { + rc = lmv_connect_mdc(obd, tgt); + if (rc) { + spin_lock(&lmv->lmv_lock); + lmv->desc.ld_tgt_count--; + memset(tgt, 0, sizeof(*tgt)); + spin_unlock(&lmv->lmv_lock); } else { int easize = sizeof(struct lmv_stripe_md) + lmv->desc.ld_tgt_count * @@ -553,12 +557,12 @@ int lmv_add_target(struct obd_device *obd, struct obd_uuid *tgt_uuid) int lmv_check_connect(struct obd_device *obd) { - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_tgt_desc *tgt; - int i; - int rc; - int easize; - ENTRY; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt; + __u32 i; + int rc; + int easize; + ENTRY; if (lmv->connected) RETURN(0); @@ -570,20 +574,31 @@ int lmv_check_connect(struct obd_device *obd) } if (lmv->desc.ld_tgt_count == 0) { + lmv_init_unlock(lmv); CERROR("%s: no targets configured.\n", obd->obd_name); RETURN(-EINVAL); } - CDEBUG(D_CONFIG, "Time to connect %s to %s\n", - lmv->cluuid.uuid, obd->obd_name); + LASSERT(lmv->tgts != NULL); - LASSERT(lmv->tgts != NULL); + if (lmv->tgts[0] == NULL) { + lmv_init_unlock(lmv); + CERROR("%s: no target configured for index 0.\n", + obd->obd_name); + RETURN(-EINVAL); + } - for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) { - rc = lmv_connect_mdc(obd, tgt); - if (rc) - GOTO(out_disc, rc); - } + CDEBUG(D_CONFIG, "Time to connect %s to %s\n", + lmv->cluuid.uuid, obd->obd_name); + + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + tgt = lmv->tgts[i]; + if (tgt == NULL) + continue; + rc = lmv_connect_mdc(obd, tgt); + if (rc) + GOTO(out_disc, rc); + } lmv_set_timeouts(obd); class_export_put(lmv->exp); @@ -596,7 +611,9 @@ int lmv_check_connect(struct obd_device *obd) out_disc: while (i-- > 0) { int rc2; - --tgt; + tgt = lmv->tgts[i]; + if (tgt == NULL) + continue; tgt->ltd_active = 0; if (tgt->ltd_exp) { --lmv->desc.ld_active_tgt_count; @@ -649,9 +666,9 @@ static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt) } } #endif - rc = obd_fid_fini(tgt->ltd_exp); - if (rc) - CERROR("Can't finanize fids factory\n"); + rc = obd_fid_fini(tgt->ltd_exp->exp_obd); + if (rc) + CERROR("Can't finanize fids factory\n"); CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n", tgt->ltd_exp->exp_obd->obd_name, @@ -673,14 +690,14 @@ static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt) static int lmv_disconnect(struct obd_export *exp) { - struct obd_device *obd = class_exp2obd(exp); + struct obd_device *obd = class_exp2obd(exp); #ifdef __KERNEL__ - struct proc_dir_entry *lmv_proc_dir; + struct proc_dir_entry *lmv_proc_dir; #endif - struct lmv_obd *lmv = &obd->u.lmv; - int rc; - int i; - ENTRY; + struct lmv_obd *lmv = &obd->u.lmv; + int rc; + __u32 i; + ENTRY; if (!lmv->tgts) goto out_local; @@ -693,9 +710,10 @@ static int lmv_disconnect(struct obd_export *exp) goto out_local; for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - if (lmv->tgts[i].ltd_exp == NULL) + if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL) continue; - lmv_disconnect_mdc(obd, &lmv->tgts[i]); + + lmv_disconnect_mdc(obd, lmv->tgts[i]); } #ifdef __KERNEL__ @@ -721,16 +739,248 @@ out_local: RETURN(rc); } +static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg) +{ + struct obd_device *obddev = class_exp2obd(exp); + struct lmv_obd *lmv = &obddev->u.lmv; + struct getinfo_fid2path *gf; + struct lmv_tgt_desc *tgt; + struct getinfo_fid2path *remote_gf = NULL; + int remote_gf_size = 0; + int rc; + + gf = (struct getinfo_fid2path *)karg; + tgt = lmv_find_target(lmv, &gf->gf_fid); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); + +repeat_fid2path: + rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg); + if (rc != 0 && rc != -EREMOTE) + GOTO(out_fid2path, rc); + + /* If remote_gf != NULL, it means just building the + * path on the remote MDT, copy this path segement to gf */ + if (remote_gf != NULL) { + struct getinfo_fid2path *ori_gf; + char *ptr; + + ori_gf = (struct getinfo_fid2path *)karg; + if (strlen(ori_gf->gf_path) + + strlen(gf->gf_path) > ori_gf->gf_pathlen) + GOTO(out_fid2path, rc = -EOVERFLOW); + + ptr = ori_gf->gf_path; + + memmove(ptr + strlen(gf->gf_path) + 1, ptr, + strlen(ori_gf->gf_path)); + + strncpy(ptr, gf->gf_path, strlen(gf->gf_path)); + ptr += strlen(gf->gf_path); + *ptr = '/'; + } + + CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n", + tgt->ltd_exp->exp_obd->obd_name, + gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno, + gf->gf_linkno); + + if (rc == 0) + GOTO(out_fid2path, rc); + + /* sigh, has to go to another MDT to do path building further */ + if (remote_gf == NULL) { + remote_gf_size = sizeof(*remote_gf) + PATH_MAX; + OBD_ALLOC(remote_gf, remote_gf_size); + if (remote_gf == NULL) + GOTO(out_fid2path, rc = -ENOMEM); + remote_gf->gf_pathlen = PATH_MAX; + } + + if (!fid_is_sane(&gf->gf_fid)) { + CERROR("%s: invalid FID "DFID": rc = %d\n", + tgt->ltd_exp->exp_obd->obd_name, + PFID(&gf->gf_fid), -EINVAL); + GOTO(out_fid2path, rc = -EINVAL); + } + + tgt = lmv_find_target(lmv, &gf->gf_fid); + if (IS_ERR(tgt)) + GOTO(out_fid2path, rc = -EINVAL); + + remote_gf->gf_fid = gf->gf_fid; + remote_gf->gf_recno = -1; + remote_gf->gf_linkno = -1; + memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen); + gf = remote_gf; + goto repeat_fid2path; + +out_fid2path: + if (remote_gf != NULL) + OBD_FREE(remote_gf, remote_gf_size); + RETURN(rc); +} + +static int lmv_hsm_req_count(struct lmv_obd *lmv, + const struct hsm_user_request *hur, + const struct lmv_tgt_desc *tgt_mds) +{ + __u32 i; + int nr = 0; + struct lmv_tgt_desc *curr_tgt; + + /* count how many requests must be sent to the given target */ + for (i = 0; i < hur->hur_request.hr_itemcount; i++) { + curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid); + if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) + nr++; + } + return nr; +} + +static void lmv_hsm_req_build(struct lmv_obd *lmv, + struct hsm_user_request *hur_in, + const struct lmv_tgt_desc *tgt_mds, + struct hsm_user_request *hur_out) +{ + __u32 i, nr_out; + struct lmv_tgt_desc *curr_tgt; + + /* build the hsm_user_request for the given target */ + hur_out->hur_request = hur_in->hur_request; + nr_out = 0; + for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) { + curr_tgt = lmv_find_target(lmv, + &hur_in->hur_user_item[i].hui_fid); + if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) { + hur_out->hur_user_item[nr_out] = + hur_in->hur_user_item[i]; + nr_out++; + } + } + hur_out->hur_request.hr_itemcount = nr_out; + memcpy(hur_data(hur_out), hur_data(hur_in), + hur_in->hur_request.hr_data_len); +} + +static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len, + struct lustre_kernelcomm *lk, void *uarg) +{ + __u32 i; + int rc; + struct kkuc_ct_data *kcd = NULL; + ENTRY; + + /* unregister request (call from llapi_hsm_copytool_fini) */ + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + struct lmv_tgt_desc *tgt = lmv->tgts[i]; + + if (tgt == NULL || tgt->ltd_exp == NULL) + continue; + /* best effort: try to clean as much as possible + * (continue on error) */ + obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg); + } + + /* Whatever the result, remove copytool from kuc groups. + * Unreached coordinators will get EPIPE on next requests + * and will unregister automatically. + */ + rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd); + if (kcd != NULL) + OBD_FREE_PTR(kcd); + + RETURN(rc); +} + +static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len, + struct lustre_kernelcomm *lk, void *uarg) +{ + struct file *filp; + __u32 i, j; + int err, rc; + bool any_set = false; + struct kkuc_ct_data *kcd; + ENTRY; + + /* All or nothing: try to register to all MDS. + * In case of failure, unregister from previous MDS, + * except if it because of inactive target. */ + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + struct lmv_tgt_desc *tgt = lmv->tgts[i]; + + if (tgt == NULL || tgt->ltd_exp == NULL) + continue; + err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg); + if (err) { + if (tgt->ltd_active) { + /* permanent error */ + CERROR("%s: iocontrol MDC %s on MDT" + " idx %d cmd %x: err = %d\n", + class_exp2obd(lmv->exp)->obd_name, + tgt->ltd_uuid.uuid, i, cmd, err); + rc = err; + lk->lk_flags |= LK_FLG_STOP; + /* unregister from previous MDS */ + for (j = 0; j < i; j++) { + tgt = lmv->tgts[j]; + if (tgt == NULL || tgt->ltd_exp == NULL) + continue; + obd_iocontrol(cmd, tgt->ltd_exp, len, + lk, uarg); + } + RETURN(rc); + } + /* else: transient error. + * kuc will register to the missing MDT + * when it is back */ + } else { + any_set = true; + } + } + + if (!any_set) + /* no registration done: return error */ + RETURN(-ENOTCONN); + + /* at least one registration done, with no failure */ + filp = fget(lk->lk_wfd); + if (filp == NULL) + RETURN(-EBADF); + + OBD_ALLOC_PTR(kcd); + if (kcd == NULL) { + fput(filp); + RETURN(-ENOMEM); + } + kcd->kcd_magic = KKUC_CT_DATA_MAGIC; + kcd->kcd_uuid = lmv->cluuid; + kcd->kcd_archive = lk->lk_data; + + rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd); + if (rc != 0) { + if (filp != NULL) + fput(filp); + OBD_FREE_PTR(kcd); + } + + RETURN(rc); +} + + + + static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, int len, void *karg, void *uarg) { - struct obd_device *obddev = class_exp2obd(exp); - struct lmv_obd *lmv = &obddev->u.lmv; - int i = 0; - int rc = 0; - int set = 0; - int count = lmv->desc.ld_tgt_count; - ENTRY; + struct obd_device *obddev = class_exp2obd(exp); + struct lmv_obd *lmv = &obddev->u.lmv; + struct lmv_tgt_desc *tgt = NULL; + __u32 i = 0; + int rc = 0; + int set = 0; + __u32 count = lmv->desc.ld_tgt_count; + ENTRY; if (count == 0) RETURN(-ENOTTY); @@ -746,48 +996,50 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, if ((index >= count)) RETURN(-ENODEV); - if (!lmv->tgts[index].ltd_active) - RETURN(-ENODATA); - - mdc_obd = class_exp2obd(lmv->tgts[index].ltd_exp); - if (!mdc_obd) - RETURN(-EINVAL); - - /* copy UUID */ - if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd), - min((int) data->ioc_plen2, - (int) sizeof(struct obd_uuid)))) - RETURN(-EFAULT); - - rc = obd_statfs(mdc_obd, &stat_buf, - cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), - 0); - if (rc) - RETURN(rc); - if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf, - min((int) data->ioc_plen1, - (int) sizeof(stat_buf)))) - RETURN(-EFAULT); - break; + tgt = lmv->tgts[index]; + if (tgt == NULL || !tgt->ltd_active) + RETURN(-ENODATA); + + mdc_obd = class_exp2obd(tgt->ltd_exp); + if (!mdc_obd) + RETURN(-EINVAL); + + /* copy UUID */ + if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd), + min((int) data->ioc_plen2, + (int) sizeof(struct obd_uuid)))) + RETURN(-EFAULT); + + rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf, + cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), + 0); + if (rc) + RETURN(rc); + if (copy_to_user(data->ioc_pbuf1, &stat_buf, + min((int) data->ioc_plen1, + (int) sizeof(stat_buf)))) + RETURN(-EFAULT); + break; } case OBD_IOC_QUOTACTL: { struct if_quotactl *qctl = karg; - struct lmv_tgt_desc *tgt = NULL; struct obd_quotactl *oqctl; - if (qctl->qc_valid == QC_MDTIDX) { - if (qctl->qc_idx < 0 || count <= qctl->qc_idx) - RETURN(-EINVAL); - - tgt = &lmv->tgts[qctl->qc_idx]; - if (!tgt->ltd_exp) - RETURN(-EINVAL); - } else if (qctl->qc_valid == QC_UUID) { - for (i = 0; i < count; i++) { - tgt = &lmv->tgts[i]; - if (!obd_uuid_equals(&tgt->ltd_uuid, - &qctl->obd_uuid)) - continue; + if (qctl->qc_valid == QC_MDTIDX) { + if (count <= qctl->qc_idx) + RETURN(-EINVAL); + + tgt = lmv->tgts[qctl->qc_idx]; + if (tgt == NULL || tgt->ltd_exp == NULL) + RETURN(-EINVAL); + } else if (qctl->qc_valid == QC_UUID) { + for (i = 0; i < count; i++) { + tgt = lmv->tgts[i]; + if (tgt == NULL) + continue; + if (!obd_uuid_equals(&tgt->ltd_uuid, + &qctl->obd_uuid)) + continue; if (tgt->ltd_exp == NULL) RETURN(-EINVAL); @@ -801,7 +1053,7 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, if (i >= count) RETURN(-EAGAIN); - LASSERT(tgt && tgt->ltd_exp); + LASSERT(tgt != NULL && tgt->ltd_exp != NULL); OBD_ALLOC_PTR(oqctl); if (!oqctl) RETURN(-ENOMEM); @@ -823,49 +1075,160 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, if (icc->icc_mdtindex >= count) RETURN(-ENODEV); - rc = obd_iocontrol(cmd, lmv->tgts[icc->icc_mdtindex].ltd_exp, - sizeof(*icc), icc, NULL); - break; - } - case LL_IOC_GET_CONNECT_FLAGS: { - rc = obd_iocontrol(cmd, lmv->tgts[0].ltd_exp, len, karg, uarg); - break; - } - - default : { - for (i = 0; i < count; i++) { - int err; - struct obd_device *mdc_obd; - - if (lmv->tgts[i].ltd_exp == NULL) - continue; - /* ll_umount_begin() sets force flag but for lmv, not - * mdc. Let's pass it through */ - mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp); - mdc_obd->obd_force = obddev->obd_force; - err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, - karg, uarg); - if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) { - RETURN(err); - } else if (err) { - if (lmv->tgts[i].ltd_active) { - CERROR("error: iocontrol MDC %s on MDT" - "idx %d cmd %x: err = %d\n", - lmv->tgts[i].ltd_uuid.uuid, - i, cmd, err); - if (!rc) - rc = err; - } - } else - set = 1; + tgt = lmv->tgts[icc->icc_mdtindex]; + if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) + RETURN(-ENODEV); + rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL); + break; + } + case LL_IOC_GET_CONNECT_FLAGS: { + tgt = lmv->tgts[0]; + if (tgt == NULL || tgt->ltd_exp == NULL) + RETURN(-ENODATA); + rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg); + break; + } + case OBD_IOC_FID2PATH: { + rc = lmv_fid2path(exp, len, karg, uarg); + break; + } + case LL_IOC_HSM_STATE_GET: + case LL_IOC_HSM_STATE_SET: + case LL_IOC_HSM_ACTION: { + struct md_op_data *op_data = karg; + + tgt = lmv_find_target(lmv, &op_data->op_fid1); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); + + if (tgt->ltd_exp == NULL) + RETURN(-EINVAL); + + rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg); + break; + } + case LL_IOC_HSM_PROGRESS: { + const struct hsm_progress_kernel *hpk = karg; + + tgt = lmv_find_target(lmv, &hpk->hpk_fid); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); + rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg); + break; + } + case LL_IOC_HSM_REQUEST: { + struct hsm_user_request *hur = karg; + unsigned int reqcount = hur->hur_request.hr_itemcount; + + if (reqcount == 0) + RETURN(0); + + /* if the request is about a single fid + * or if there is a single MDS, no need to split + * the request. */ + if (reqcount == 1 || count == 1) { + tgt = lmv_find_target(lmv, + &hur->hur_user_item[0].hui_fid); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); + rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg); + } else { + /* split fid list to their respective MDS */ + for (i = 0; i < count; i++) { + unsigned int nr, reqlen; + int rc1; + struct hsm_user_request *req; + + tgt = lmv->tgts[i]; + if (tgt == NULL || tgt->ltd_exp == NULL) + continue; + + nr = lmv_hsm_req_count(lmv, hur, tgt); + if (nr == 0) /* nothing for this MDS */ + continue; + + /* build a request with fids for this MDS */ + reqlen = offsetof(typeof(*hur), + hur_user_item[nr]) + + hur->hur_request.hr_data_len; + OBD_ALLOC_LARGE(req, reqlen); + if (req == NULL) + RETURN(-ENOMEM); + + lmv_hsm_req_build(lmv, hur, tgt, req); + + rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen, + req, uarg); + if (rc1 != 0 && rc == 0) + rc = rc1; + OBD_FREE_LARGE(req, reqlen); + } + } + break; + } + case LL_IOC_LOV_SWAP_LAYOUTS: { + struct md_op_data *op_data = karg; + struct lmv_tgt_desc *tgt1, *tgt2; + + tgt1 = lmv_find_target(lmv, &op_data->op_fid1); + if (IS_ERR(tgt1)) + RETURN(PTR_ERR(tgt1)); + + tgt2 = lmv_find_target(lmv, &op_data->op_fid2); + if (IS_ERR(tgt2)) + RETURN(PTR_ERR(tgt2)); + + if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL)) + RETURN(-EINVAL); + + /* only files on same MDT can have their layouts swapped */ + if (tgt1->ltd_idx != tgt2->ltd_idx) + RETURN(-EPERM); + + rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg); + break; + } + case LL_IOC_HSM_CT_START: { + struct lustre_kernelcomm *lk = karg; + if (lk->lk_flags & LK_FLG_STOP) + rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg); + else + rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg); + break; + } + default: + for (i = 0; i < count; i++) { + struct obd_device *mdc_obd; + int err; + + tgt = lmv->tgts[i]; + if (tgt == NULL || tgt->ltd_exp == NULL) + continue; + /* ll_umount_begin() sets force flag but for lmv, not + * mdc. Let's pass it through */ + mdc_obd = class_exp2obd(tgt->ltd_exp); + mdc_obd->obd_force = obddev->obd_force; + err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg); + if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) { + RETURN(err); + } else if (err) { + if (tgt->ltd_active) { + CERROR("error: iocontrol MDC %s on MDT" + " idx %d cmd %x: err = %d\n", + tgt->ltd_uuid.uuid, i, cmd, err); + if (!rc) + rc = err; + } + } else + set = 1; } if (!set && !rc) rc = -EIO; } - } RETURN(rc); } +#if 0 static int lmv_all_chars_policy(int count, const char *name, int len) { @@ -908,6 +1271,7 @@ static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data, CERROR("Unsupported placement policy %x\n", placement); return -EINVAL; } +#endif /** * This is _inode_ placement policy function (not name). @@ -916,86 +1280,63 @@ static int lmv_placement_policy(struct obd_device *obd, struct md_op_data *op_data, mdsno_t *mds) { - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_object *obj; - int rc; - ENTRY; - - LASSERT(mds != NULL); - - if (lmv->desc.ld_tgt_count == 1) { - *mds = 0; - RETURN(0); - } - - /* - * Allocate new fid on target according to operation type and parent - * home mds. - */ - obj = lmv_object_find(obd, &op_data->op_fid1); - if (obj != NULL || op_data->op_name == NULL || - op_data->op_opc != LUSTRE_OPC_MKDIR) { - /* - * Allocate fid for non-dir or for null name or for case parent - * dir is split. - */ - if (obj) { - lmv_object_put(obj); - - /* - * If we have this flag turned on, and we see that - * parent dir is split, this means, that caller did not - * notice split yet. This is race and we would like to - * let caller know that. - */ - if (op_data->op_bias & MDS_CHECK_SPLIT) - RETURN(-ERESTART); - } - - /* - * Allocate new fid on same mds where parent fid is located and - * where operation will be sent. In case of split dir, ->op_fid1 - * and ->op_mds here will contain fid and mds of slave directory - * object (assigned by caller). - */ - *mds = op_data->op_mds; - rc = 0; - } else { - /* - * Parent directory is not split and we want to create a - * directory in it. Let's calculate where to place it according - * to operation data @op_data. - */ - *mds = lmv_choose_mds(lmv, op_data, lmv->lmv_placement); - rc = 0; - } - - if (rc) { - CERROR("Can't choose MDS, err = %d\n", rc); - } else { - LASSERT(*mds < lmv->desc.ld_tgt_count); - } - - RETURN(rc); + struct lmv_obd *lmv = &obd->u.lmv; + ENTRY; + + LASSERT(mds != NULL); + + if (lmv->desc.ld_tgt_count == 1) { + *mds = 0; + RETURN(0); + } + + /** + * If stripe_offset is provided during setdirstripe + * (setdirstripe -i xx), xx MDS will be choosen. + */ + if (op_data->op_cli_flags & CLI_SET_MEA) { + struct lmv_user_md *lum; + + lum = (struct lmv_user_md *)op_data->op_data; + if (lum->lum_type == LMV_STRIPE_TYPE && + lum->lum_stripe_offset != -1) { + if (lum->lum_stripe_offset >= lmv->desc.ld_tgt_count) { + CERROR("%s: Stripe_offset %d > MDT count %d:" + " rc = %d\n", obd->obd_name, + lum->lum_stripe_offset, + lmv->desc.ld_tgt_count, -ERANGE); + RETURN(-ERANGE); + } + *mds = lum->lum_stripe_offset; + RETURN(0); + } + } + + /* Allocate new fid on target according to operation type and parent + * home mds. */ + *mds = op_data->op_mds; + RETURN(0); } int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, - mdsno_t mds) + mdsno_t mds) { - struct lmv_tgt_desc *tgt; - int rc; - ENTRY; + struct lmv_tgt_desc *tgt; + int rc; + ENTRY; - tgt = lmv_get_target(lmv, mds); + tgt = lmv_get_target(lmv, mds); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); - /* - * New seq alloc and FLD setup should be atomic. Otherwise we may find - * on server that seq in new allocated fid is not yet known. - */ - cfs_down(&tgt->ltd_fid_sem); + /* + * New seq alloc and FLD setup should be atomic. Otherwise we may find + * on server that seq in new allocated fid is not yet known. + */ + mutex_lock(&tgt->ltd_fid_mutex); - if (!tgt->ltd_active) - GOTO(out, rc = -ENODEV); + if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL) + GOTO(out, rc = -ENODEV); /* * Asking underlaying tgt layer to allocate new fid. @@ -1008,7 +1349,7 @@ int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, EXIT; out: - cfs_up(&tgt->ltd_fid_sem); + mutex_unlock(&tgt->ltd_fid_mutex); return rc; } @@ -1017,7 +1358,7 @@ int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid, { struct obd_device *obd = class_exp2obd(exp); struct lmv_obd *lmv = &obd->u.lmv; - mdsno_t mds; + mdsno_t mds = 0; int rc; ENTRY; @@ -1040,24 +1381,12 @@ int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid, RETURN(rc); } -static int lmv_fid_delete(struct obd_export *exp, const struct lu_fid *fid) -{ - ENTRY; - LASSERT(exp != NULL && fid != NULL); - if (lmv_object_delete(exp, fid)) { - CDEBUG(D_INODE, "Object "DFID" is destroyed.\n", - PFID(fid)); - } - RETURN(0); -} - static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg) { struct lmv_obd *lmv = &obd->u.lmv; struct lprocfs_static_vars lvars; struct lmv_desc *desc; int rc; - int i = 0; ENTRY; if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) { @@ -1072,113 +1401,108 @@ static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg) RETURN(-EINVAL); } - lmv->tgts_size = LMV_MAX_TGT_COUNT * sizeof(struct lmv_tgt_desc); - - OBD_ALLOC(lmv->tgts, lmv->tgts_size); - if (lmv->tgts == NULL) - RETURN(-ENOMEM); - - for (i = 0; i < LMV_MAX_TGT_COUNT; i++) { - cfs_sema_init(&lmv->tgts[i].ltd_fid_sem, 1); - lmv->tgts[i].ltd_idx = i; - } - - lmv->datas_size = LMV_MAX_TGT_COUNT * sizeof(struct obd_connect_data); - - OBD_ALLOC(lmv->datas, lmv->datas_size); - if (lmv->datas == NULL) - GOTO(out_free_tgts, rc = -ENOMEM); + OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32); + if (lmv->tgts == NULL) + RETURN(-ENOMEM); + lmv->tgts_size = 32; - obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid); - lmv->desc.ld_tgt_count = 0; - lmv->desc.ld_active_tgt_count = 0; - lmv->max_cookiesize = 0; - lmv->max_def_easize = 0; - lmv->max_easize = 0; - lmv->lmv_placement = PLACEMENT_CHAR_POLICY; + obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid); + lmv->desc.ld_tgt_count = 0; + lmv->desc.ld_active_tgt_count = 0; + lmv->max_cookiesize = 0; + lmv->max_def_easize = 0; + lmv->max_easize = 0; + lmv->lmv_placement = PLACEMENT_CHAR_POLICY; - cfs_spin_lock_init(&lmv->lmv_lock); - cfs_sema_init(&lmv->init_sem, 1); + spin_lock_init(&lmv->lmv_lock); + mutex_init(&lmv->init_mutex); - rc = lmv_object_setup(obd); - if (rc) { - CERROR("Can't setup LMV object manager, error %d.\n", rc); - GOTO(out_free_datas, rc); - } + lprocfs_lmv_init_vars(&lvars); - lprocfs_lmv_init_vars(&lvars); - lprocfs_obd_setup(obd, lvars.obd_vars); + lprocfs_obd_setup(obd, lvars.obd_vars); + lprocfs_alloc_md_stats(obd, 0); #ifdef LPROCFS - { - rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd_status", - 0444, &lmv_proc_target_fops, obd); - if (rc) - CWARN("Error adding target_obd_stats file (%d)\n", rc); - } + { + rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd", + 0444, &lmv_proc_target_fops, obd); + if (rc) + CWARN("%s: error adding LMV target_obd file: rc = %d\n", + obd->obd_name, rc); + } #endif - rc = fld_client_init(&lmv->lmv_fld, obd->obd_name, - LUSTRE_CLI_FLD_HASH_DHT); - if (rc) { - CERROR("Can't init FLD, err %d\n", rc); - GOTO(out_free_datas, rc); - } + rc = fld_client_init(&lmv->lmv_fld, obd->obd_name, + LUSTRE_CLI_FLD_HASH_DHT); + if (rc) { + CERROR("Can't init FLD, err %d\n", rc); + GOTO(out, rc); + } RETURN(0); -out_free_datas: - OBD_FREE(lmv->datas, lmv->datas_size); - lmv->datas = NULL; -out_free_tgts: - OBD_FREE(lmv->tgts, lmv->tgts_size); - lmv->tgts = NULL; +out: return rc; } static int lmv_cleanup(struct obd_device *obd) { - struct lmv_obd *lmv = &obd->u.lmv; - ENTRY; - - fld_client_fini(&lmv->lmv_fld); - lmv_object_cleanup(obd); - OBD_FREE(lmv->datas, lmv->datas_size); - OBD_FREE(lmv->tgts, lmv->tgts_size); + struct lmv_obd *lmv = &obd->u.lmv; + ENTRY; - RETURN(0); + fld_client_fini(&lmv->lmv_fld); + if (lmv->tgts != NULL) { + int i; + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + if (lmv->tgts[i] == NULL) + continue; + lmv_del_target(lmv, i); + } + OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size); + lmv->tgts_size = 0; + } + RETURN(0); } static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf) { - struct lustre_cfg *lcfg = buf; - struct obd_uuid tgt_uuid; - int rc; - ENTRY; - - switch(lcfg->lcfg_command) { - case LCFG_ADD_MDC: - if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(tgt_uuid.uuid)) - GOTO(out, rc = -EINVAL); - - obd_str2uuid(&tgt_uuid, lustre_cfg_string(lcfg, 1)); - rc = lmv_add_target(obd, &tgt_uuid); - GOTO(out, rc); - default: { - CERROR("Unknown command: %d\n", lcfg->lcfg_command); - GOTO(out, rc = -EINVAL); - } - } + struct lustre_cfg *lcfg = buf; + struct obd_uuid obd_uuid; + int gen; + __u32 index; + int rc; + ENTRY; + + switch (lcfg->lcfg_command) { + case LCFG_ADD_MDC: + /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID + * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */ + if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid)) + GOTO(out, rc = -EINVAL); + + obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1)); + + if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1) + GOTO(out, rc = -EINVAL); + if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1) + GOTO(out, rc = -EINVAL); + rc = lmv_add_target(obd, &obd_uuid, index, gen); + GOTO(out, rc); + default: + CERROR("Unknown command: %d\n", lcfg->lcfg_command); + GOTO(out, rc = -EINVAL); + } out: - RETURN(rc); + RETURN(rc); } -static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs, - __u64 max_age, __u32 flags) +static int lmv_statfs(const struct lu_env *env, struct obd_export *exp, + struct obd_statfs *osfs, __u64 max_age, __u32 flags) { - struct lmv_obd *lmv = &obd->u.lmv; - struct obd_statfs *temp; - int rc = 0; - int i; - ENTRY; + struct obd_device *obd = class_exp2obd(exp); + struct lmv_obd *lmv = &obd->u.lmv; + struct obd_statfs *temp; + int rc = 0; + __u32 i; + ENTRY; rc = lmv_check_connect(obd); if (rc) @@ -1189,19 +1513,28 @@ static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs, RETURN(-ENOMEM); for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - if (lmv->tgts[i].ltd_exp == NULL) - continue; - - rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, - max_age, flags); - if (rc) { - CERROR("can't stat MDS #%d (%s), error %d\n", i, - lmv->tgts[i].ltd_exp->exp_obd->obd_name, - rc); - GOTO(out_free_temp, rc); - } - if (i == 0) { - *osfs = *temp; + if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL) + continue; + + rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp, + max_age, flags); + if (rc) { + CERROR("can't stat MDS #%d (%s), error %d\n", i, + lmv->tgts[i]->ltd_exp->exp_obd->obd_name, + rc); + GOTO(out_free_temp, rc); + } + + if (i == 0) { + *osfs = *temp; + /* If the statfs is from mount, it will needs + * retrieve necessary information from MDT0. + * i.e. mount does not need the merged osfs + * from all of MDT. + * And also clients can be mounted as long as + * MDT0 is in service*/ + if (flags & OBD_STATFS_FOR_MDT0) + GOTO(out_free_temp, rc); } else { osfs->os_bavail += temp->os_bavail; osfs->os_blocks += temp->os_blocks; @@ -1229,8 +1562,8 @@ static int lmv_getstatus(struct obd_export *exp, if (rc) RETURN(rc); - rc = md_getstatus(lmv->tgts[0].ltd_exp, fid, pc); - RETURN(rc); + rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc); + RETURN(rc); } static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid, @@ -1291,9 +1624,7 @@ static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data, struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; struct lmv_tgt_desc *tgt; - struct lmv_object *obj; int rc; - int i; ENTRY; rc = lmv_check_connect(obd); @@ -1304,67 +1635,21 @@ static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data, if (IS_ERR(tgt)) RETURN(PTR_ERR(tgt)); - if (op_data->op_valid & OBD_MD_MDTIDX) { - op_data->op_mds = tgt->ltd_idx; - RETURN(0); - } + if (op_data->op_flags & MF_GET_MDT_IDX) { + op_data->op_mds = tgt->ltd_idx; + RETURN(0); + } rc = md_getattr(tgt->ltd_exp, op_data, request); - if (rc) - RETURN(rc); - - obj = lmv_object_find_lock(obd, &op_data->op_fid1); - - CDEBUG(D_INODE, "GETATTR for "DFID" %s\n", PFID(&op_data->op_fid1), - obj ? "(split)" : ""); - - /* - * If object is split, then we loop over all the slaves and gather size - * attribute. In ideal world we would have to gather also mds field from - * all slaves, as object is spread over the cluster and this is - * definitely interesting information and it is not good to loss it, - * but... - */ - if (obj) { - struct mdt_body *body; - - if (*request == NULL) { - lmv_object_put(obj); - RETURN(rc); - } - - body = req_capsule_server_get(&(*request)->rq_pill, - &RMF_MDT_BODY); - LASSERT(body != NULL); - - for (i = 0; i < obj->lo_objcount; i++) { - if (lmv->tgts[i].ltd_exp == NULL) { - CWARN("%s: NULL export for %d\n", - obd->obd_name, i); - continue; - } - - /* - * Skip master object. - */ - if (lu_fid_eq(&obj->lo_fid, &obj->lo_stripes[i].ls_fid)) - continue; - - body->size += obj->lo_stripes[i].ls_size; - } - - lmv_object_put_unlock(obj); - } RETURN(rc); } -static int lmv_change_cbdata(struct obd_export *exp, const struct lu_fid *fid, - ldlm_iterator_t it, void *data) +static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; - int i; + __u32 i; int rc; ENTRY; @@ -1374,15 +1659,18 @@ static int lmv_change_cbdata(struct obd_export *exp, const struct lu_fid *fid, CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid)); - /* - * With CMD every object can have two locks in different namespaces: - * lookup lock in space of mds storing direntry and update/open lock in - * space of mds storing inode. - */ - for (i = 0; i < lmv->desc.ld_tgt_count; i++) - md_change_cbdata(lmv->tgts[i].ltd_exp, fid, it, data); + /* + * With DNE every object can have two locks in different namespaces: + * lookup lock in space of MDT storing direntry and update/open lock in + * space of MDT storing inode. + */ + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL) + continue; + md_null_inode(lmv->tgts[i]->ltd_exp, fid); + } - RETURN(0); + RETURN(0); } static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid, @@ -1390,7 +1678,7 @@ static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid, { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; - int i; + __u32 i; int rc; ENTRY; @@ -1400,18 +1688,20 @@ static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid, CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid)); - /* - * With CMD every object can have two locks in different namespaces: - * lookup lock in space of mds storing direntry and update/open lock in - * space of mds storing inode. - */ - for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - rc = md_find_cbdata(lmv->tgts[i].ltd_exp, fid, it, data); - if (rc) - RETURN(rc); - } + /* + * With DNE every object can have two locks in different namespaces: + * lookup lock in space of MDT storing direntry and update/open lock in + * space of MDT storing inode. + */ + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL) + continue; + rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data); + if (rc) + RETURN(rc); + } - RETURN(rc); + RETURN(rc); } @@ -1437,160 +1727,66 @@ static int lmv_close(struct obd_export *exp, struct md_op_data *op_data, RETURN(rc); } -/** - * Called in the case MDS returns -ERESTART on create on open, what means that - * directory is split and its LMV presentation object has to be updated. - */ -int lmv_handle_split(struct obd_export *exp, const struct lu_fid *fid) +struct lmv_tgt_desc +*lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data, + struct lu_fid *fid) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct ptlrpc_request *req = NULL; - struct lmv_tgt_desc *tgt; - struct lmv_object *obj; - struct lustre_md md; - struct md_op_data *op_data; - int mealen; - int rc; - __u64 valid; - ENTRY; + struct lmv_tgt_desc *tgt; - md.mea = NULL; - mealen = lmv_get_easize(lmv); + tgt = lmv_find_target(lmv, fid); + if (IS_ERR(tgt)) + return tgt; - valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA; + op_data->op_mds = tgt->ltd_idx; - tgt = lmv_find_target(lmv, fid); - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); + return tgt; +} - /* - * Time to update mea of parent fid. - */ +int lmv_create(struct obd_export *exp, struct md_op_data *op_data, + const void *data, int datalen, int mode, __u32 uid, + __u32 gid, cfs_cap_t cap_effective, __u64 rdev, + struct ptlrpc_request **request) +{ + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt; + int rc; + ENTRY; - OBD_ALLOC_PTR(op_data); - if (op_data == NULL) - RETURN(-ENOMEM); + rc = lmv_check_connect(obd); + if (rc) + RETURN(rc); - op_data->op_fid1 = *fid; - op_data->op_mode = mealen; - op_data->op_valid = valid; + if (!lmv->desc.ld_active_tgt_count) + RETURN(-EIO); - rc = md_getattr(tgt->ltd_exp, op_data, &req); - OBD_FREE_PTR(op_data); - if (rc) { - CERROR("md_getattr() failed, error %d\n", rc); - GOTO(cleanup, rc); - } + tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); - rc = md_get_lustre_md(tgt->ltd_exp, req, NULL, exp, &md); - if (rc) { - CERROR("md_get_lustre_md() failed, error %d\n", rc); - GOTO(cleanup, rc); - } + rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data); + if (rc) + RETURN(rc); - if (md.mea == NULL) - GOTO(cleanup, rc = -ENODATA); + CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n", + op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1), + op_data->op_mds); - obj = lmv_object_create(exp, fid, md.mea); - if (IS_ERR(obj)) - rc = PTR_ERR(obj); - else - lmv_object_put(obj); + op_data->op_flags |= MF_MDC_CANCEL_FID1; + rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid, + cap_effective, rdev, request); - obd_free_memmd(exp, (void *)&md.mea); - EXIT; -cleanup: - if (req) - ptlrpc_req_finished(req); - return rc; + if (rc == 0) { + if (*request == NULL) + RETURN(rc); + CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2)); + } + RETURN(rc); } -int lmv_create(struct obd_export *exp, struct md_op_data *op_data, - const void *data, int datalen, int mode, __u32 uid, - __u32 gid, cfs_cap_t cap_effective, __u64 rdev, - struct ptlrpc_request **request) -{ - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_tgt_desc *tgt; - struct lmv_object *obj; - int rc; - int loop = 0; - int sidx; - ENTRY; - - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); - - if (!lmv->desc.ld_active_tgt_count) - RETURN(-EIO); -repeat: - ++loop; - LASSERT(loop <= 2); - - obj = lmv_object_find(obd, &op_data->op_fid1); - if (obj) { - sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount, - op_data->op_name, op_data->op_namelen); - op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid; - op_data->op_bias &= ~MDS_CHECK_SPLIT; - op_data->op_mds = obj->lo_stripes[sidx].ls_mds; - tgt = lmv_get_target(lmv, op_data->op_mds); - lmv_object_put(obj); - } else { - tgt = lmv_find_target(lmv, &op_data->op_fid1); - op_data->op_bias |= MDS_CHECK_SPLIT; - op_data->op_mds = tgt->ltd_idx; - } - - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); - - rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data); - if (rc == -ERESTART) - goto repeat; - else if (rc) - RETURN(rc); - - CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n", - op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1), - op_data->op_mds); - - op_data->op_flags |= MF_MDC_CANCEL_FID1; - rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid, - cap_effective, rdev, request); - if (rc == 0) { - if (*request == NULL) - RETURN(rc); - CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2)); - } else if (rc == -ERESTART) { - LASSERT(*request != NULL); - DEBUG_REQ(D_WARNING|D_RPCTRACE, *request, - "Got -ERESTART during create!\n"); - ptlrpc_req_finished(*request); - *request = NULL; - - /* - * Directory got split. Time to update local object and repeat - * the request with proper MDS. - */ - rc = lmv_handle_split(exp, &op_data->op_fid1); - if (rc == 0) { - rc = lmv_allocate_slaves(obd, &op_data->op_fid1, - op_data, &op_data->op_fid2); - if (rc) - RETURN(rc); - goto repeat; - } - } - RETURN(rc); -} - -static int lmv_done_writing(struct obd_export *exp, - struct md_op_data *op_data, - struct md_open_data *mod) +static int lmv_done_writing(struct obd_export *exp, + struct md_op_data *op_data, + struct md_open_data *mod) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; @@ -1611,77 +1807,10 @@ static int lmv_done_writing(struct obd_export *exp, } static int -lmv_enqueue_slaves(struct obd_export *exp, struct ldlm_enqueue_info *einfo, - struct lookup_intent *it, struct md_op_data *op_data, - struct lustre_handle *lockh, void *lmm, int lmmsize) -{ - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_stripe_md *mea = op_data->op_mea1; - struct md_op_data *op_data2; - struct lmv_tgt_desc *tgt; - int i; - int rc = 0; - ENTRY; - - OBD_ALLOC_PTR(op_data2); - if (op_data2 == NULL) - RETURN(-ENOMEM); - - LASSERT(mea != NULL); - for (i = 0; i < mea->mea_count; i++) { - memset(op_data2, 0, sizeof(*op_data2)); - op_data2->op_fid1 = mea->mea_ids[i]; - op_data2->op_bias = 0; - - tgt = lmv_find_target(lmv, &op_data2->op_fid1); - if (IS_ERR(tgt)) - GOTO(cleanup, rc = PTR_ERR(tgt)); - - if (tgt->ltd_exp == NULL) - continue; - - rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data2, - lockh + i, lmm, lmmsize, NULL, 0); - - CDEBUG(D_INODE, "Take lock on slave "DFID" -> %d/%d\n", - PFID(&mea->mea_ids[i]), rc, it->d.lustre.it_status); - - if (rc) - GOTO(cleanup, rc); - - if (it->d.lustre.it_data) { - struct ptlrpc_request *req; - req = (struct ptlrpc_request *)it->d.lustre.it_data; - ptlrpc_req_finished(req); - } - - if (it->d.lustre.it_status) - GOTO(cleanup, rc = it->d.lustre.it_status); - } - - EXIT; -cleanup: - OBD_FREE_PTR(op_data2); - - if (rc != 0) { - /* - * Drop all taken locks. - */ - while (--i >= 0) { - if (lockh[i].cookie) - ldlm_lock_decref(lockh + i, einfo->ei_mode); - lockh[i].cookie = 0; - } - } - return rc; -} - -static int lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo, - struct lookup_intent *it, struct md_op_data *op_data, - struct lustre_handle *lockh, void *lmm, int lmmsize, - int extra_lock_flags) + struct lookup_intent *it, struct md_op_data *op_data, + struct lustre_handle *lockh, void *lmm, int lmmsize, + __u64 extra_lock_flags) { struct ptlrpc_request *req = it->d.lustre.it_data; struct obd_device *obd = exp->exp_obd; @@ -1741,147 +1870,91 @@ static int lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo, struct lookup_intent *it, struct md_op_data *op_data, struct lustre_handle *lockh, void *lmm, int lmmsize, - struct ptlrpc_request **req, int extra_lock_flags) -{ - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_tgt_desc *tgt; - struct lmv_object *obj; - int sidx; - int rc; - ENTRY; + struct ptlrpc_request **req, __u64 extra_lock_flags) +{ + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt; + int rc; + ENTRY; - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); + rc = lmv_check_connect(obd); + if (rc) + RETURN(rc); - CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n", - LL_IT2STR(it), PFID(&op_data->op_fid1)); + CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n", + LL_IT2STR(it), PFID(&op_data->op_fid1)); - if (op_data->op_mea1 && it && it->it_op == IT_UNLINK) { - rc = lmv_enqueue_slaves(exp, einfo, it, op_data, - lockh, lmm, lmmsize); - RETURN(rc); - } + tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); - obj = lmv_object_find(obd, &op_data->op_fid1); - if (obj && op_data->op_namelen) { - sidx = raw_name2idx(obj->lo_hashtype, - obj->lo_objcount, - (char *)op_data->op_name, - op_data->op_namelen); - op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid; - tgt = lmv_get_target(lmv, obj->lo_stripes[sidx].ls_mds); - } else { - tgt = lmv_find_target(lmv, &op_data->op_fid1); - } - if (obj) - lmv_object_put(obj); + CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n", + LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx); - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); + rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh, + lmm, lmmsize, req, extra_lock_flags); - CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n", - LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx); - - rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh, - lmm, lmmsize, req, extra_lock_flags); - - if (rc == 0 && it && it->it_op == IT_OPEN) { - rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh, - lmm, lmmsize, extra_lock_flags); - } - RETURN(rc); + if (rc == 0 && it && it->it_op == IT_OPEN) { + rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh, + lmm, lmmsize, extra_lock_flags); + } + RETURN(rc); } static int lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data, struct ptlrpc_request **request) { - struct ptlrpc_request *req = NULL; - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lu_fid rid = op_data->op_fid1; - struct lmv_tgt_desc *tgt; - struct mdt_body *body; - struct lmv_object *obj; - obd_valid valid = op_data->op_valid; - int rc; - int loop = 0; - int sidx; - ENTRY; - - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); - -repeat: - ++loop; - LASSERT(loop <= 2); - obj = lmv_object_find(obd, &rid); - if (obj) { - sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount, - op_data->op_name, op_data->op_namelen); - rid = obj->lo_stripes[sidx].ls_fid; - tgt = lmv_get_target(lmv, obj->lo_stripes[sidx].ls_mds); - op_data->op_mds = obj->lo_stripes[sidx].ls_mds; - valid &= ~OBD_MD_FLCKSPLIT; - lmv_object_put(obj); - } else { - tgt = lmv_find_target(lmv, &rid); - valid |= OBD_MD_FLCKSPLIT; - op_data->op_mds = tgt->ltd_idx; - } - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); - - CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" - "DFID" -> mds #%d\n", - op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1), - PFID(&rid), tgt->ltd_idx); - - op_data->op_valid = valid; - op_data->op_fid1 = rid; - rc = md_getattr_name(tgt->ltd_exp, op_data, request); - if (rc == 0) { - body = req_capsule_server_get(&(*request)->rq_pill, - &RMF_MDT_BODY); - LASSERT(body != NULL); - - if (body->valid & OBD_MD_MDS) { - rid = body->fid1; - CDEBUG(D_INODE, "Request attrs for "DFID"\n", - PFID(&rid)); - - tgt = lmv_find_target(lmv, &rid); - if (IS_ERR(tgt)) { - ptlrpc_req_finished(*request); - RETURN(PTR_ERR(tgt)); - } - - op_data->op_fid1 = rid; - op_data->op_valid |= OBD_MD_FLCROSSREF; - op_data->op_namelen = 0; - op_data->op_name = NULL; - rc = md_getattr_name(tgt->ltd_exp, op_data, &req); - ptlrpc_req_finished(*request); - *request = req; - } - } else if (rc == -ERESTART) { - LASSERT(*request != NULL); - DEBUG_REQ(D_WARNING|D_RPCTRACE, *request, - "Got -ERESTART during getattr!\n"); - ptlrpc_req_finished(*request); - *request = NULL; - - /* - * Directory got split. Time to update local object and repeat - * the request with proper MDS. - */ - rc = lmv_handle_split(exp, &rid); - if (rc == 0) - goto repeat; - } - RETURN(rc); + struct ptlrpc_request *req = NULL; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt; + struct mdt_body *body; + int rc; + ENTRY; + + rc = lmv_check_connect(obd); + if (rc) + RETURN(rc); + + tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); + + CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n", + op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1), + tgt->ltd_idx); + + rc = md_getattr_name(tgt->ltd_exp, op_data, request); + if (rc != 0) + RETURN(rc); + + body = req_capsule_server_get(&(*request)->rq_pill, + &RMF_MDT_BODY); + LASSERT(body != NULL); + + if (body->valid & OBD_MD_MDS) { + struct lu_fid rid = body->fid1; + CDEBUG(D_INODE, "Request attrs for "DFID"\n", + PFID(&rid)); + + tgt = lmv_find_target(lmv, &rid); + if (IS_ERR(tgt)) { + ptlrpc_req_finished(*request); + RETURN(PTR_ERR(tgt)); + } + + op_data->op_fid1 = rid; + op_data->op_valid |= OBD_MD_FLCROSSREF; + op_data->op_namelen = 0; + op_data->op_name = NULL; + rc = md_getattr_name(tgt->ltd_exp, op_data, &req); + ptlrpc_req_finished(*request); + *request = req; + } + + RETURN(rc); } #define md_op_data_fid(op_data, fl) \ @@ -1891,59 +1964,6 @@ repeat: fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \ NULL) -static int lmv_early_cancel_slaves(struct obd_export *exp, - struct md_op_data *op_data, int op_tgt, - ldlm_mode_t mode, int bits, int flag) -{ - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - ldlm_policy_data_t policy = {{0}}; - struct lu_fid *op_fid; - struct lu_fid *st_fid; - struct lmv_tgt_desc *tgt; - struct lmv_object *obj; - int rc = 0; - int i; - ENTRY; - - op_fid = md_op_data_fid(op_data, flag); - if (!fid_is_sane(op_fid)) - RETURN(0); - - obj = lmv_object_find(obd, op_fid); - if (obj == NULL) - RETURN(-EALREADY); - - policy.l_inodebits.bits = bits; - for (i = 0; i < obj->lo_objcount; i++) { - tgt = lmv_get_target(lmv, obj->lo_stripes[i].ls_mds); - st_fid = &obj->lo_stripes[i].ls_fid; - if (op_tgt != tgt->ltd_idx) { - CDEBUG(D_INODE, "EARLY_CANCEL slave "DFID" -> mds #%d\n", - PFID(st_fid), tgt->ltd_idx); - rc = md_cancel_unused(tgt->ltd_exp, st_fid, &policy, - mode, LCF_ASYNC, NULL); - if (rc) - GOTO(out_put_obj, rc); - } else { - CDEBUG(D_INODE, - "EARLY_CANCEL skip operation target %d on "DFID"\n", - op_tgt, PFID(st_fid)); - /* - * Do not cancel locks for operation target, they will - * be handled later in underlaying layer when calling - * function we run on behalf of. - */ - *op_fid = *st_fid; - op_data->op_flags |= flag; - } - } - EXIT; -out_put_obj: - lmv_object_put(obj); - return rc; -} - static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data, int op_tgt, ldlm_mode_t mode, int bits, int flag) { @@ -1952,38 +1972,30 @@ static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data, struct lmv_obd *lmv = &obd->u.lmv; struct lmv_tgt_desc *tgt; ldlm_policy_data_t policy = {{0}}; - struct lmv_object *obj; int rc = 0; ENTRY; if (!fid_is_sane(fid)) RETURN(0); - obj = lmv_object_find(obd, fid); - if (obj) { - rc = lmv_early_cancel_slaves(exp, op_data, op_tgt, mode, - bits, flag); - lmv_object_put(obj); - } else { - tgt = lmv_find_target(lmv, fid); - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); - - if (tgt->ltd_idx != op_tgt) { - CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid)); - policy.l_inodebits.bits = bits; - rc = md_cancel_unused(tgt->ltd_exp, fid, &policy, - mode, LCF_ASYNC, NULL); - } else { - CDEBUG(D_INODE, - "EARLY_CANCEL skip operation target %d on "DFID"\n", - op_tgt, PFID(fid)); - op_data->op_flags |= flag; - rc = 0; - } + tgt = lmv_find_target(lmv, fid); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); - } - RETURN(rc); + if (tgt->ltd_idx != op_tgt) { + CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid)); + policy.l_inodebits.bits = bits; + rc = md_cancel_unused(tgt->ltd_exp, fid, &policy, + mode, LCF_ASYNC, NULL); + } else { + CDEBUG(D_INODE, + "EARLY_CANCEL skip operation target %d on "DFID"\n", + op_tgt, PFID(fid)); + op_data->op_flags |= flag; + rc = 0; + } + + RETURN(rc); } /* @@ -1993,77 +2005,41 @@ static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data, static int lmv_link(struct obd_export *exp, struct md_op_data *op_data, struct ptlrpc_request **request) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_tgt_desc *tgt; - struct lmv_object *obj; - int rc; - int loop = 0; - mdsno_t mds; - int sidx; - ENTRY; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt; + int rc; + ENTRY; - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); + rc = lmv_check_connect(obd); + if (rc) + RETURN(rc); -repeat: - ++loop; - LASSERT(loop <= 2); - LASSERT(op_data->op_namelen != 0); - - CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n", - PFID(&op_data->op_fid2), op_data->op_namelen, - op_data->op_name, PFID(&op_data->op_fid1)); - - obj = lmv_object_find(obd, &op_data->op_fid2); - if (obj) { - sidx = raw_name2idx(obj->lo_hashtype, - obj->lo_objcount, - op_data->op_name, - op_data->op_namelen); - op_data->op_fid2 = obj->lo_stripes[sidx].ls_fid; - mds = obj->lo_stripes[sidx].ls_mds; - lmv_object_put(obj); - } else { - rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds); - if (rc) - RETURN(rc); - } + LASSERT(op_data->op_namelen != 0); - CDEBUG(D_INODE, "Forward to mds #%x ("DFID")\n", - mds, PFID(&op_data->op_fid1)); + CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n", + PFID(&op_data->op_fid2), op_data->op_namelen, + op_data->op_name, PFID(&op_data->op_fid1)); - op_data->op_fsuid = cfs_curproc_fsuid(); - op_data->op_fsgid = cfs_curproc_fsgid(); - op_data->op_cap = cfs_curproc_cap_pack(); - tgt = lmv_get_target(lmv, mds); + op_data->op_fsuid = current_fsuid(); + op_data->op_fsgid = current_fsgid(); + op_data->op_cap = cfs_curproc_cap_pack(); + tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); - /* - * Cancel UPDATE lock on child (fid1). - */ - op_data->op_flags |= MF_MDC_CANCEL_FID2; - rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX, - MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1); - if (rc == 0) - rc = md_link(tgt->ltd_exp, op_data, request); - if (rc == -ERESTART) { - LASSERT(*request != NULL); - DEBUG_REQ(D_WARNING|D_RPCTRACE, *request, - "Got -ERESTART during link!\n"); - ptlrpc_req_finished(*request); - *request = NULL; + /* + * Cancel UPDATE lock on child (fid1). + */ + op_data->op_flags |= MF_MDC_CANCEL_FID2; + rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX, + MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1); + if (rc != 0) + RETURN(rc); - /* - * Directory got split. Time to update local object and repeat - * the request with proper MDS. - */ - rc = lmv_handle_split(exp, &op_data->op_fid2); - if (rc == 0) - goto repeat; - } + rc = md_link(tgt->ltd_exp, op_data, request); - RETURN(rc); + RETURN(rc); } static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data, @@ -2073,14 +2049,9 @@ static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data, struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; struct lmv_tgt_desc *src_tgt; - struct lmv_tgt_desc *tgt_tgt; - int rc; - int sidx; - int loop = 0; - struct lmv_object *obj; - mdsno_t mds1; - mdsno_t mds2; - ENTRY; + struct lmv_tgt_desc *tgt_tgt; + int rc; + ENTRY; LASSERT(oldlen != 0); @@ -2092,100 +2063,51 @@ static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data, if (rc) RETURN(rc); -repeat: - ++loop; - LASSERT(loop <= 2); - obj = lmv_object_find(obd, &op_data->op_fid1); - if (obj) { - sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount, - (char *)old, oldlen); - op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid; - mds1 = obj->lo_stripes[sidx].ls_mds; - CDEBUG(D_INODE, "Parent obj "DFID"\n", PFID(&op_data->op_fid1)); - lmv_object_put(obj); - } else { - rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds1); - if (rc) - RETURN(rc); - } - - obj = lmv_object_find(obd, &op_data->op_fid2); - if (obj) { - /* - * Directory is already split, so we have to forward request to - * the right MDS. - */ - sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount, - (char *)new, newlen); - - mds2 = obj->lo_stripes[sidx].ls_mds; - op_data->op_fid2 = obj->lo_stripes[sidx].ls_fid; - CDEBUG(D_INODE, "Parent obj "DFID"\n", PFID(&op_data->op_fid2)); - lmv_object_put(obj); - } else { - rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds2); - if (rc) - RETURN(rc); - } - - op_data->op_fsuid = cfs_curproc_fsuid(); - op_data->op_fsgid = cfs_curproc_fsgid(); - op_data->op_cap = cfs_curproc_cap_pack(); - - src_tgt = lmv_get_target(lmv, mds1); - tgt_tgt = lmv_get_target(lmv, mds2); - - /* - * LOOKUP lock on src child (fid3) should also be cancelled for - * src_tgt in mdc_rename. - */ - op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3; - - /* - * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its - * own target. - */ - rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx, - LCK_EX, MDS_INODELOCK_UPDATE, - MF_MDC_CANCEL_FID2); - - /* - * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt. - */ - if (rc == 0) { - rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx, - LCK_EX, MDS_INODELOCK_LOOKUP, - MF_MDC_CANCEL_FID4); - } - - /* - * Cancel all the locks on tgt child (fid4). - */ - if (rc == 0) - rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx, - LCK_EX, MDS_INODELOCK_FULL, - MF_MDC_CANCEL_FID4); - - if (rc == 0) - rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, - new, newlen, request); - - if (rc == -ERESTART) { - LASSERT(*request != NULL); - DEBUG_REQ(D_WARNING|D_RPCTRACE, *request, - "Got -ERESTART during rename!\n"); - ptlrpc_req_finished(*request); - *request = NULL; - - /* - * Directory got split. Time to update local object and repeat - * the request with proper MDS. - */ - rc = lmv_handle_split(exp, &op_data->op_fid1); - if (rc == 0) - goto repeat; - } - RETURN(rc); + op_data->op_fsuid = current_fsuid(); + op_data->op_fsgid = current_fsgid(); + op_data->op_cap = cfs_curproc_cap_pack(); + src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1); + if (IS_ERR(src_tgt)) + RETURN(PTR_ERR(src_tgt)); + + tgt_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2); + if (IS_ERR(tgt_tgt)) + RETURN(PTR_ERR(tgt_tgt)); + /* + * LOOKUP lock on src child (fid3) should also be cancelled for + * src_tgt in mdc_rename. + */ + op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3; + + /* + * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its + * own target. + */ + rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx, + LCK_EX, MDS_INODELOCK_UPDATE, + MF_MDC_CANCEL_FID2); + + /* + * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt. + */ + if (rc == 0) { + rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx, + LCK_EX, MDS_INODELOCK_LOOKUP, + MF_MDC_CANCEL_FID4); + } + + /* + * Cancel all the locks on tgt child (fid4). + */ + if (rc == 0) + rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx, + LCK_EX, MDS_INODELOCK_FULL, + MF_MDC_CANCEL_FID4); + + if (rc == 0) + rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, + new, newlen, request); + RETURN(rc); } static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data, @@ -2193,439 +2115,281 @@ static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data, struct ptlrpc_request **request, struct md_open_data **mod) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct ptlrpc_request *req; - struct lmv_tgt_desc *tgt; - struct lmv_object *obj; - int rc = 0; - int i; - ENTRY; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt; + int rc = 0; + ENTRY; - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); + rc = lmv_check_connect(obd); + if (rc) + RETURN(rc); - obj = lmv_object_find(obd, &op_data->op_fid1); + CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n", + PFID(&op_data->op_fid1), op_data->op_attr.ia_valid); - CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x%s\n", - PFID(&op_data->op_fid1), op_data->op_attr.ia_valid, - obj ? ", split" : ""); + op_data->op_flags |= MF_MDC_CANCEL_FID1; + tgt = lmv_find_target(lmv, &op_data->op_fid1); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); - op_data->op_flags |= MF_MDC_CANCEL_FID1; - if (obj) { - for (i = 0; i < obj->lo_objcount; i++) { - op_data->op_fid1 = obj->lo_stripes[i].ls_fid; + rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2, + ea2len, request, mod); - tgt = lmv_get_target(lmv, obj->lo_stripes[i].ls_mds); - if (IS_ERR(tgt)) { - rc = PTR_ERR(tgt); - break; - } - - rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, - ea2, ea2len, &req, mod); - - if (lu_fid_eq(&obj->lo_fid, &obj->lo_stripes[i].ls_fid)) { - /* - * This is master object and this request should - * be returned back to llite. - */ - *request = req; - } else { - ptlrpc_req_finished(req); - } - - if (rc) - break; - } - lmv_object_put(obj); - } else { - tgt = lmv_find_target(lmv, &op_data->op_fid1); - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); - - rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2, - ea2len, request, mod); - } - RETURN(rc); + RETURN(rc); } -static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid, - struct obd_capa *oc, struct ptlrpc_request **request) +static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid, + struct obd_capa *oc, struct ptlrpc_request **request) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_tgt_desc *tgt; - int rc; - ENTRY; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt; + int rc; + ENTRY; - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); + rc = lmv_check_connect(obd); + if (rc != 0) + RETURN(rc); - tgt = lmv_find_target(lmv, fid); - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); + tgt = lmv_find_target(lmv, fid); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); - rc = md_sync(tgt->ltd_exp, fid, oc, request); - RETURN(rc); + rc = md_fsync(tgt->ltd_exp, fid, oc, request); + RETURN(rc); } -/** - * Main purpose of LMV blocking ast is to remove split directory LMV - * presentation object (struct lmv_object) attached to the lock being revoked. +/* + * Adjust a set of pages, each page containing an array of lu_dirpages, + * so that each page can be used as a single logical lu_dirpage. + * + * A lu_dirpage is laid out as follows, where s = ldp_hash_start, + * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a + * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end + * value is used as a cookie to request the next lu_dirpage in a + * directory listing that spans multiple pages (two in this example): + * ________ + * | | + * .|--------v------- -----. + * |s|e|f|p|ent|ent| ... |ent| + * '--|-------------- -----' Each CFS_PAGE contains a single + * '------. lu_dirpage. + * .---------v------- -----. + * |s|e|f|p|ent| 0 | ... | 0 | + * '----------------- -----' + * + * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is + * larger than LU_PAGE_SIZE, a single host page may contain multiple + * lu_dirpages. After reading the lu_dirpages from the MDS, the + * ldp_hash_end of the first lu_dirpage refers to the one immediately + * after it in the same CFS_PAGE (arrows simplified for brevity, but + * in general e0==s1, e1==s2, etc.): + * + * .-------------------- -----. + * |s0|e0|f0|p|ent|ent| ... |ent| + * |---v---------------- -----| + * |s1|e1|f1|p|ent|ent| ... |ent| + * |---v---------------- -----| Here, each CFS_PAGE contains + * ... multiple lu_dirpages. + * |---v---------------- -----| + * |s'|e'|f'|p|ent|ent| ... |ent| + * '---|---------------- -----' + * v + * .----------------------------. + * | next CFS_PAGE | + * + * This structure is transformed into a single logical lu_dirpage as follows: + * + * - Replace e0 with e' so the request for the next lu_dirpage gets the page + * labeled 'next CFS_PAGE'. + * + * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether + * a hash collision with the next page exists. + * + * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span + * to the first entry of the next lu_dirpage. */ -int lmv_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, - void *data, int flag) -{ - struct lustre_handle lockh; - struct lmv_object *obj; - int rc; - ENTRY; - - switch (flag) { - case LDLM_CB_BLOCKING: - ldlm_lock2handle(lock, &lockh); - rc = ldlm_cli_cancel(&lockh); - if (rc < 0) { - CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc); - RETURN(rc); - } - break; - case LDLM_CB_CANCELING: - /* - * Time to drop cached attrs for split directory object - */ - obj = lock->l_ast_data; - if (obj) { - CDEBUG(D_INODE, "Cancel %s on "LPU64"/"LPU64 - ", master "DFID"\n", - lock->l_resource->lr_name.name[3] == 1 ? - "LOOKUP" : "UPDATE", - lock->l_resource->lr_name.name[0], - lock->l_resource->lr_name.name[1], - PFID(&obj->lo_fid)); - lmv_object_put(obj); - } - break; - default: - LBUG(); - } - RETURN(0); -} - -static void lmv_hash_adjust(__u64 *hash, __u64 hash_adj) -{ - __u64 val; - - val = le64_to_cpu(*hash); - if (val < hash_adj) - val += MAX_HASH_SIZE; - if (val != MDS_DIR_END_OFF) - *hash = cpu_to_le64(val - hash_adj); -} - -static __u32 lmv_node_rank(struct obd_export *exp, const struct lu_fid *fid) -{ - __u64 id; - struct obd_import *imp; - - /* - * XXX: to get nid we assume that underlying obd device is mdc. - */ - imp = class_exp2cliimp(exp); - id = imp->imp_connection->c_self + fid_flatten(fid); - - CDEBUG(D_INODE, "Readpage node rank: "LPX64" "DFID" "LPX64" "LPX64"\n", - imp->imp_connection->c_self, PFID(fid), id, id ^ (id >> 32)); - - return id ^ (id >> 32); +#if PAGE_CACHE_SIZE > LU_PAGE_SIZE +static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs) +{ + int i; + + for (i = 0; i < ncfspgs; i++) { + struct lu_dirpage *dp = kmap(pages[i]); + struct lu_dirpage *first = dp; + struct lu_dirent *end_dirent = NULL; + struct lu_dirent *ent; + __u64 hash_end = dp->ldp_hash_end; + __u32 flags = dp->ldp_flags; + + while (--nlupgs > 0) { + ent = lu_dirent_start(dp); + for (end_dirent = ent; ent != NULL; + end_dirent = ent, ent = lu_dirent_next(ent)); + + /* Advance dp to next lu_dirpage. */ + dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE); + + /* Check if we've reached the end of the CFS_PAGE. */ + if (!((unsigned long)dp & ~CFS_PAGE_MASK)) + break; + + /* Save the hash and flags of this lu_dirpage. */ + hash_end = dp->ldp_hash_end; + flags = dp->ldp_flags; + + /* Check if lu_dirpage contains no entries. */ + if (!end_dirent) + break; + + /* Enlarge the end entry lde_reclen from 0 to + * first entry of next lu_dirpage. */ + LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0); + end_dirent->lde_reclen = + cpu_to_le16((char *)(dp->ldp_entries) - + (char *)end_dirent); + } + + first->ldp_hash_end = hash_end; + first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE); + first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE); + + kunmap(pages[i]); + } + LASSERTF(nlupgs == 0, "left = %d", nlupgs); } +#else +#define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0) +#endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */ -static int lmv_readpage(struct obd_export *exp, const struct lu_fid *fid, - struct obd_capa *oc, __u64 offset64, - struct page **pages, unsigned npages, - struct ptlrpc_request **request) +static int lmv_readpage(struct obd_export *exp, struct md_op_data *op_data, + struct page **pages, struct ptlrpc_request **request) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lu_fid rid = *fid; - struct lmv_object *obj; - __u64 offset; - __u64 hash_adj = 0; - __u32 rank = 0; - __u64 seg_size = 0; - __u64 tgt_tmp = 0; - int tgt_idx = 0; - int tgt0_idx = 0; - int rc; - int nr = 0; - int i; - /* number of pages read, in CFS_PAGE_SIZE */ - int nrdpgs; - /* number of pages transferred in LU_PAGE_SIZE */ - int nlupgs; - struct lmv_stripe *los; - struct lmv_tgt_desc *tgt; - struct lu_dirpage *dp; - struct lu_dirent *ent; - ENTRY; - - offset = offset64; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + __u64 offset = op_data->op_offset; + int rc; + int ncfspgs; /* pages read in PAGE_CACHE_SIZE */ + int nlupgs; /* pages read in LU_PAGE_SIZE */ + struct lmv_tgt_desc *tgt; + ENTRY; - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); + rc = lmv_check_connect(obd); + if (rc) + RETURN(rc); - CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n", offset, PFID(&rid)); - - /* - * This case handle directory lookup in clustered metadata case (i.e. - * split directory is located on multiple md servers.) - * each server keeps directory entries for certain range of hashes. - * E.g. we have N server and suppose hash range is 0 to MAX_HASH. - * first server will keep records with hashes [ 0 ... MAX_HASH / N - 1], - * second one with hashes [MAX_HASH / N ... 2 * MAX_HASH / N] and - * so on.... - * readdir can simply start reading entries from 0 - N server in - * order but that will not scale well as all client will request dir in - * to server in same order. - * Following algorithm does optimization: - * Instead of doing readdir in 1, 2, ...., N order, client with a - * rank R does readdir in R, R + 1, ..., N, 1, ... R - 1 order. - * (every client has rank R) - * But ll_readdir() expect offset range [0 to MAX_HASH/N) but - * since client ask dir from MDS{R} client has pages with offsets - * [R*MAX_HASH/N ... (R + 1)*MAX_HASH/N] there for we do hash_adj - * on hash values that we get. - */ - obj = lmv_object_find_lock(obd, fid); - if (obj) { - nr = obj->lo_objcount; - LASSERT(nr > 0); - seg_size = MAX_HASH_SIZE; - do_div(seg_size, nr); - los = obj->lo_stripes; - tgt = lmv_get_target(lmv, los[0].ls_mds); - rank = lmv_node_rank(tgt->ltd_exp, fid) % nr; - tgt_tmp = offset; - do_div(tgt_tmp, seg_size); - tgt0_idx = do_div(tgt_tmp, nr); - tgt_idx = (tgt0_idx + rank) % nr; - - if (tgt_idx < tgt0_idx) - /* - * Wrap around. - * - * Last segment has unusual length due to division - * rounding. - */ - hash_adj = MAX_HASH_SIZE - seg_size * nr; - else - hash_adj = 0; - - hash_adj += rank * seg_size; - - CDEBUG(D_INODE, "Readpage hash adjustment: %x "LPX64" " - LPX64"/%x -> "LPX64"/%x\n", rank, hash_adj, - offset, tgt0_idx, offset + hash_adj, tgt_idx); - - offset = (offset + hash_adj) & MAX_HASH_SIZE; - rid = obj->lo_stripes[tgt_idx].ls_fid; - tgt = lmv_get_target(lmv, los[tgt_idx].ls_mds); - - CDEBUG(D_INODE, "Forward to "DFID" with offset %lu i %d\n", - PFID(&rid), (unsigned long)offset, tgt_idx); - } else - tgt = lmv_find_target(lmv, &rid); + CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n", + offset, PFID(&op_data->op_fid1)); - if (IS_ERR(tgt)) - GOTO(cleanup, rc = PTR_ERR(tgt)); + tgt = lmv_find_target(lmv, &op_data->op_fid1); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); - rc = md_readpage(tgt->ltd_exp, &rid, oc, offset, pages, npages, - request); - if (rc) - GOTO(cleanup, rc); + rc = md_readpage(tgt->ltd_exp, op_data, pages, request); + if (rc != 0) + RETURN(rc); - nrdpgs = ((*request)->rq_bulk->bd_nob_transferred + CFS_PAGE_SIZE - 1) - >> CFS_PAGE_SHIFT; - nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT; - LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK)); - LASSERT(nrdpgs > 0 && nrdpgs <= npages); + ncfspgs = ((*request)->rq_bulk->bd_nob_transferred + + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; + nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT; + LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK)); + LASSERT(ncfspgs > 0 && ncfspgs <= op_data->op_npages); - CDEBUG(D_INODE, "read %d(%d)/%d pages\n", nrdpgs, nlupgs, npages); + CDEBUG(D_INODE, "read %d(%d)/%d pages\n", ncfspgs, nlupgs, + op_data->op_npages); - for (i = 0; i < nrdpgs; i++) { -#if CFS_PAGE_SIZE > LU_PAGE_SIZE - struct lu_dirpage *first; - __u64 hash_end = 0; - __u32 flags = 0; -#endif - struct lu_dirent *tmp = NULL; - - dp = cfs_kmap(pages[i]); - if (obj) { - lmv_hash_adjust(&dp->ldp_hash_start, hash_adj); - lmv_hash_adjust(&dp->ldp_hash_end, hash_adj); - LASSERT(le64_to_cpu(dp->ldp_hash_start) <= offset64); - - if ((tgt0_idx != nr - 1) && - (le64_to_cpu(dp->ldp_hash_end) == MDS_DIR_END_OFF)) - { - dp->ldp_hash_end = cpu_to_le32(seg_size * - (tgt0_idx + 1)); - CDEBUG(D_INODE, - ""DFID" reset end "LPX64" tgt %d\n", - PFID(&rid), - (__u64)le64_to_cpu(dp->ldp_hash_end), - tgt_idx); - } - } + lmv_adjust_dirpages(pages, ncfspgs, nlupgs); - ent = lu_dirent_start(dp); -#if CFS_PAGE_SIZE > LU_PAGE_SIZE - first = dp; - hash_end = dp->ldp_hash_end; -repeat: -#endif - nlupgs--; - for (tmp = ent; ent != NULL; - tmp = ent, ent = lu_dirent_next(ent)) { - if (obj) - lmv_hash_adjust(&ent->lde_hash, hash_adj); - } - -#if CFS_PAGE_SIZE > LU_PAGE_SIZE - dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE); - if (((unsigned long)dp & ~CFS_PAGE_MASK) && nlupgs > 0) { - ent = lu_dirent_start(dp); - - if (obj) { - lmv_hash_adjust(&dp->ldp_hash_end, hash_adj); - if ((tgt0_idx != nr - 1) && - (le64_to_cpu(dp->ldp_hash_end) == - MDS_DIR_END_OFF)) { - hash_end = cpu_to_le32(seg_size * - (tgt0_idx + 1)); - CDEBUG(D_INODE, - ""DFID" reset end "LPX64" tgt %d\n", - PFID(&rid), - (__u64)le64_to_cpu(hash_end), - tgt_idx); - } - } - hash_end = dp->ldp_hash_end; - flags = dp->ldp_flags; - - if (tmp) { - /* enlarge the end entry lde_reclen from 0 to - * first entry of next lu_dirpage, in this way - * several lu_dirpages can be stored into one - * client page on client. */ - tmp = ((void *)tmp) + - le16_to_cpu(tmp->lde_reclen); - tmp->lde_reclen = - cpu_to_le16((char *)(dp->ldp_entries) - - (char *)tmp); - goto repeat; - } - } - first->ldp_hash_end = hash_end; - first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE); - first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE); -#endif - cfs_kunmap(pages[i]); - } - EXIT; -cleanup: - if (obj) - lmv_object_put_unlock(obj); - return rc; + RETURN(rc); } static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data, struct ptlrpc_request **request) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_tgt_desc *tgt = NULL; - struct lmv_object *obj; - int rc; - int sidx; - int loop = 0; - ENTRY; - - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); - -repeat: - ++loop; - LASSERT(loop <= 2); - LASSERT(op_data->op_namelen != 0); - - obj = lmv_object_find(obd, &op_data->op_fid1); - if (obj) { - sidx = raw_name2idx(obj->lo_hashtype, - obj->lo_objcount, - op_data->op_name, - op_data->op_namelen); - op_data->op_bias &= ~MDS_CHECK_SPLIT; - op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid; - tgt = lmv_get_target(lmv, - obj->lo_stripes[sidx].ls_mds); - lmv_object_put(obj); - CDEBUG(D_INODE, "UNLINK '%*s' in "DFID" -> %u\n", - op_data->op_namelen, op_data->op_name, - PFID(&op_data->op_fid1), sidx); - } - - if (tgt == NULL) { - tgt = lmv_find_target(lmv, &op_data->op_fid1); - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); - op_data->op_bias |= MDS_CHECK_SPLIT; - } - - op_data->op_fsuid = cfs_curproc_fsuid(); - op_data->op_fsgid = cfs_curproc_fsgid(); - op_data->op_cap = cfs_curproc_cap_pack(); - - /* - * If child's fid is given, cancel unused locks for it if it is from - * another export than parent. - * - * LOOKUP lock for child (fid3) should also be cancelled on parent - * tgt_tgt in mdc_unlink(). - */ - op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3; - - /* - * Cancel FULL locks on child (fid3). - */ - rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX, - MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3); - - if (rc == 0) - rc = md_unlink(tgt->ltd_exp, op_data, request); - - if (rc == -ERESTART) { - LASSERT(*request != NULL); - DEBUG_REQ(D_WARNING|D_RPCTRACE, *request, - "Got -ERESTART during unlink!\n"); - ptlrpc_req_finished(*request); - *request = NULL; - - /* - * Directory got split. Time to update local object and repeat - * the request with proper MDS. - */ - rc = lmv_handle_split(exp, &op_data->op_fid1); - if (rc == 0) - goto repeat; - } - RETURN(rc); + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt = NULL; + struct mdt_body *body; + int rc; + ENTRY; + + rc = lmv_check_connect(obd); + if (rc) + RETURN(rc); +retry: + /* Send unlink requests to the MDT where the child is located */ + if (likely(!fid_is_zero(&op_data->op_fid2))) + tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2); + else + tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); + + op_data->op_fsuid = current_fsuid(); + op_data->op_fsgid = current_fsgid(); + op_data->op_cap = cfs_curproc_cap_pack(); + + /* + * If child's fid is given, cancel unused locks for it if it is from + * another export than parent. + * + * LOOKUP lock for child (fid3) should also be cancelled on parent + * tgt_tgt in mdc_unlink(). + */ + op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3; + + /* + * Cancel FULL locks on child (fid3). + */ + rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX, + MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3); + + if (rc != 0) + RETURN(rc); + + CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n", + PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx); + + rc = md_unlink(tgt->ltd_exp, op_data, request); + if (rc != 0 && rc != -EREMOTE) + RETURN(rc); + + body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY); + if (body == NULL) + RETURN(-EPROTO); + + /* Not cross-ref case, just get out of here. */ + if (likely(!(body->valid & OBD_MD_MDS))) + RETURN(0); + + CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n", + exp->exp_obd->obd_name, PFID(&body->fid1)); + + /* This is a remote object, try remote MDT, Note: it may + * try more than 1 time here, Considering following case + * /mnt/lustre is root on MDT0, remote1 is on MDT1 + * 1. Initially A does not know where remote1 is, it send + * unlink RPC to MDT0, MDT0 return -EREMOTE, it will + * resend unlink RPC to MDT1 (retry 1st time). + * + * 2. During the unlink RPC in flight, + * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2 + * and create new remote1, but on MDT0 + * + * 3. MDT1 get unlink RPC(from A), then do remote lock on + * /mnt/lustre, then lookup get fid of remote1, and find + * it is remote dir again, and replay -EREMOTE again. + * + * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times). + * + * In theory, it might try unlimited time here, but it should + * be very rare case. */ + op_data->op_fid2 = body->fid1; + ptlrpc_req_finished(*request); + *request = NULL; + + goto retry; } static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) @@ -2641,9 +2405,7 @@ static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) case OBD_CLEANUP_EXPORTS: fld_client_proc_fini(&lmv->lmv_fld); lprocfs_obd_cleanup(obd); - rc = obd_llog_finish(obd, 0); - if (rc != 0) - CERROR("failed to cleanup llogging subsystems\n"); + lprocfs_free_md_stats(obd); break; default: break; @@ -2651,8 +2413,8 @@ static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) RETURN(rc); } -static int lmv_get_info(struct obd_export *exp, __u32 keylen, - void *key, __u32 *vallen, void *val, +static int lmv_get_info(const struct lu_env *env, struct obd_export *exp, + __u32 keylen, void *key, __u32 *vallen, void *val, struct lov_stripe_md *lsm) { struct obd_device *obd; @@ -2669,7 +2431,6 @@ static int lmv_get_info(struct obd_export *exp, __u32 keylen, lmv = &obd->u.lmv; if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) { - struct lmv_tgt_desc *tgts; int i; rc = lmv_check_connect(obd); @@ -2677,20 +2438,17 @@ static int lmv_get_info(struct obd_export *exp, __u32 keylen, RETURN(rc); LASSERT(*vallen == sizeof(__u32)); - for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; - i++, tgts++) { - - /* - * All tgts should be connected when this gets called. - */ - if (!tgts || !tgts->ltd_exp) { - CERROR("target not setup?\n"); - continue; - } - - if (!obd_get_info(tgts->ltd_exp, keylen, key, - vallen, val, NULL)) - RETURN(0); + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + struct lmv_tgt_desc *tgt = lmv->tgts[i]; + /* + * All tgts should be connected when this gets called. + */ + if (tgt == NULL || tgt->ltd_exp == NULL) + continue; + + if (!obd_get_info(env, tgt->ltd_exp, keylen, key, + vallen, val, NULL)) + RETURN(0); } RETURN(-EINVAL); } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) { @@ -2698,16 +2456,14 @@ static int lmv_get_info(struct obd_export *exp, __u32 keylen, if (rc) RETURN(rc); - /* - * Forwarding this request to first MDS, it should know LOV - * desc. - */ - rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key, - vallen, val, NULL); - if (!rc && KEY_IS(KEY_CONN_DATA)) { - exp->exp_connect_flags = - ((struct obd_connect_data *)val)->ocd_connect_flags; - } + /* + * Forwarding this request to first MDS, it should know LOV + * desc. + */ + rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key, + vallen, val, NULL); + if (!rc && KEY_IS(KEY_CONN_DATA)) + exp->exp_connect_data = *(struct obd_connect_data *)val; RETURN(rc); } else if (KEY_IS(KEY_TGT_COUNT)) { *((int *)val) = lmv->desc.ld_tgt_count; @@ -2718,9 +2474,9 @@ static int lmv_get_info(struct obd_export *exp, __u32 keylen, RETURN(-EINVAL); } -int lmv_set_info_async(struct obd_export *exp, obd_count keylen, - void *key, obd_count vallen, void *val, - struct ptlrpc_request_set *set) +int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp, + obd_count keylen, void *key, obd_count vallen, + void *val, struct ptlrpc_request_set *set) { struct lmv_tgt_desc *tgt; struct obd_device *obd; @@ -2739,13 +2495,13 @@ int lmv_set_info_async(struct obd_export *exp, obd_count keylen, if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) { int i, err = 0; - for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - tgt = &lmv->tgts[i]; + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + tgt = lmv->tgts[i]; - if (!tgt->ltd_exp) - continue; + if (tgt == NULL || tgt->ltd_exp == NULL) + continue; - err = obd_set_info_async(tgt->ltd_exp, + err = obd_set_info_async(env, tgt->ltd_exp, keylen, key, vallen, val, set); if (err && rc == 0) rc = err; @@ -2765,7 +2521,7 @@ int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp, struct lmv_stripe_md *meap; struct lmv_stripe_md *lsmp; int mea_size; - int i; + __u32 i; ENTRY; mea_size = lmv_get_easize(lmv); @@ -2798,10 +2554,10 @@ int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp, meap->mea_count = cpu_to_le32(lsmp->mea_count); meap->mea_master = cpu_to_le32(lsmp->mea_master); - for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - meap->mea_ids[i] = meap->mea_ids[i]; - fid_cpu_to_le(&meap->mea_ids[i], &meap->mea_ids[i]); - } + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + meap->mea_ids[i] = lsmp->mea_ids[i]; + fid_cpu_to_le(&meap->mea_ids[i], &lsmp->mea_ids[i]); + } RETURN(mea_size); } @@ -2814,7 +2570,7 @@ int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp, struct lmv_stripe_md *mea = (struct lmv_stripe_md *)lmm; struct lmv_obd *lmv = &obd->u.lmv; int mea_size; - int i; + __u32 i; __u32 magic; ENTRY; @@ -2869,36 +2625,40 @@ static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid, struct lmv_obd *lmv = &obd->u.lmv; int rc = 0; int err; - int i; + __u32 i; ENTRY; LASSERT(fid != NULL); - for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].ltd_active) - continue; + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + struct lmv_tgt_desc *tgt = lmv->tgts[i]; - err = md_cancel_unused(lmv->tgts[i].ltd_exp, fid, - policy, mode, flags, opaque); - if (!rc) - rc = err; - } - RETURN(rc); + if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) + continue; + + err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags, + opaque); + if (!rc) + rc = err; + } + RETURN(rc); } int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data, __u64 *bits) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - int rc; - ENTRY; + struct lmv_obd *lmv = &exp->exp_obd->u.lmv; + struct lmv_tgt_desc *tgt = lmv->tgts[0]; + int rc; + ENTRY; - rc = md_set_lock_data(lmv->tgts[0].ltd_exp, lockh, data, bits); - RETURN(rc); + if (tgt == NULL || tgt->ltd_exp == NULL) + RETURN(-EINVAL); + rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits); + RETURN(rc); } -ldlm_mode_t lmv_lock_match(struct obd_export *exp, int flags, +ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags, const struct lu_fid *fid, ldlm_type_t type, ldlm_policy_data_t *policy, ldlm_mode_t mode, struct lustre_handle *lockh) @@ -2906,7 +2666,7 @@ ldlm_mode_t lmv_lock_match(struct obd_export *exp, int flags, struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; ldlm_mode_t rc; - int i; + __u32 i; ENTRY; CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid)); @@ -2918,8 +2678,13 @@ ldlm_mode_t lmv_lock_match(struct obd_export *exp, int flags, * one fid was created in. */ for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - rc = md_lock_match(lmv->tgts[i].ltd_exp, flags, fid, - type, policy, mode, lockh); + struct lmv_tgt_desc *tgt = lmv->tgts[i]; + + if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) + continue; + + rc = md_lock_match(tgt->ltd_exp, flags, fid, type, policy, mode, + lockh); if (rc) RETURN(rc); } @@ -2928,26 +2693,29 @@ ldlm_mode_t lmv_lock_match(struct obd_export *exp, int flags, } int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req, - struct obd_export *dt_exp, struct obd_export *md_exp, - struct lustre_md *md) + struct obd_export *dt_exp, struct obd_export *md_exp, + struct lustre_md *md) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - int rc; - ENTRY; - rc = md_get_lustre_md(lmv->tgts[0].ltd_exp, req, dt_exp, md_exp, md); - RETURN(rc); + struct lmv_obd *lmv = &exp->exp_obd->u.lmv; + struct lmv_tgt_desc *tgt = lmv->tgts[0]; + + if (tgt == NULL || tgt->ltd_exp == NULL) + RETURN(-EINVAL); + return md_get_lustre_md(tgt->ltd_exp, req, dt_exp, md_exp, md); } int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - ENTRY; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt = lmv->tgts[0]; + ENTRY; - if (md->mea) - obd_free_memmd(exp, (void *)&md->mea); - RETURN(md_free_lustre_md(lmv->tgts[0].ltd_exp, md)); + if (md->mea) + obd_free_memmd(exp, (void *)&md->mea); + if (tgt == NULL || tgt->ltd_exp == NULL) + RETURN(-EINVAL); + RETURN(md_free_lustre_md(tgt->ltd_exp, md)); } int lmv_set_open_replay_data(struct obd_export *exp, @@ -3026,54 +2794,37 @@ static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc, } int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req, - const struct req_msg_field *field, struct obd_capa **oc) + const struct req_msg_field *field, struct obd_capa **oc) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - int rc; + struct lmv_obd *lmv = &exp->exp_obd->u.lmv; + struct lmv_tgt_desc *tgt = lmv->tgts[0]; - ENTRY; - rc = md_unpack_capa(lmv->tgts[0].ltd_exp, req, field, oc); - RETURN(rc); + if (tgt == NULL || tgt->ltd_exp == NULL) + RETURN(-EINVAL); + return md_unpack_capa(tgt->ltd_exp, req, field, oc); } int lmv_intent_getattr_async(struct obd_export *exp, struct md_enqueue_info *minfo, struct ldlm_enqueue_info *einfo) { - struct md_op_data *op_data = &minfo->mi_data; - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_object *obj; - struct lmv_tgt_desc *tgt = NULL; - int rc; - int sidx; - ENTRY; - - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); + struct md_op_data *op_data = &minfo->mi_data; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt = NULL; + int rc; + ENTRY; - if (op_data->op_namelen) { - obj = lmv_object_find(obd, &op_data->op_fid1); - if (obj) { - sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount, - (char *)op_data->op_name, - op_data->op_namelen); - op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid; - tgt = lmv_get_target(lmv, obj->lo_stripes[sidx].ls_mds); - lmv_object_put(obj); - } - } + rc = lmv_check_connect(obd); + if (rc) + RETURN(rc); - if (tgt == NULL) - tgt = lmv_find_target(lmv, &op_data->op_fid1); - - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); + tgt = lmv_find_target(lmv, &op_data->op_fid1); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); - rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo); - RETURN(rc); + rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo); + RETURN(rc); } int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it, @@ -3097,6 +2848,84 @@ int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it, RETURN(rc); } +/** + * For lmv, only need to send request to master MDT, and the master MDT will + * process with other slave MDTs. The only exception is Q_GETOQUOTA for which + * we directly fetch data from the slave MDTs. + */ +int lmv_quotactl(struct obd_device *unused, struct obd_export *exp, + struct obd_quotactl *oqctl) +{ + struct obd_device *obd = class_exp2obd(exp); + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt = lmv->tgts[0]; + int rc = 0; + __u32 i; + __u64 curspace, curinodes; + ENTRY; + + if (tgt == NULL || + tgt->ltd_exp == NULL || + !tgt->ltd_active || + lmv->desc.ld_tgt_count == 0) { + CERROR("master lmv inactive\n"); + RETURN(-EIO); + } + + if (oqctl->qc_cmd != Q_GETOQUOTA) { + rc = obd_quotactl(tgt->ltd_exp, oqctl); + RETURN(rc); + } + + curspace = curinodes = 0; + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + int err; + tgt = lmv->tgts[i]; + + if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) + continue; + + err = obd_quotactl(tgt->ltd_exp, oqctl); + if (err) { + CERROR("getquota on mdt %d failed. %d\n", i, err); + if (!rc) + rc = err; + } else { + curspace += oqctl->qc_dqblk.dqb_curspace; + curinodes += oqctl->qc_dqblk.dqb_curinodes; + } + } + oqctl->qc_dqblk.dqb_curspace = curspace; + oqctl->qc_dqblk.dqb_curinodes = curinodes; + + RETURN(rc); +} + +int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp, + struct obd_quotactl *oqctl) +{ + struct obd_device *obd = class_exp2obd(exp); + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt; + __u32 i; + int rc = 0; + ENTRY; + + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + int err; + tgt = lmv->tgts[i]; + if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) { + CERROR("lmv idx %d inactive\n", i); + RETURN(-EIO); + } + + err = obd_quotacheck(tgt->ltd_exp, oqctl); + if (err && !rc) + rc = err; + } + + RETURN(rc); +} struct obd_ops lmv_obd_ops = { .o_owner = THIS_MODULE, @@ -3114,12 +2943,13 @@ struct obd_ops lmv_obd_ops = { .o_notify = lmv_notify, .o_get_uuid = lmv_get_uuid, .o_iocontrol = lmv_iocontrol, - .o_fid_delete = lmv_fid_delete + .o_quotacheck = lmv_quotacheck, + .o_quotactl = lmv_quotactl }; struct md_ops lmv_md_ops = { .m_getstatus = lmv_getstatus, - .m_change_cbdata = lmv_change_cbdata, + .m_null_inode = lmv_null_inode, .m_find_cbdata = lmv_find_cbdata, .m_close = lmv_close, .m_create = lmv_create, @@ -3133,7 +2963,7 @@ struct md_ops lmv_md_ops = { .m_rename = lmv_rename, .m_setattr = lmv_setattr, .m_setxattr = lmv_setxattr, - .m_sync = lmv_sync, + .m_fsync = lmv_fsync, .m_readpage = lmv_readpage, .m_unlink = lmv_unlink, .m_init_ea_size = lmv_init_ea_size, @@ -3151,51 +2981,22 @@ struct md_ops lmv_md_ops = { .m_revalidate_lock = lmv_revalidate_lock }; -static quota_interface_t *quota_interface; -extern quota_interface_t lmv_quota_interface; - int __init lmv_init(void) { struct lprocfs_static_vars lvars; int rc; - lmv_object_cache = cfs_mem_cache_create("lmv_objects", - sizeof(struct lmv_object), - 0, 0); - if (!lmv_object_cache) { - CERROR("Error allocating lmv objects cache\n"); - return -ENOMEM; - } - lprocfs_lmv_init_vars(&lvars); - cfs_request_module("lquota"); - quota_interface = PORTAL_SYMBOL_GET(lmv_quota_interface); - init_obd_quota_ops(quota_interface, &lmv_obd_ops); - rc = class_register_type(&lmv_obd_ops, &lmv_md_ops, lvars.module_vars, LUSTRE_LMV_NAME, NULL); - if (rc) { - if (quota_interface) - PORTAL_SYMBOL_PUT(lmv_quota_interface); - cfs_mem_cache_destroy(lmv_object_cache); - } - return rc; } #ifdef __KERNEL__ static void lmv_exit(void) { - if (quota_interface) - PORTAL_SYMBOL_PUT(lmv_quota_interface); - class_unregister_type(LUSTRE_LMV_NAME); - - LASSERTF(cfs_atomic_read(&lmv_object_count) == 0, - "Can't free lmv objects cache, %d object(s) busy\n", - cfs_atomic_read(&lmv_object_count)); - cfs_mem_cache_destroy(lmv_object_cache); } MODULE_AUTHOR("Sun Microsystems, Inc. ");