X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Flov%2Flov_obd.c;h=135458d85ca8b8a9a3a0cbbee89a43a8b2facc49;hp=a3a45637d96feedeb0b6214e168eb4d16573fb98;hb=b3f5156084edcebb714bb6d77fca4152fe3b045f;hpb=f2b264f2049554d69c1cc6f0c0ebc5d3e196a068 diff --git a/lustre/lov/lov_obd.c b/lustre/lov/lov_obd.c index a3a4563..135458d 100644 --- a/lustre/lov/lov_obd.c +++ b/lustre/lov/lov_obd.c @@ -1,29 +1,44 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * Copyright (C) 2002-2006 Cluster File Systems, Inc. - * Author: Phil Schwan - * Peter Braam - * Mike Shaver - * Nathan Rutman + * GPL HEADER START + * + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 only, + * as published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). * - * This file is part of the Lustre file system, http://www.lustre.org - * Lustre is a trademark of Cluster File Systems, Inc. + * You should have received a copy of the GNU General Public License + * version 2 along with this program; If not, see + * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf * - * You may have signed or agreed to another license before downloading - * this software. If so, you are bound by the terms and conditions - * of that agreement, and the following does not apply to you. See the - * LICENSE file included with this distribution for more information. + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. * - * If you did not agree to a different license, then this copy of Lustre - * is open source software; you can redistribute it and/or modify it - * under the terms of version 2 of the GNU General Public License as - * published by the Free Software Foundation. + * GPL HEADER END + */ +/* + * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Use is subject to license terms. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. * - * In either case, Lustre is distributed in the hope that it will be - * useful, but WITHOUT ANY WARRANTY; without even the implied warranty - * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * license text for more details. + * lustre/lov/lov_obd.c + * + * Author: Phil Schwan + * Author: Peter Braam + * Author: Mike Shaver + * Author: Nathan Rutman */ #ifndef EXPORT_SYMTAB @@ -48,54 +63,77 @@ #include #include #include +#include +#include #include "lov_internal.h" - /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion. Any function that expects lov_tgts to remain stationary must take a ref. */ -void lov_getref(struct obd_device *obd) +static void lov_getref(struct obd_device *obd) { struct lov_obd *lov = &obd->u.lov; /* nobody gets through here until lov_putref is done */ - mutex_down(&lov->lov_lock); - atomic_inc(&lov->lov_refcount); - mutex_up(&lov->lov_lock); + cfs_mutex_down(&lov->lov_lock); + cfs_atomic_inc(&lov->lov_refcount); + cfs_mutex_up(&lov->lov_lock); return; } -static void __lov_del_obd(struct obd_device *obd, __u32 index); +static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt); -void lov_putref(struct obd_device *obd) +static void lov_putref(struct obd_device *obd) { struct lov_obd *lov = &obd->u.lov; - mutex_down(&lov->lov_lock); + + cfs_mutex_down(&lov->lov_lock); /* ok to dec to 0 more than once -- ltd_exp's will be null */ - if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) { + if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) { + CFS_LIST_HEAD(kill); int i; + struct lov_tgt_desc *tgt, *n; CDEBUG(D_CONFIG, "destroying %d lov targets\n", lov->lov_death_row); for (i = 0; i < lov->desc.ld_tgt_count; i++) { - if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_reap) + tgt = lov->lov_tgts[i]; + + if (!tgt || !tgt->ltd_reap) continue; - /* Disconnect and delete from list */ - __lov_del_obd(obd, i); + cfs_list_add(&tgt->ltd_kill, &kill); + /* XXX - right now there is a dependency on ld_tgt_count + * being the maximum tgt index for computing the + * mds_max_easize. So we can't shrink it. */ + lov_ost_pool_remove(&lov->lov_packed, i); + lov->lov_tgts[i] = NULL; lov->lov_death_row--; } + cfs_mutex_up(&lov->lov_lock); + + cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) { + cfs_list_del(&tgt->ltd_kill); + /* Disconnect */ + __lov_del_obd(obd, tgt); + } + } else { + cfs_mutex_up(&lov->lov_lock); } - mutex_up(&lov->lov_lock); } +static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid, + int activate); +static int lov_notify(struct obd_device *obd, struct obd_device *watched, + enum obd_notify_event ev, void *data); + + #define MAX_STRING_SIZE 128 -static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate, - struct obd_connect_data *data) +int lov_connect_obd(struct obd_device *obd, __u32 index, int activate, + struct obd_connect_data *data) { struct lov_obd *lov = &obd->u.lov; - struct obd_uuid tgt_uuid = lov->lov_tgts[index]->ltd_uuid; + struct obd_uuid tgt_uuid; struct obd_device *tgt_obd; - struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" }; - struct lustre_handle conn = {0, }; + static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" }; struct obd_import *imp; #ifdef __KERNEL__ @@ -107,18 +145,17 @@ static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate, if (!lov->lov_tgts[index]) RETURN(-EINVAL); - tgt_obd = class_find_client_obd(&tgt_uuid, LUSTRE_OSC_NAME, - &obd->obd_uuid); + tgt_uuid = lov->lov_tgts[index]->ltd_uuid; + tgt_obd = lov->lov_tgts[index]->ltd_obd; - if (!tgt_obd) { - CERROR("Target %s not attached\n", obd_uuid2str(&tgt_uuid)); - RETURN(-EINVAL); - } if (!tgt_obd->obd_set_up) { CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid)); RETURN(-EINVAL); } + /* override the sp_me from lov */ + tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me; + if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX)) data->ocd_index = index; @@ -129,57 +166,42 @@ static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate, if (activate) { tgt_obd->obd_no_recov = 0; - /* FIXME this is probably supposed to be + /* FIXME this is probably supposed to be ptlrpc_set_import_active. Horrible naming. */ ptlrpc_activate_import(imp); } + rc = obd_register_observer(tgt_obd, obd); + if (rc) { + CERROR("Target %s register_observer error %d\n", + obd_uuid2str(&tgt_uuid), rc); + RETURN(rc); + } + + if (imp->imp_invalid) { CERROR("not connecting OSC %s; administratively " "disabled\n", obd_uuid2str(&tgt_uuid)); - rc = obd_register_observer(tgt_obd, obd); - if (rc) { - CERROR("Target %s register_observer error %d; " - "will not be able to reactivate\n", - obd_uuid2str(&tgt_uuid), rc); - } RETURN(0); } - rc = obd_connect(NULL, &conn, tgt_obd, &lov_osc_uuid, data); - if (rc) { + rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd, + &lov_osc_uuid, data, NULL); + if (rc || !lov->lov_tgts[index]->ltd_exp) { CERROR("Target %s connect error %d\n", obd_uuid2str(&tgt_uuid), rc); - RETURN(rc); - } - lov->lov_tgts[index]->ltd_exp = class_conn2export(&conn); - if (!lov->lov_tgts[index]->ltd_exp) { - CERROR("Target %s: null export!\n", obd_uuid2str(&tgt_uuid)); RETURN(-ENODEV); } - rc = obd_register_observer(tgt_obd, obd); - if (rc) { - CERROR("Target %s register_observer error %d\n", - obd_uuid2str(&tgt_uuid), rc); - obd_disconnect(lov->lov_tgts[index]->ltd_exp); - lov->lov_tgts[index]->ltd_exp = NULL; - RETURN(rc); - } - lov->lov_tgts[index]->ltd_reap = 0; - if (activate) { - lov->lov_tgts[index]->ltd_active = 1; - lov->desc.ld_active_tgt_count++; - lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0; - } + CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index, obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in"); #ifdef __KERNEL__ lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds"); if (lov_proc_dir) { - struct obd_device *osc_obd = class_conn2obd(&conn); + struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd; cfs_proc_dir_entry_t *osc_symlink; char name[MAX_STRING_SIZE]; @@ -189,8 +211,8 @@ static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate, snprintf(name, MAX_STRING_SIZE, "../../../%s/%s", osc_obd->obd_type->typ_name, osc_obd->obd_name); - osc_symlink = proc_symlink(osc_obd->obd_name, lov_proc_dir, - name); + osc_symlink = lprocfs_add_symlink(osc_obd->obd_name, lov_proc_dir, + name); if (osc_symlink == NULL) { CERROR("could not register LOV target " "/proc/fs/lustre/%s/%s/target_obds/%s.", @@ -209,20 +231,24 @@ static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate, } static int lov_connect(const struct lu_env *env, - struct lustre_handle *conn, struct obd_device *obd, - struct obd_uuid *cluuid, struct obd_connect_data *data) + struct obd_export **exp, struct obd_device *obd, + struct obd_uuid *cluuid, struct obd_connect_data *data, + void *localdata) { struct lov_obd *lov = &obd->u.lov; struct lov_tgt_desc *tgt; + struct lustre_handle conn; int i, rc; ENTRY; CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects); - rc = class_connect(conn, obd, cluuid); + rc = class_connect(&conn, obd, cluuid); if (rc) RETURN(rc); + *exp = class_conn2export(&conn); + /* Why should there ever be more than 1 connect? */ lov->lov_connects++; LASSERT(lov->lov_connects == 1); @@ -231,7 +257,7 @@ static int lov_connect(const struct lu_env *env, if (data) lov->lov_ocd = *data; - lov_getref(obd); + obd_getref(obd); for (i = 0; i < lov->desc.ld_tgt_count; i++) { tgt = lov->lov_tgts[i]; if (!tgt || obd_uuid_empty(&tgt->ltd_uuid)) @@ -243,29 +269,38 @@ static int lov_connect(const struct lu_env *env, obd->obd_name, i, rc); continue; } + /* connect to administrative disabled ost */ + if (!lov->lov_tgts[i]->ltd_exp) + continue; + + rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd, + OBD_NOTIFY_CONNECT, (void *)&i); + if (rc) { + CERROR("%s error sending notify %d\n", + obd->obd_name, rc); + } } - lov_putref(obd); + obd_putref(obd); RETURN(0); } -static int lov_disconnect_obd(struct obd_device *obd, __u32 index) +static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt) { cfs_proc_dir_entry_t *lov_proc_dir; struct lov_obd *lov = &obd->u.lov; - struct obd_device *osc_obd = - class_exp2obd(lov->lov_tgts[index]->ltd_exp); + struct obd_device *osc_obd; int rc; - ENTRY; + osc_obd = class_exp2obd(tgt->ltd_exp); CDEBUG(D_CONFIG, "%s: disconnecting target %s\n", obd->obd_name, osc_obd->obd_name); - if (lov->lov_tgts[index]->ltd_active) { - lov->lov_tgts[index]->ltd_active = 0; + if (tgt->ltd_active) { + tgt->ltd_active = 0; lov->desc.ld_active_tgt_count--; - lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1; + tgt->ltd_exp->exp_obd->obd_inactive = 1; } lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds"); @@ -294,22 +329,19 @@ static int lov_disconnect_obd(struct obd_device *obd, __u32 index) obd_register_observer(osc_obd, NULL); - rc = obd_disconnect(lov->lov_tgts[index]->ltd_exp); + rc = obd_disconnect(tgt->ltd_exp); if (rc) { CERROR("Target %s disconnect error %d\n", - lov_uuid2str(lov, index), rc); + tgt->ltd_uuid.uuid, rc); rc = 0; } - qos_del_tgt(obd, index); + qos_del_tgt(obd, tgt); - lov->lov_tgts[index]->ltd_exp = NULL; + tgt->ltd_exp = NULL; RETURN(0); } -static int lov_del_target(struct obd_device *obd, __u32 index, - struct obd_uuid *uuidp, int gen); - static int lov_disconnect(struct obd_export *exp) { struct obd_device *obd = class_exp2obd(exp); @@ -330,14 +362,15 @@ static int lov_disconnect(struct obd_export *exp) /* Let's hold another reference so lov_del_obd doesn't spin through putref every time */ - lov_getref(obd); + obd_getref(obd); + for (i = 0; i < lov->desc.ld_tgt_count; i++) { if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) { /* Disconnection is the last we know about an obd */ lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen); } } - lov_putref(obd); + obd_putref(obd); out: rc = class_disconnect(exp); /* bz 9811 */ @@ -349,58 +382,59 @@ out: * -EINVAL : UUID can't be found in the LOV's target list * -ENOTCONN: The UUID is found, but the target connection is bad (!) * -EBADF : The UUID is found, but the OBD is the wrong type (!) + * any >= 0 : is log target index */ static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid, int activate) { struct lov_obd *lov = &obd->u.lov; struct lov_tgt_desc *tgt; - int i, rc = 0; + int index; ENTRY; CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n", lov, uuid->uuid, activate); - lov_getref(obd); - for (i = 0; i < lov->desc.ld_tgt_count; i++) { - tgt = lov->lov_tgts[i]; + obd_getref(obd); + for (index = 0; index < lov->desc.ld_tgt_count; index++) { + tgt = lov->lov_tgts[index]; if (!tgt || !tgt->ltd_exp) continue; CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n", - i, obd_uuid2str(&tgt->ltd_uuid), + index, obd_uuid2str(&tgt->ltd_uuid), tgt->ltd_exp->exp_handle.h_cookie); if (obd_uuid_equals(uuid, &tgt->ltd_uuid)) break; } - if (i == lov->desc.ld_tgt_count) - GOTO(out, rc = -EINVAL); + if (index == lov->desc.ld_tgt_count) + GOTO(out, index = -EINVAL); - if (lov->lov_tgts[i]->ltd_active == activate) { + if (lov->lov_tgts[index]->ltd_active == activate) { CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid, activate ? "" : "in"); - GOTO(out, rc); + GOTO(out, index); } CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid), activate ? "" : "in"); - lov->lov_tgts[i]->ltd_active = activate; + lov->lov_tgts[index]->ltd_active = activate; if (activate) { lov->desc.ld_active_tgt_count++; - lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 0; + lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0; } else { lov->desc.ld_active_tgt_count--; - lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 1; + lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1; } /* remove any old qos penalty */ - lov->lov_tgts[i]->ltd_qos.ltq_penalty = 0; + lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0; out: - lov_putref(obd); - RETURN(rc); + obd_putref(obd); + RETURN(index); } static int lov_notify(struct obd_device *obd, struct obd_device *watched, @@ -426,12 +460,14 @@ static int lov_notify(struct obd_device *obd, struct obd_device *watched, * observer can use the OSC normally. */ rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE); - if (rc) { + if (rc < 0) { CERROR("%sactivation of %s failed: %d\n", (ev == OBD_NOTIFY_ACTIVE) ? "" : "de", obd_uuid2str(uuid), rc); RETURN(rc); } + /* active event should be pass lov target index as data */ + data = &rc; } /* Pass the notification up the chain. */ @@ -439,14 +475,24 @@ static int lov_notify(struct obd_device *obd, struct obd_device *watched, rc = obd_notify_observer(obd, watched, ev, data); } else { /* NULL watched means all osc's in the lov (only for syncs) */ + /* sync event should be send lov idx as data */ struct lov_obd *lov = &obd->u.lov; struct obd_device *tgt_obd; int i; - lov_getref(obd); + obd_getref(obd); for (i = 0; i < lov->desc.ld_tgt_count; i++) { - if (!lov->lov_tgts[i]) - continue; + /* don't send sync event if target not + * connected/activated */ + if (!lov->lov_tgts[i] || + !lov->lov_tgts[i]->ltd_active) + continue; + + if ((ev == OBD_NOTIFY_SYNC) || + (ev == OBD_NOTIFY_SYNC_NONBLOCK)) + data = &i; + tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp); + rc = obd_notify_observer(obd, tgt_obd, ev, data); if (rc) { CERROR("%s: notify %s of %s failed %d\n", @@ -456,17 +502,18 @@ static int lov_notify(struct obd_device *obd, struct obd_device *watched, break; } } - lov_putref(obd); + obd_putref(obd); } RETURN(rc); } -static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp, - __u32 index, int gen, int active) +int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp, + __u32 index, int gen, int active) { struct lov_obd *lov = &obd->u.lov; struct lov_tgt_desc *tgt; + struct obd_device *tgt_obd; int rc; ENTRY; @@ -479,13 +526,18 @@ static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp, RETURN(-EINVAL); } - mutex_down(&lov->lov_lock); + tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME, + &obd->obd_uuid); + if (tgt_obd == NULL) + RETURN(-EINVAL); + + cfs_mutex_down(&lov->lov_lock); if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) { tgt = lov->lov_tgts[index]; CERROR("UUID %s already assigned at LOV target index %d\n", obd_uuid2str(&tgt->ltd_uuid), index); - mutex_up(&lov->lov_lock); + cfs_mutex_up(&lov->lov_lock); RETURN(-EEXIST); } @@ -499,7 +551,7 @@ static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp, newsize = newsize << 1; OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize); if (newtgts == NULL) { - mutex_up(&lov->lov_lock); + cfs_mutex_up(&lov->lov_lock); RETURN(-ENOMEM); } @@ -522,15 +574,22 @@ static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp, lov->lov_tgts, lov->lov_tgt_size); } - OBD_ALLOC_PTR(tgt); if (!tgt) { - mutex_up(&lov->lov_lock); + cfs_mutex_up(&lov->lov_lock); RETURN(-ENOMEM); } + rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size); + if (rc) { + cfs_mutex_up(&lov->lov_lock); + OBD_FREE_PTR(tgt); + RETURN(rc); + } + memset(tgt, 0, sizeof(*tgt)); tgt->ltd_uuid = *uuidp; + tgt->ltd_obd = tgt_obd; /* XXX - add a sanity check on the generation number. */ tgt->ltd_gen = gen; tgt->ltd_index = index; @@ -538,7 +597,8 @@ static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp, lov->lov_tgts[index] = tgt; if (index >= lov->desc.ld_tgt_count) lov->desc.ld_tgt_count = index + 1; - mutex_up(&lov->lov_lock); + + cfs_mutex_up(&lov->lov_lock); CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n", index, tgt->ltd_gen, lov->desc.ld_tgt_count); @@ -550,14 +610,18 @@ static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp, RETURN(0); } - lov_getref(obd); + obd_getref(obd); rc = lov_connect_obd(obd, index, active, &lov->lov_ocd); if (rc) GOTO(out, rc); + /* connect to administrative disabled ost */ + if (!tgt->ltd_exp) + GOTO(out, rc = 0); + rc = lov_notify(obd, tgt->ltd_exp->exp_obd, - active ? OBD_NOTIFY_ACTIVE : OBD_NOTIFY_INACTIVE, + active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE, (void *)&index); out: @@ -566,13 +630,13 @@ out: obd_uuid2str(&tgt->ltd_uuid)); lov_del_target(obd, index, 0, 0); } - lov_putref(obd); + obd_putref(obd); RETURN(rc); } /* Schedule a target for deletion */ -static int lov_del_target(struct obd_device *obd, __u32 index, - struct obd_uuid *uuidp, int gen) +int lov_del_target(struct obd_device *obd, __u32 index, + struct obd_uuid *uuidp, int gen) { struct lov_obd *lov = &obd->u.lov; int count = lov->desc.ld_tgt_count; @@ -585,7 +649,7 @@ static int lov_del_target(struct obd_device *obd, __u32 index, RETURN(-EINVAL); } - lov_getref(obd); + obd_getref(obd); if (!lov->lov_tgts[index]) { CERROR("LOV target at index %d is not setup.\n", index); @@ -606,19 +670,16 @@ static int lov_del_target(struct obd_device *obd, __u32 index, lov->lov_tgts[index]->ltd_reap = 1; lov->lov_death_row++; - /* we really delete it from lov_putref */ + /* we really delete it from obd_putref */ out: - lov_putref(obd); + obd_putref(obd); RETURN(rc); } -/* We are holding lov_lock */ -static void __lov_del_obd(struct obd_device *obd, __u32 index) +static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt) { - struct lov_obd *lov = &obd->u.lov; struct obd_device *osc_obd; - struct lov_tgt_desc *tgt = lov->lov_tgts[index]; LASSERT(tgt); LASSERT(tgt->ltd_reap); @@ -626,17 +687,12 @@ static void __lov_del_obd(struct obd_device *obd, __u32 index) osc_obd = class_exp2obd(tgt->ltd_exp); CDEBUG(D_CONFIG, "Removing tgt %s : %s\n", - lov_uuid2str(lov, index), + tgt->ltd_uuid.uuid, osc_obd ? osc_obd->obd_name : ""); if (tgt->ltd_exp) - lov_disconnect_obd(obd, index); + lov_disconnect_obd(obd, tgt); - /* XXX - right now there is a dependency on ld_tgt_count being the - * maximum tgt index for computing the mds_max_easize. So we can't - * shrink it. */ - - lov->lov_tgts[index] = NULL; OBD_FREE_PTR(tgt); /* Manual cleanup - no cleanup logs to clean up the osc's. We must @@ -646,39 +702,56 @@ static void __lov_del_obd(struct obd_device *obd, __u32 index) class_manual_cleanup(osc_obd); } -void lov_fix_desc(struct lov_desc *desc) +void lov_fix_desc_stripe_size(__u64 *val) { - if (desc->ld_default_stripe_size < PTLRPC_MAX_BRW_SIZE) { + if (*val < PTLRPC_MAX_BRW_SIZE) { LCONSOLE_WARN("Increasing default stripe size to min %u\n", PTLRPC_MAX_BRW_SIZE); - desc->ld_default_stripe_size = PTLRPC_MAX_BRW_SIZE; - } else if (desc->ld_default_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) { - desc->ld_default_stripe_size &= ~(LOV_MIN_STRIPE_SIZE - 1); + *val = PTLRPC_MAX_BRW_SIZE; + } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) { + *val &= ~(LOV_MIN_STRIPE_SIZE - 1); LCONSOLE_WARN("Changing default stripe size to "LPU64" (a " "multiple of %u)\n", - desc->ld_default_stripe_size,LOV_MIN_STRIPE_SIZE); + *val, LOV_MIN_STRIPE_SIZE); } +} - if (desc->ld_default_stripe_count == 0) - desc->ld_default_stripe_count = 1; +void lov_fix_desc_stripe_count(__u32 *val) +{ + if (*val == 0) + *val = 1; +} +void lov_fix_desc_pattern(__u32 *val) +{ /* from lov_setstripe */ - if ((desc->ld_pattern != 0) && - (desc->ld_pattern != LOV_PATTERN_RAID0)) { - LCONSOLE_WARN("Unknown stripe pattern: %#x\n",desc->ld_pattern); - desc->ld_pattern = 0; + if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) { + LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val); + *val = 0; } +} + +void lov_fix_desc_qos_maxage(__u32 *val) +{ + /* fix qos_maxage */ + if (*val == 0) + *val = QOS_DEFAULT_MAXAGE; +} - if (desc->ld_qos_maxage == 0) - desc->ld_qos_maxage = QOS_DEFAULT_MAXAGE; +void lov_fix_desc(struct lov_desc *desc) +{ + lov_fix_desc_stripe_size(&desc->ld_default_stripe_size); + lov_fix_desc_stripe_count(&desc->ld_default_stripe_count); + lov_fix_desc_pattern(&desc->ld_pattern); + lov_fix_desc_qos_maxage(&desc->ld_qos_maxage); } -static int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg) +int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg) { struct lprocfs_static_vars lvars = { 0 }; struct lov_desc *desc; struct lov_obd *lov = &obd->u.lov; - int count; + int rc; ENTRY; if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) { @@ -708,44 +781,57 @@ static int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg) lov_fix_desc(desc); - /* Because of 64-bit divide/mod operations only work with a 32-bit - * divisor in a 32-bit kernel, we cannot support a stripe width - * of 4GB or larger on 32-bit CPUs. */ - count = desc->ld_default_stripe_count; - if ((count > 0 ? count : desc->ld_tgt_count) * - desc->ld_default_stripe_size > 0xffffffff) { - CERROR("LOV: stripe width "LPU64"x%u > 4294967295 bytes\n", - desc->ld_default_stripe_size, count); - RETURN(-EINVAL); - } - desc->ld_active_tgt_count = 0; lov->desc = *desc; lov->lov_tgt_size = 0; - sema_init(&lov->lov_lock, 1); - atomic_set(&lov->lov_refcount, 0); - INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list); - init_rwsem(&lov->lov_qos.lq_rw_sem); + + cfs_sema_init(&lov->lov_lock, 1); + cfs_atomic_set(&lov->lov_refcount, 0); + CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list); + cfs_init_rwsem(&lov->lov_qos.lq_rw_sem); + lov->lov_sp_me = LUSTRE_SP_CLI; lov->lov_qos.lq_dirty = 1; - lov->lov_qos.lq_dirty_rr = 1; + lov->lov_qos.lq_rr.lqr_dirty = 1; lov->lov_qos.lq_reset = 1; /* Default priority is toward free space balance */ lov->lov_qos.lq_prio_free = 232; + /* Default threshold for rr (roughly 17%) */ + lov->lov_qos.lq_threshold_rr = 43; + /* Init statfs fields */ + OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data); + if (NULL == lov->lov_qos.lq_statfs_data) + RETURN(-ENOMEM); + cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq); + + lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS, + HASH_POOLS_MAX_BITS, + &pool_hash_operations, CFS_HASH_REHASH); + CFS_INIT_LIST_HEAD(&lov->lov_pool_list); + lov->lov_pool_count = 0; + rc = lov_ost_pool_init(&lov->lov_packed, 0); + if (rc) + RETURN(rc); + rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0); + if (rc) { + lov_ost_pool_free(&lov->lov_packed); + RETURN(rc); + } lprocfs_lov_init_vars(&lvars); lprocfs_obd_setup(obd, lvars.obd_vars); #ifdef LPROCFS { - cfs_proc_dir_entry_t *entry; + int rc; - entry = create_proc_entry("target_obd", 0444, - obd->obd_proc_entry); - if (entry != NULL) { - entry->proc_fops = &lov_proc_target_fops; - entry->data = obd; - } + rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd", + 0444, &lov_proc_target_fops, obd); + if (rc) + CWARN("Error adding the target_obd file\n"); } #endif + lov->lov_pool_proc_entry = lprocfs_register("pools", + obd->obd_proc_entry, + NULL, NULL); RETURN(0); } @@ -753,11 +839,12 @@ static int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg) static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) { int rc = 0; + struct lov_obd *lov = &obd->u.lov; + ENTRY; switch (stage) { case OBD_CLEANUP_EARLY: { - struct lov_obd *lov = &obd->u.lov; int i; for (i = 0; i < lov->desc.ld_tgt_count; i++) { if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) @@ -768,14 +855,10 @@ static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) break; } case OBD_CLEANUP_EXPORTS: - break; - case OBD_CLEANUP_SELF_EXP: rc = obd_llog_finish(obd, 0); if (rc != 0) CERROR("failed to cleanup llogging subsystems\n"); break; - case OBD_CLEANUP_OBD: - break; } RETURN(rc); } @@ -783,39 +866,54 @@ static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) static int lov_cleanup(struct obd_device *obd) { struct lov_obd *lov = &obd->u.lov; + cfs_list_t *pos, *tmp; + struct pool_desc *pool; + + cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) { + pool = cfs_list_entry(pos, struct pool_desc, pool_list); + /* free pool structs */ + CDEBUG(D_INFO, "delete pool %p\n", pool); + lov_pool_del(obd, pool->pool_name); + } + cfs_hash_destroy(lov->lov_pools_hash_body); + lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool)); + lov_ost_pool_free(&lov->lov_packed); - lprocfs_obd_cleanup(obd); if (lov->lov_tgts) { int i; + obd_getref(obd); for (i = 0; i < lov->desc.ld_tgt_count; i++) { - if (lov->lov_tgts[i]) { - /* Inactive targets may never have connected */ - if (lov->lov_tgts[i]->ltd_active || - atomic_read(&lov->lov_refcount)) - /* We should never get here - these - should have been removed in the - disconnect. */ - CERROR("lov tgt %d not cleaned!" - " deathrow=%d, lovrc=%d\n", - i, lov->lov_death_row, - atomic_read(&lov->lov_refcount)); - lov_del_target(obd, i, 0, 0); - } + if (!lov->lov_tgts[i]) + continue; + + /* Inactive targets may never have connected */ + if (lov->lov_tgts[i]->ltd_active || + cfs_atomic_read(&lov->lov_refcount)) + /* We should never get here - these + should have been removed in the + disconnect. */ + CERROR("lov tgt %d not cleaned!" + " deathrow=%d, lovrc=%d\n", + i, lov->lov_death_row, + cfs_atomic_read(&lov->lov_refcount)); + lov_del_target(obd, i, 0, 0); } + obd_putref(obd); OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) * lov->lov_tgt_size); lov->lov_tgt_size = 0; } - if (lov->lov_qos.lq_rr_size) - OBD_FREE(lov->lov_qos.lq_rr_array, lov->lov_qos.lq_rr_size); + /* clear pools parent proc entry only after all pools is killed */ + lprocfs_obd_cleanup(obd); + OBD_FREE_PTR(lov->lov_qos.lq_statfs_data); RETURN(0); } -static int lov_process_config(struct obd_device *obd, obd_count len, void *buf) +int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg, + __u32 *indexp, int *genp) { - struct lustre_cfg *lcfg = buf; struct obd_uuid obd_uuid; int cmd; int rc = 0; @@ -833,10 +931,12 @@ static int lov_process_config(struct obd_device *obd, obd_count len, void *buf) obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1)); - if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1) + if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1) GOTO(out, rc = -EINVAL); - if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1) + if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1) GOTO(out, rc = -EINVAL); + index = *indexp; + gen = *genp; if (cmd == LCFG_LOV_ADD_OBD) rc = lov_add_target(obd, &obd_uuid, index, gen, 1); else if (cmd == LCFG_LOV_ADD_INA) @@ -856,8 +956,16 @@ static int lov_process_config(struct obd_device *obd, obd_count len, void *buf) rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars, lcfg, obd); + if (rc > 0) + rc = 0; GOTO(out, rc); } + case LCFG_POOL_NEW: + case LCFG_POOL_ADD: + case LCFG_POOL_DEL: + case LCFG_POOL_REM: + GOTO(out, rc); + default: { CERROR("Unknown command: %d\n", lcfg->lcfg_command); GOTO(out, rc = -EINVAL); @@ -869,7 +977,7 @@ out: } #ifndef log2 -#define log2(n) ffz(~(n)) +#define log2(n) cfs_ffz(~(n)) #endif static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa, @@ -891,13 +999,13 @@ static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa, if (tmp_oa == NULL) RETURN(-ENOMEM); - if (src_oa->o_valid & OBD_MD_FLINLINE) { - ost_uuid = (struct obd_uuid *)src_oa->o_inline; + if (oti->oti_ost_uuid) { + ost_uuid = oti->oti_ost_uuid; CDEBUG(D_HA, "clearing orphans only for %s\n", ost_uuid->uuid); } - lov_getref(export->exp_obd); + obd_getref(export->exp_obd); for (i = 0; i < lov->desc.ld_tgt_count; i++) { struct lov_stripe_md obj_md; struct lov_stripe_md *obj_mdp = &obj_md; @@ -927,16 +1035,18 @@ static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa, /* XXX: LOV STACKING: use real "obj_mdp" sub-data */ err = obd_create(lov->lov_tgts[i]->ltd_exp, tmp_oa, &obj_mdp, oti); - if (err) + if (err) { /* This export will be disabled until it is recovered, and then orphan recovery will be completed. */ CERROR("error in orphan recovery on OST idx %d/%d: " "rc = %d\n", i, lov->desc.ld_tgt_count, err); + rc = err; + } if (ost_uuid) break; } - lov_putref(export->exp_obd); + obd_putref(export->exp_obd); OBDO_FREE(tmp_oa); RETURN(rc); @@ -990,8 +1100,7 @@ static int lov_create(struct obd_export *exp, struct obdo *src_oa, struct obd_info oinfo; struct lov_request_set *set = NULL; struct lov_request *req; - struct obd_statfs osfs; - __u64 maxage; + struct l_wait_info lwi = { 0 }; int rc = 0; ENTRY; @@ -1009,48 +1118,63 @@ static int lov_create(struct obd_export *exp, struct obdo *src_oa, if (!lov->desc.ld_active_tgt_count) RETURN(-EIO); + obd_getref(exp->exp_obd); /* Recreate a specific object id at the given OST index */ if ((src_oa->o_valid & OBD_MD_FLFLAGS) && (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) { rc = lov_recreate(exp, src_oa, ea, oti); - RETURN(rc); + GOTO(out, rc); } - maxage = cfs_time_shift_64(-lov->desc.ld_qos_maxage); - obd_statfs_rqset(exp->exp_obd, &osfs, maxage); + /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s, + * later in alloc_qos(), we will wait for those rpcs to complete if + * the osfs age is older than 2 * qos_maxage */ + qos_statfs_update(exp->exp_obd, + cfs_time_shift_64(-lov->desc.ld_qos_maxage) + CFS_HZ, + 0); rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set); if (rc) - RETURN(rc); + GOTO(out, rc); - list_for_each_entry(req, &set->set_list, rq_link) { + cfs_list_for_each_entry(req, &set->set_list, rq_link) { /* XXX: LOV STACKING: use real "obj_mdp" sub-data */ - rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp, - req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti); - lov_update_create_set(set, req, rc); + rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp, + &req->rq_oi, &req->rq_oi.oi_md, oti); } + + /* osc_create have timeout equ obd_timeout/2 so waiting don't be + * longer then this */ + l_wait_event(set->set_waitq, lov_finished_set(set), &lwi); + + /* we not have ptlrpc set for assign set->interpret and should + * be call interpret function himself. calling from cb_create_update + * not permited because lov_fini_create_set can sleep for long time, + * but we must avoid sleeping in ptlrpcd interpret function. */ rc = lov_fini_create_set(set, ea); +out: + obd_putref(exp->exp_obd); RETURN(rc); } #define ASSERT_LSM_MAGIC(lsmp) \ do { \ LASSERT((lsmp) != NULL); \ - LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC || \ - (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n", \ - (lsmp), (lsmp)->lsm_magic); \ + LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \ + (lsmp)->lsm_magic == LOV_MAGIC_V3), \ + "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \ } while (0) static int lov_destroy(struct obd_export *exp, struct obdo *oa, struct lov_stripe_md *lsm, struct obd_trans_info *oti, - struct obd_export *md_exp) + struct obd_export *md_exp, void *capa) { struct lov_request_set *set; struct obd_info oinfo; struct lov_request *req; - struct list_head *pos; + cfs_list_t *pos; struct lov_obd *lov; - int rc = 0, err; + int rc = 0, err = 0; ENTRY; ASSERT_LSM_MAGIC(lsm); @@ -1064,19 +1188,19 @@ static int lov_destroy(struct obd_export *exp, struct obdo *oa, } lov = &exp->exp_obd->u.lov; + obd_getref(exp->exp_obd); rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set); if (rc) - RETURN(rc); + GOTO(out, rc); - list_for_each (pos, &set->set_list) { - int err; - req = list_entry(pos, struct lov_request, rq_link); + cfs_list_for_each (pos, &set->set_list) { + req = cfs_list_entry(pos, struct lov_request, rq_link); if (oa->o_valid & OBD_MD_FLCOOKIE) oti->oti_logcookies = set->set_cookies + req->rq_stripe; err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp, - req->rq_oi.oi_oa, NULL, oti, NULL); + req->rq_oi.oi_oa, NULL, oti, NULL, capa); err = lov_update_common_set(set, req, err); if (err) { CERROR("error: destroying objid "LPX64" subobj " @@ -1093,6 +1217,8 @@ static int lov_destroy(struct obd_export *exp, struct obdo *oa, rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp); } err = lov_fini_destroy_set(set); +out: + obd_putref(exp->exp_obd); RETURN(rc ? rc : err); } @@ -1100,7 +1226,7 @@ static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo) { struct lov_request_set *set; struct lov_request *req; - struct list_head *pos; + cfs_list_t *pos; struct lov_obd *lov; int err = 0, rc = 0; ENTRY; @@ -1117,8 +1243,8 @@ static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo) if (rc) RETURN(rc); - list_for_each (pos, &set->set_list) { - req = list_entry(pos, struct lov_request, rq_link); + cfs_list_for_each (pos, &set->set_list) { + req = cfs_list_entry(pos, struct lov_request, rq_link); CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx " "%u\n", oinfo->oi_oa->o_id, req->rq_stripe, @@ -1161,7 +1287,7 @@ static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo, { struct lov_request_set *lovset; struct lov_obd *lov; - struct list_head *pos; + cfs_list_t *pos; struct lov_request *req; int rc = 0, err; ENTRY; @@ -1182,8 +1308,8 @@ static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo, oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count, oinfo->oi_md->lsm_stripe_size); - list_for_each (pos, &lovset->set_list) { - req = list_entry(pos, struct lov_request, rq_link); + cfs_list_for_each (pos, &lovset->set_list) { + req = cfs_list_entry(pos, struct lov_request, rq_link); CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx " "%u\n", oinfo->oi_oa->o_id, req->rq_stripe, @@ -1199,7 +1325,7 @@ static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo, } } - if (!list_empty(&rqset->set_requests)) { + if (!cfs_list_empty(&rqset->set_requests)) { LASSERT(rc == 0); LASSERT (rqset->set_interpret == NULL); rqset->set_interpret = lov_getattr_interpret; @@ -1218,7 +1344,7 @@ static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo, { struct lov_request_set *set; struct lov_obd *lov; - struct list_head *pos; + cfs_list_t *pos; struct lov_request *req; int err = 0, rc = 0; ENTRY; @@ -1235,15 +1361,15 @@ static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo, OBD_MD_FLMTIME | OBD_MD_FLCTIME | OBD_MD_FLFLAGS | OBD_MD_FLSIZE | OBD_MD_FLGROUP | OBD_MD_FLUID | - OBD_MD_FLGID | OBD_MD_FLINLINE | - OBD_MD_FLFID | OBD_MD_FLGENER))); + OBD_MD_FLGID | OBD_MD_FLFID | + OBD_MD_FLGENER))); lov = &exp->exp_obd->u.lov; rc = lov_prep_setattr_set(exp, oinfo, oti, &set); if (rc) RETURN(rc); - list_for_each (pos, &set->set_list) { - req = list_entry(pos, struct lov_request, rq_link); + cfs_list_for_each (pos, &set->set_list) { + req = cfs_list_entry(pos, struct lov_request, rq_link); rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp, &req->rq_oi, NULL); @@ -1284,7 +1410,7 @@ static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo, { struct lov_request_set *set; struct lov_request *req; - struct list_head *pos; + cfs_list_t *pos; struct lov_obd *lov; int rc = 0; ENTRY; @@ -1308,8 +1434,8 @@ static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo, oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count, oinfo->oi_md->lsm_stripe_size); - list_for_each (pos, &set->set_list) { - req = list_entry(pos, struct lov_request, rq_link); + cfs_list_for_each (pos, &set->set_list) { + req = cfs_list_entry(pos, struct lov_request, rq_link); if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) oti->oti_logcookies = set->set_cookies + req->rq_stripe; @@ -1331,7 +1457,7 @@ static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo, } /* If we are not waiting for responses on async requests, return. */ - if (rc || !rqset || list_empty(&rqset->set_requests)) { + if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) { int err; if (rc) set->set_completes = 0; @@ -1368,7 +1494,7 @@ static int lov_punch(struct obd_export *exp, struct obd_info *oinfo, { struct lov_request_set *set; struct lov_obd *lov; - struct list_head *pos; + cfs_list_t *pos; struct lov_request *req; int rc = 0; ENTRY; @@ -1384,8 +1510,8 @@ static int lov_punch(struct obd_export *exp, struct obd_info *oinfo, if (rc) RETURN(rc); - list_for_each (pos, &set->set_list) { - req = list_entry(pos, struct lov_request, rq_link); + cfs_list_for_each (pos, &set->set_list) { + req = cfs_list_entry(pos, struct lov_request, rq_link); rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp, &req->rq_oi, NULL, rqset); @@ -1398,7 +1524,7 @@ static int lov_punch(struct obd_export *exp, struct obd_info *oinfo, } } - if (rc || list_empty(&rqset->set_requests)) { + if (rc || cfs_list_empty(&rqset->set_requests)) { int err; err = lov_fini_punch_set(set); RETURN(rc ? rc : err); @@ -1418,7 +1544,7 @@ static int lov_sync(struct obd_export *exp, struct obdo *oa, struct lov_request_set *set; struct obd_info oinfo; struct lov_obd *lov; - struct list_head *pos; + cfs_list_t *pos; struct lov_request *req; int err = 0, rc = 0; ENTRY; @@ -1433,8 +1559,8 @@ static int lov_sync(struct obd_export *exp, struct obdo *oa, if (rc) RETURN(rc); - list_for_each (pos, &set->set_list) { - req = list_entry(pos, struct lov_request, rq_link); + cfs_list_for_each (pos, &set->set_list) { + req = cfs_list_entry(pos, struct lov_request, rq_link); rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp, req->rq_oi.oi_oa, NULL, @@ -1472,7 +1598,7 @@ static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo, obd_off start, end; if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off, - pga[i].off + pga[i].count, + pga[i].off + pga[i].count - 1, &start, &end)) continue; @@ -1495,7 +1621,7 @@ static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo, { struct lov_request_set *set; struct lov_request *req; - struct list_head *pos; + cfs_list_t *pos; struct lov_obd *lov = &exp->exp_obd->u.lov; int err, rc = 0; ENTRY; @@ -1511,10 +1637,10 @@ static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo, if (rc) RETURN(rc); - list_for_each (pos, &set->set_list) { + cfs_list_for_each (pos, &set->set_list) { struct obd_export *sub_exp; struct brw_page *sub_pga; - req = list_entry(pos, struct lov_request, rq_link); + req = cfs_list_entry(pos, struct lov_request, rq_link); sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp; sub_pga = set->set_pga + req->rq_pgaidx; @@ -1531,478 +1657,124 @@ static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo, RETURN(rc); } -static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data, - int rc) +static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset, + void *data, int rc) { struct lov_request_set *lovset = (struct lov_request_set *)data; ENTRY; - - if (rc) { - lovset->set_completes = 0; - lov_fini_brw_set(lovset); - } else { - rc = lov_fini_brw_set(lovset); - } - + rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset); RETURN(rc); } -static int lov_brw_async(int cmd, struct obd_export *exp, - struct obd_info *oinfo, obd_count oa_bufs, - struct brw_page *pga, struct obd_trans_info *oti, - struct ptlrpc_request_set *set) +static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo, + struct ldlm_enqueue_info *einfo, + struct ptlrpc_request_set *rqset) { - struct lov_request_set *lovset; + ldlm_mode_t mode = einfo->ei_mode; + struct lov_request_set *set; struct lov_request *req; - struct list_head *pos; - struct lov_obd *lov = &exp->exp_obd->u.lov; - int rc = 0; + cfs_list_t *pos; + struct lov_obd *lov; + ldlm_error_t rc; ENTRY; LASSERT(oinfo); ASSERT_LSM_MAGIC(oinfo->oi_md); + LASSERT(mode == (mode & -mode)); - if (cmd == OBD_BRW_CHECK) { - rc = lov_brw_check(lov, oinfo, oa_bufs, pga); - RETURN(rc); - } + /* we should never be asked to replay a lock this way. */ + LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0); + + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); - rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset); + lov = &exp->exp_obd->u.lov; + rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set); if (rc) RETURN(rc); - list_for_each (pos, &lovset->set_list) { - struct obd_export *sub_exp; - struct brw_page *sub_pga; - req = list_entry(pos, struct lov_request, rq_link); + cfs_list_for_each (pos, &set->set_list) { + req = cfs_list_entry(pos, struct lov_request, rq_link); - sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp; - sub_pga = lovset->set_pga + req->rq_pgaidx; - rc = obd_brw_async(cmd, sub_exp, &req->rq_oi, req->rq_oabufs, - sub_pga, oti, set); - if (rc) + rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp, + &req->rq_oi, einfo, rqset); + if (rc != ELDLM_OK) GOTO(out, rc); - lov_update_common_set(lovset, req, rc); } - LASSERT(rc == 0); - LASSERT(set->set_interpret == NULL); - set->set_interpret = (set_interpreter_func)lov_brw_interpret; - set->set_arg = (void *)lovset; - RETURN(rc); + if (rqset && !cfs_list_empty(&rqset->set_requests)) { + LASSERT(rc == 0); + LASSERT(rqset->set_interpret == NULL); + rqset->set_interpret = lov_enqueue_interpret; + rqset->set_arg = (void *)set; + RETURN(rc); + } out: - lov_fini_brw_set(lovset); + rc = lov_fini_enqueue_set(set, mode, rc, rqset); RETURN(rc); } -static int lov_ap_make_ready(void *data, int cmd) -{ - struct lov_async_page *lap = LAP_FROM_COOKIE(data); - - return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd); -} - -static int lov_ap_refresh_count(void *data, int cmd) +static int lov_change_cbdata(struct obd_export *exp, + struct lov_stripe_md *lsm, ldlm_iterator_t it, + void *data) { - struct lov_async_page *lap = LAP_FROM_COOKIE(data); - - return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data, - cmd); -} + struct lov_obd *lov; + int rc = 0, i; + ENTRY; -static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa) -{ - struct lov_async_page *lap = LAP_FROM_COOKIE(data); - - lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa); - /* XXX woah, shouldn't we be altering more here? size? */ - oa->o_id = lap->lap_loi_id; - oa->o_gr = lap->lap_loi_gr; - oa->o_valid |= OBD_MD_FLGROUP; - oa->o_stripe_idx = lap->lap_stripe; -} + ASSERT_LSM_MAGIC(lsm); -static void lov_ap_update_obdo(void *data, int cmd, struct obdo *oa, - obd_valid valid) -{ - struct lov_async_page *lap = LAP_FROM_COOKIE(data); + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); - lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid); -} + LASSERT_MDS_GROUP(lsm->lsm_object_gr); -static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc) -{ - struct lov_async_page *lap = LAP_FROM_COOKIE(data); + lov = &exp->exp_obd->u.lov; + for (i = 0; i < lsm->lsm_stripe_count; i++) { + struct lov_stripe_md submd; + struct lov_oinfo *loi = lsm->lsm_oinfo[i]; - /* in a raid1 regime this would down a count of many ios - * in flight, onl calling the caller_ops completion when all - * the raid1 ios are complete */ - rc = lap->lap_caller_ops->ap_completion(lap->lap_caller_data,cmd,oa,rc); - return rc; -} + if (!lov->lov_tgts[loi->loi_ost_idx]) { + CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx); + continue; + } -static struct obd_capa *lov_ap_lookup_capa(void *data, int cmd) -{ - struct lov_async_page *lap = LAP_FROM_COOKIE(data); - return lap->lap_caller_ops->ap_lookup_capa(lap->lap_caller_data, cmd); + submd.lsm_object_id = loi->loi_id; + submd.lsm_object_gr = lsm->lsm_object_gr; + submd.lsm_stripe_count = 0; + rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, + &submd, it, data); + } + RETURN(rc); } -static struct obd_async_page_ops lov_async_page_ops = { - .ap_make_ready = lov_ap_make_ready, - .ap_refresh_count = lov_ap_refresh_count, - .ap_fill_obdo = lov_ap_fill_obdo, - .ap_update_obdo = lov_ap_update_obdo, - .ap_completion = lov_ap_completion, - .ap_lookup_capa = lov_ap_lookup_capa, -}; - -int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm, - struct lov_oinfo *loi, cfs_page_t *page, - obd_off offset, struct obd_async_page_ops *ops, - void *data, void **res) +static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm, + __u32 mode, struct lustre_handle *lockh) { - struct lov_obd *lov = &exp->exp_obd->u.lov; - struct lov_async_page *lap; - int rc = 0; + struct lov_request_set *set; + struct obd_info oinfo; + struct lov_request *req; + cfs_list_t *pos; + struct lov_obd *lov; + struct lustre_handle *lov_lockhp; + int err = 0, rc = 0; ENTRY; - if (!page) { - int i = 0; - /* Find an existing osc so we can get it's stupid sizeof(*oap). - Only because of this layering limitation will a client - mount with no osts fail */ - while (!lov->lov_tgts || !lov->lov_tgts[i] || - !lov->lov_tgts[i]->ltd_exp) { - i++; - if (i >= lov->desc.ld_tgt_count) - RETURN(-ENOMEDIUM); - } - rc = size_round(sizeof(*lap)) + - obd_prep_async_page(lov->lov_tgts[i]->ltd_exp, NULL, - NULL, NULL, 0, NULL, NULL, NULL); - RETURN(rc); - } ASSERT_LSM_MAGIC(lsm); - LASSERT(loi == NULL); - - lap = *res; - lap->lap_magic = LOV_AP_MAGIC; - lap->lap_caller_ops = ops; - lap->lap_caller_data = data; - - /* for now only raid 0 which passes through */ - lap->lap_stripe = lov_stripe_number(lsm, offset); - lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset); - loi = lsm->lsm_oinfo[lap->lap_stripe]; - - /* so the callback doesn't need the lsm */ - lap->lap_loi_id = loi->loi_id; - lap->lap_loi_gr = lsm->lsm_object_gr; - LASSERT(lsm->lsm_object_gr > 0); - - lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap)); - - rc = obd_prep_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, - lsm, loi, page, lap->lap_sub_offset, - &lov_async_page_ops, lap, - &lap->lap_sub_cookie); + + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + LASSERT_MDS_GROUP(lsm->lsm_object_gr); + LASSERT(lockh); + lov = &exp->exp_obd->u.lov; + rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set); if (rc) RETURN(rc); - CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page, - lap->lap_sub_cookie, offset); - RETURN(0); -} - -static int lov_queue_async_io(struct obd_export *exp, - struct lov_stripe_md *lsm, - struct lov_oinfo *loi, void *cookie, - int cmd, obd_off off, int count, - obd_flag brw_flags, obd_flag async_flags) -{ - struct lov_obd *lov = &exp->exp_obd->u.lov; - struct lov_async_page *lap; - int rc; - LASSERT(loi == NULL); - - ASSERT_LSM_MAGIC(lsm); - - lap = LAP_FROM_COOKIE(cookie); - - loi = lsm->lsm_oinfo[lap->lap_stripe]; - - rc = obd_queue_async_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm, - loi, lap->lap_sub_cookie, cmd, off, count, - brw_flags, async_flags); - RETURN(rc); -} - -static int lov_set_async_flags(struct obd_export *exp, - struct lov_stripe_md *lsm, - struct lov_oinfo *loi, void *cookie, - obd_flag async_flags) -{ - struct lov_obd *lov = &exp->exp_obd->u.lov; - struct lov_async_page *lap; - int rc; - - LASSERT(loi == NULL); - - ASSERT_LSM_MAGIC(lsm); - - lap = LAP_FROM_COOKIE(cookie); - - loi = lsm->lsm_oinfo[lap->lap_stripe]; - - rc = obd_set_async_flags(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, - lsm, loi, lap->lap_sub_cookie, async_flags); - RETURN(rc); -} - -static int lov_queue_group_io(struct obd_export *exp, - struct lov_stripe_md *lsm, - struct lov_oinfo *loi, - struct obd_io_group *oig, void *cookie, - int cmd, obd_off off, int count, - obd_flag brw_flags, obd_flag async_flags) -{ - struct lov_obd *lov = &exp->exp_obd->u.lov; - struct lov_async_page *lap; - int rc; - - LASSERT(loi == NULL); - - ASSERT_LSM_MAGIC(lsm); - - lap = LAP_FROM_COOKIE(cookie); - - loi = lsm->lsm_oinfo[lap->lap_stripe]; - - rc = obd_queue_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm, - loi, oig, lap->lap_sub_cookie, cmd, off, count, - brw_flags, async_flags); - RETURN(rc); -} - -/* this isn't exactly optimal. we may have queued sync io in oscs on - * all stripes, but we don't record that fact at queue time. so we - * trigger sync io on all stripes. */ -static int lov_trigger_group_io(struct obd_export *exp, - struct lov_stripe_md *lsm, - struct lov_oinfo *loi, - struct obd_io_group *oig) -{ - struct lov_obd *lov = &exp->exp_obd->u.lov; - int rc = 0, i, err; - - LASSERT(loi == NULL); - - ASSERT_LSM_MAGIC(lsm); - - for (i = 0; i < lsm->lsm_stripe_count; i++) { - loi = lsm->lsm_oinfo[i]; - if (!lov->lov_tgts[loi->loi_ost_idx] || - !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) { - CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx); - continue; - } - - err = obd_trigger_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, - lsm, loi, oig); - if (rc == 0 && err != 0) - rc = err; - }; - RETURN(rc); -} - -static int lov_teardown_async_page(struct obd_export *exp, - struct lov_stripe_md *lsm, - struct lov_oinfo *loi, void *cookie) -{ - struct lov_obd *lov = &exp->exp_obd->u.lov; - struct lov_async_page *lap; - int rc; - - LASSERT(loi == NULL); - - ASSERT_LSM_MAGIC(lsm); - - lap = LAP_FROM_COOKIE(cookie); - - loi = lsm->lsm_oinfo[lap->lap_stripe]; - - rc = obd_teardown_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, - lsm, loi, lap->lap_sub_cookie); - if (rc) { - CERROR("unable to teardown sub cookie %p: %d\n", - lap->lap_sub_cookie, rc); - RETURN(rc); - } - RETURN(rc); -} - -static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset, - void *data, int rc) -{ - struct lov_request_set *lovset = (struct lov_request_set *)data; - ENTRY; - rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset); - RETURN(rc); -} - -static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo, - struct ldlm_enqueue_info *einfo, - struct ptlrpc_request_set *rqset) -{ - ldlm_mode_t mode = einfo->ei_mode; - struct lov_request_set *set; - struct lov_request *req; - struct list_head *pos; - struct lov_obd *lov; - ldlm_error_t rc; - ENTRY; - - LASSERT(oinfo); - ASSERT_LSM_MAGIC(oinfo->oi_md); - LASSERT(mode == (mode & -mode)); - - /* we should never be asked to replay a lock this way. */ - LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0); - - if (!exp || !exp->exp_obd) - RETURN(-ENODEV); - - lov = &exp->exp_obd->u.lov; - rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set); - if (rc) - RETURN(rc); - - list_for_each (pos, &set->set_list) { - req = list_entry(pos, struct lov_request, rq_link); - - rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp, - &req->rq_oi, einfo, rqset); - if (rc != ELDLM_OK) - GOTO(out, rc); - } - - if (rqset && !list_empty(&rqset->set_requests)) { - LASSERT(rc == 0); - LASSERT(rqset->set_interpret == NULL); - rqset->set_interpret = lov_enqueue_interpret; - rqset->set_arg = (void *)set; - RETURN(rc); - } -out: - rc = lov_fini_enqueue_set(set, mode, rc, rqset); - RETURN(rc); -} - -static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm, - __u32 type, ldlm_policy_data_t *policy, __u32 mode, - int *flags, void *data, struct lustre_handle *lockh) -{ - struct lov_request_set *set; - struct obd_info oinfo; - struct lov_request *req; - struct list_head *pos; - struct lov_obd *lov = &exp->exp_obd->u.lov; - struct lustre_handle *lov_lockhp; - int lov_flags, rc = 0; - ENTRY; - - ASSERT_LSM_MAGIC(lsm); - LASSERT((*flags & LDLM_FL_TEST_LOCK) || mode == (mode & -mode)); - - if (!exp || !exp->exp_obd) - RETURN(-ENODEV); - - lov = &exp->exp_obd->u.lov; - rc = lov_prep_match_set(exp, &oinfo, lsm, policy, mode, lockh, &set); - if (rc) - RETURN(rc); - - list_for_each (pos, &set->set_list) { - ldlm_policy_data_t sub_policy; - req = list_entry(pos, struct lov_request, rq_link); - lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe; - LASSERT(lov_lockhp); - - lov_flags = *flags; - sub_policy.l_extent = req->rq_oi.oi_policy.l_extent; - - rc = obd_match(lov->lov_tgts[req->rq_idx]->ltd_exp, - req->rq_oi.oi_md, type, &sub_policy, - mode, &lov_flags, data, lov_lockhp); - rc = lov_update_match_set(set, req, rc); - if (rc <= 0) - break; - } - lov_fini_match_set(set, mode, *flags); - RETURN(rc); -} - -static int lov_change_cbdata(struct obd_export *exp, - struct lov_stripe_md *lsm, ldlm_iterator_t it, - void *data) -{ - struct lov_obd *lov; - int rc = 0, i; - ENTRY; - - ASSERT_LSM_MAGIC(lsm); - - if (!exp || !exp->exp_obd) - RETURN(-ENODEV); - - LASSERT(lsm->lsm_object_gr > 0); - - lov = &exp->exp_obd->u.lov; - for (i = 0; i < lsm->lsm_stripe_count; i++) { - struct lov_stripe_md submd; - struct lov_oinfo *loi = lsm->lsm_oinfo[i]; - - if (!lov->lov_tgts[loi->loi_ost_idx]) { - CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx); - continue; - } - - submd.lsm_object_id = loi->loi_id; - submd.lsm_object_gr = lsm->lsm_object_gr; - submd.lsm_stripe_count = 0; - rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, - &submd, it, data); - } - RETURN(rc); -} - -static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm, - __u32 mode, struct lustre_handle *lockh) -{ - struct lov_request_set *set; - struct obd_info oinfo; - struct lov_request *req; - struct list_head *pos; - struct lov_obd *lov = &exp->exp_obd->u.lov; - struct lustre_handle *lov_lockhp; - int err = 0, rc = 0; - ENTRY; - - ASSERT_LSM_MAGIC(lsm); - - if (!exp || !exp->exp_obd) - RETURN(-ENODEV); - - LASSERT(lsm->lsm_object_gr > 0); - LASSERT(lockh); - lov = &exp->exp_obd->u.lov; - rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set); - if (rc) - RETURN(rc); - - list_for_each (pos, &set->set_list) { - req = list_entry(pos, struct lov_request, rq_link); - lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe; + cfs_list_for_each (pos, &set->set_list) { + req = cfs_list_entry(pos, struct lov_request, rq_link); + lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe; rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp, req->rq_oi.oi_md, mode, lov_lockhp); @@ -2049,7 +1821,7 @@ static int lov_cancel_unused(struct obd_export *exp, ASSERT_LSM_MAGIC(lsm); - LASSERT(lsm->lsm_object_gr > 0); + LASSERT_MDS_GROUP(lsm->lsm_object_gr); for (i = 0; i < lsm->lsm_stripe_count; i++) { struct lov_stripe_md submd; struct lov_oinfo *loi = lsm->lsm_oinfo[i]; @@ -2079,50 +1851,7 @@ static int lov_cancel_unused(struct obd_export *exp, RETURN(rc); } -static int lov_join_lru(struct obd_export *exp, - struct lov_stripe_md *lsm, int join) -{ - struct lov_obd *lov; - int i, count = 0; - ENTRY; - - ASSERT_LSM_MAGIC(lsm); - if (!exp || !exp->exp_obd) - RETURN(-ENODEV); - - lov = &exp->exp_obd->u.lov; - for (i = 0; i < lsm->lsm_stripe_count; i++) { - struct lov_stripe_md submd; - struct lov_oinfo *loi = lsm->lsm_oinfo[i]; - int rc = 0; - - if (!lov->lov_tgts[loi->loi_ost_idx]) { - CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx); - continue; - } - - if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active) - CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx); - - submd.lsm_object_id = loi->loi_id; - submd.lsm_object_gr = lsm->lsm_object_gr; - submd.lsm_stripe_count = 0; - rc = obd_join_lru(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, - &submd, join); - if (rc < 0) { - CERROR("join lru failed. objid: "LPX64" subobj: "LPX64 - " ostidx: %d rc: %d\n", lsm->lsm_object_id, - loi->loi_id, loi->loi_ost_idx, rc); - return rc; - } else { - count += rc; - } - } - RETURN(count); -} - -static int lov_statfs_interpret(struct ptlrpc_request_set *rqset, - void *data, int rc) +int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc) { struct lov_request_set *lovset = (struct lov_request_set *)data; int err; @@ -2140,11 +1869,11 @@ static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo, { struct lov_request_set *set; struct lov_request *req; - struct list_head *pos; + cfs_list_t *pos; struct lov_obd *lov; int rc = 0; ENTRY; - + LASSERT(oinfo != NULL); LASSERT(oinfo->oi_osfs != NULL); @@ -2153,10 +1882,10 @@ static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo, if (rc) RETURN(rc); - list_for_each (pos, &set->set_list) { + cfs_list_for_each (pos, &set->set_list) { struct obd_device *osc_obd; - req = list_entry(pos, struct lov_request, rq_link); + req = cfs_list_entry(pos, struct lov_request, rq_link); osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp); rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset); @@ -2164,7 +1893,7 @@ static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo, break; } - if (rc || list_empty(&rqset->set_requests)) { + if (rc || cfs_list_empty(&rqset->set_requests)) { int err; if (rc) set->set_completes = 0; @@ -2179,38 +1908,28 @@ static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo, } static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs, - __u64 max_age) + __u64 max_age, __u32 flags) { - struct lov_obd *lov = &obd->u.lov; - struct obd_statfs lov_sfs; - int set = 0; - int rc = 0, err; - int i; + struct ptlrpc_request_set *set = NULL; + struct obd_info oinfo = { { { 0 } } }; + int rc = 0; ENTRY; - /* We only get block data from the OBD */ - for (i = 0; i < lov->desc.ld_tgt_count; i++) { - if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) { - CDEBUG(D_HA, "lov idx %d inactive\n", i); - continue; - } - err = obd_statfs(class_exp2obd(lov->lov_tgts[i]->ltd_exp), - &lov_sfs, max_age); - if (err) { - if (lov->lov_tgts[i]->ltd_active && !rc) - rc = err; - continue; - } - lov_update_statfs(class_exp2obd(lov->lov_tgts[i]->ltd_exp), - osfs, &lov_sfs, set); - set++; - } + /* for obdclass we forbid using obd_statfs_rqset, but prefer using async + * statfs requests */ + set = ptlrpc_prep_set(); + if (set == NULL) + RETURN(-ENOMEM); - err = lov_fini_statfs(obd, osfs, set); - qos_update(lov); + oinfo.oi_osfs = osfs; + oinfo.oi_flags = flags; + rc = lov_statfs_async(obd, &oinfo, max_age, set); + if (rc == 0) + rc = ptlrpc_set_wait(set); + ptlrpc_set_destroy(set); - RETURN(rc ? rc : err); + RETURN(rc); } static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len, @@ -2218,11 +1937,47 @@ static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len, { struct obd_device *obddev = class_exp2obd(exp); struct lov_obd *lov = &obddev->u.lov; - int i, rc, count = lov->desc.ld_tgt_count; + int i = 0, rc = 0, count = lov->desc.ld_tgt_count; struct obd_uuid *uuidp; ENTRY; switch (cmd) { + case IOC_OBD_STATFS: { + struct obd_ioctl_data *data = karg; + struct obd_device *osc_obd; + struct obd_statfs stat_buf = {0}; + __u32 index; + + memcpy(&index, data->ioc_inlbuf2, sizeof(__u32)); + LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs)); + + if ((index >= count)) + RETURN(-ENODEV); + + if (!lov->lov_tgts[index]) + /* Try again with the next index */ + RETURN(-EAGAIN); + if (!lov->lov_tgts[index]->ltd_active) + RETURN(-ENODATA); + + osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp); + if (!osc_obd) + RETURN(-EINVAL); + + /* got statfs data */ + rc = obd_statfs(osc_obd, &stat_buf, + cfs_time_current_64() - CFS_HZ, 0); + if (rc) + RETURN(rc); + if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf, + data->ioc_plen1)) + RETURN(-EFAULT); + /* copy UUID */ + if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd), + data->ioc_plen2)) + RETURN(-EFAULT); + break; + } case OBD_IOC_LOV_GET_CONFIG: { struct obd_ioctl_data *data; struct lov_desc *desc; @@ -2263,14 +2018,13 @@ static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len, *genp = lov->lov_tgts[i]->ltd_gen; } - rc = copy_to_user((void *)uarg, buf, len); - if (rc) + if (cfs_copy_to_user((void *)uarg, buf, len)) rc = -EFAULT; obd_ioctl_freedata(buf, len); break; } case LL_IOC_LOV_SETSTRIPE: - rc = lov_setstripe(exp, karg, uarg); + rc = lov_setstripe(exp, len, karg, uarg); break; case LL_IOC_LOV_GETSTRIPE: rc = lov_getstripe(exp, karg, uarg); @@ -2278,13 +2032,59 @@ static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len, case LL_IOC_LOV_SETEA: rc = lov_setea(exp, karg, uarg); break; + case OBD_IOC_QUOTACTL: { + struct if_quotactl *qctl = karg; + struct lov_tgt_desc *tgt = NULL; + struct obd_quotactl *oqctl; + + if (qctl->qc_valid == QC_OSTIDX) { + if (qctl->qc_idx < 0 || count <= qctl->qc_idx) + RETURN(-EINVAL); + + tgt = lov->lov_tgts[qctl->qc_idx]; + if (!tgt || !tgt->ltd_exp) + RETURN(-EINVAL); + } else if (qctl->qc_valid == QC_UUID) { + for (i = 0; i < count; i++) { + tgt = lov->lov_tgts[i]; + if (!tgt || + !obd_uuid_equals(&tgt->ltd_uuid, + &qctl->obd_uuid)) + continue; + + if (tgt->ltd_exp == NULL) + RETURN(-EINVAL); + + break; + } + } else { + RETURN(-EINVAL); + } + + if (i >= count) + RETURN(-EAGAIN); + + LASSERT(tgt && tgt->ltd_exp); + OBD_ALLOC_PTR(oqctl); + if (!oqctl) + RETURN(-ENOMEM); + + QCTL_COPY(oqctl, qctl); + rc = obd_quotactl(tgt->ltd_exp, oqctl); + if (rc == 0) { + QCTL_COPY(qctl, oqctl); + qctl->qc_valid = QC_OSTIDX; + qctl->obd_uuid = tgt->ltd_uuid; + } + OBD_FREE_PTR(oqctl); + break; + } default: { int set = 0; if (count == 0) RETURN(-ENOTTY); - rc = 0; for (i = 0; i < count; i++) { int err; @@ -2319,8 +2119,361 @@ static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len, RETURN(rc); } +#define FIEMAP_BUFFER_SIZE 4096 + +/** + * Non-zero fe_logical indicates that this is a continuation FIEMAP + * call. The local end offset and the device are sent in the first + * fm_extent. This function calculates the stripe number from the index. + * This function returns a stripe_no on which mapping is to be restarted. + * + * This function returns fm_end_offset which is the in-OST offset at which + * mapping should be restarted. If fm_end_offset=0 is returned then caller + * will re-calculate proper offset in next stripe. + * Note that the first extent is passed to lov_get_info via the value field. + * + * \param fiemap fiemap request header + * \param lsm striping information for the file + * \param fm_start logical start of mapping + * \param fm_end logical end of mapping + * \param start_stripe starting stripe will be returned in this + */ +obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap, + struct lov_stripe_md *lsm, obd_size fm_start, + obd_size fm_end, int *start_stripe) +{ + obd_size local_end = fiemap->fm_extents[0].fe_logical; + obd_off lun_start, lun_end; + obd_size fm_end_offset; + int stripe_no = -1, i; + + if (fiemap->fm_extent_count == 0 || + fiemap->fm_extents[0].fe_logical == 0) + return 0; + + /* Find out stripe_no from ost_index saved in the fe_device */ + for (i = 0; i < lsm->lsm_stripe_count; i++) { + if (lsm->lsm_oinfo[i]->loi_ost_idx == + fiemap->fm_extents[0].fe_device) { + stripe_no = i; + break; + } + } + + /* If we have finished mapping on previous device, shift logical + * offset to start of next device */ + if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end, + &lun_start, &lun_end)) != 0 && + local_end < lun_end) { + fm_end_offset = local_end; + *start_stripe = stripe_no; + } else { + /* This is a special value to indicate that caller should + * calculate offset in next stripe. */ + fm_end_offset = 0; + *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count; + } + + return fm_end_offset; +} + +/** + * We calculate on which OST the mapping will end. If the length of mapping + * is greater than (stripe_size * stripe_count) then the last_stripe will + * will be one just before start_stripe. Else we check if the mapping + * intersects each OST and find last_stripe. + * This function returns the last_stripe and also sets the stripe_count + * over which the mapping is spread + * + * \param lsm striping information for the file + * \param fm_start logical start of mapping + * \param fm_end logical end of mapping + * \param start_stripe starting stripe of the mapping + * \param stripe_count the number of stripes across which to map is returned + * + * \retval last_stripe return the last stripe of the mapping + */ +int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start, + obd_size fm_end, int start_stripe, + int *stripe_count) +{ + int last_stripe; + obd_off obd_start, obd_end; + int i, j; + + if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) { + last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 : + start_stripe - 1); + *stripe_count = lsm->lsm_stripe_count; + } else { + for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count; + i = (i + 1) % lsm->lsm_stripe_count, j++) { + if ((lov_stripe_intersects(lsm, i, fm_start, fm_end, + &obd_start, &obd_end)) == 0) + break; + } + *stripe_count = j; + last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count; + } + + return last_stripe; +} + +/** + * Set fe_device and copy extents from local buffer into main return buffer. + * + * \param fiemap fiemap request header + * \param lcl_fm_ext array of local fiemap extents to be copied + * \param ost_index OST index to be written into the fm_device field for each + extent + * \param ext_count number of extents to be copied + * \param current_extent where to start copying in main extent array + */ +void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap, + struct ll_fiemap_extent *lcl_fm_ext, + int ost_index, unsigned int ext_count, + int current_extent) +{ + char *to; + int ext; + + for (ext = 0; ext < ext_count; ext++) { + lcl_fm_ext[ext].fe_device = ost_index; + lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET; + } + + /* Copy fm_extent's from fm_local to return buffer */ + to = (char *)fiemap + fiemap_count_to_size(current_extent); + memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent)); +} + +/** + * Break down the FIEMAP request and send appropriate calls to individual OSTs. + * This also handles the restarting of FIEMAP calls in case mapping overflows + * the available number of extents in single call. + */ +static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key, + __u32 *vallen, void *val, struct lov_stripe_md *lsm) +{ + struct ll_fiemap_info_key *fm_key = key; + struct ll_user_fiemap *fiemap = val; + struct ll_user_fiemap *fm_local = NULL; + struct ll_fiemap_extent *lcl_fm_ext; + int count_local; + unsigned int get_num_extents = 0; + int ost_index = 0, actual_start_stripe, start_stripe; + obd_size fm_start, fm_end, fm_length, fm_end_offset = 0; + obd_size curr_loc; + int current_extent = 0, rc = 0, i; + int ost_eof = 0; /* EOF for object */ + int ost_done = 0; /* done with required mapping for this OST? */ + int last_stripe; + int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count; + unsigned int buffer_size = FIEMAP_BUFFER_SIZE; + + if (lsm == NULL) + GOTO(out, rc = 0); + + if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size) + buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count); + + OBD_ALLOC(fm_local, buffer_size); + if (fm_local == NULL) + GOTO(out, rc = -ENOMEM); + lcl_fm_ext = &fm_local->fm_extents[0]; + + count_local = fiemap_size_to_count(buffer_size); + + memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap)); + fm_start = fiemap->fm_start; + fm_length = fiemap->fm_length; + /* Calculate start stripe, last stripe and length of mapping */ + actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start); + fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size : + fm_start + fm_length - 1); + /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */ + if (fm_end > fm_key->oa.o_size) + fm_end = fm_key->oa.o_size; + + last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end, + actual_start_stripe, &stripe_count); + + fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end, + &start_stripe); + + if (fiemap->fm_extent_count == 0) { + get_num_extents = 1; + count_local = 0; + } + + /* Check each stripe */ + for (cur_stripe = start_stripe, i = 0; i < stripe_count; + i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) { + obd_size req_fm_len; /* Stores length of required mapping */ + obd_size len_mapped_single_call; + obd_off lun_start, lun_end, obd_object_end; + unsigned int ext_count; + + cur_stripe_wrap = cur_stripe; + + /* Find out range of mapping on this stripe */ + if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end, + &lun_start, &obd_object_end)) == 0) + continue; + + /* If this is a continuation FIEMAP call and we are on + * starting stripe then lun_start needs to be set to + * fm_end_offset */ + if (fm_end_offset != 0 && cur_stripe == start_stripe) + lun_start = fm_end_offset; + + if (fm_length != ~0ULL) { + /* Handle fm_start + fm_length overflow */ + if (fm_start + fm_length < fm_start) + fm_length = ~0ULL - fm_start; + lun_end = lov_size_to_stripe(lsm, fm_start + fm_length, + cur_stripe); + } else { + lun_end = ~0ULL; + } + + if (lun_start == lun_end) + continue; + + req_fm_len = obd_object_end - lun_start; + fm_local->fm_length = 0; + len_mapped_single_call = 0; + + /* If the output buffer is very large and the objects have many + * extents we may need to loop on a single OST repeatedly */ + ost_eof = 0; + ost_done = 0; + do { + if (get_num_extents == 0) { + /* Don't get too many extents. */ + if (current_extent + count_local > + fiemap->fm_extent_count) + count_local = fiemap->fm_extent_count - + current_extent; + } + + lun_start += len_mapped_single_call; + fm_local->fm_length = req_fm_len - len_mapped_single_call; + req_fm_len = fm_local->fm_length; + fm_local->fm_extent_count = count_local; + fm_local->fm_mapped_extents = 0; + fm_local->fm_flags = fiemap->fm_flags; + + fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id; + ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx; + + if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count) + GOTO(out, rc = -EINVAL); + + /* If OST is inactive, return extent with UNKNOWN flag */ + if (!lov->lov_tgts[ost_index]->ltd_active) { + fm_local->fm_flags |= FIEMAP_EXTENT_LAST; + fm_local->fm_mapped_extents = 1; + + lcl_fm_ext[0].fe_logical = lun_start; + lcl_fm_ext[0].fe_length = obd_object_end - + lun_start; + lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN; + + goto inactive_tgt; + } + + fm_local->fm_start = lun_start; + fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER; + memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local)); + *vallen=fiemap_count_to_size(fm_local->fm_extent_count); + rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp, + keylen, key, vallen, fm_local, lsm); + if (rc != 0) + GOTO(out, rc); + +inactive_tgt: + ext_count = fm_local->fm_mapped_extents; + if (ext_count == 0) { + ost_done = 1; + /* If last stripe has hole at the end, + * then we need to return */ + if (cur_stripe_wrap == last_stripe) { + fiemap->fm_mapped_extents = 0; + goto finish; + } + break; + } + + /* If we just need num of extents then go to next device */ + if (get_num_extents) { + current_extent += ext_count; + break; + } + + len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical - + lun_start + lcl_fm_ext[ext_count - 1].fe_length; + + /* Have we finished mapping on this device? */ + if (req_fm_len <= len_mapped_single_call) + ost_done = 1; + + /* Clear the EXTENT_LAST flag which can be present on + * last extent */ + if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST) + lcl_fm_ext[ext_count - 1].fe_flags &= + ~FIEMAP_EXTENT_LAST; + + curr_loc = lov_stripe_size(lsm, + lcl_fm_ext[ext_count - 1].fe_logical+ + lcl_fm_ext[ext_count - 1].fe_length, + cur_stripe); + if (curr_loc >= fm_key->oa.o_size) + ost_eof = 1; + + fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext, + ost_index, ext_count, + current_extent); + + current_extent += ext_count; + + /* Ran out of available extents? */ + if (current_extent >= fiemap->fm_extent_count) + goto finish; + } while (ost_done == 0 && ost_eof == 0); + + if (cur_stripe_wrap == last_stripe) + goto finish; + } + +finish: + /* Indicate that we are returning device offsets unless file just has + * single stripe */ + if (lsm->lsm_stripe_count > 1) + fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER; + + if (get_num_extents) + goto skip_last_device_calc; + + /* Check if we have reached the last stripe and whether mapping for that + * stripe is done. */ + if (cur_stripe_wrap == last_stripe) { + if (ost_done || ost_eof) + fiemap->fm_extents[current_extent - 1].fe_flags |= + FIEMAP_EXTENT_LAST; + } + +skip_last_device_calc: + fiemap->fm_mapped_extents = current_extent; + +out: + OBD_FREE(fm_local, buffer_size); + return rc; +} + static int lov_get_info(struct obd_export *exp, __u32 keylen, - void *key, __u32 *vallen, void *val) + void *key, __u32 *vallen, void *val, + struct lov_stripe_md *lsm) { struct obd_device *obddev = class_exp2obd(exp); struct lov_obd *lov = &obddev->u.lov; @@ -2330,13 +2483,12 @@ static int lov_get_info(struct obd_export *exp, __u32 keylen, if (!vallen || !val) RETURN(-EFAULT); - lov_getref(obddev); + obd_getref(obddev); - if (KEY_IS("lock_to_stripe")) { + if (KEY_IS(KEY_LOCK_TO_STRIPE)) { struct { char name[16]; struct ldlm_lock *lock; - struct lov_stripe_md *lsm; } *data = key; struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name; struct lov_oinfo *loi; @@ -2352,24 +2504,23 @@ static int lov_get_info(struct obd_export *exp, __u32 keylen, /* XXX - it's assumed all the locks for deleted OSTs have * been cancelled. Also, the export for deleted OSTs will * be NULL and won't match the lock's export. */ - for (i = 0; i < data->lsm->lsm_stripe_count; i++) { - loi = data->lsm->lsm_oinfo[i]; + for (i = 0; i < lsm->lsm_stripe_count; i++) { + loi = lsm->lsm_oinfo[i]; if (!lov->lov_tgts[loi->loi_ost_idx]) continue; if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp == data->lock->l_conn_export && - loi->loi_id == res_id->name[0] && - loi->loi_gr == res_id->name[2]) { + osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) { *stripe = i; GOTO(out, rc = 0); } } LDLM_ERROR(data->lock, "lock on inode without such object"); - dump_lsm(D_ERROR, data->lsm); + dump_lsm(D_ERROR, lsm); GOTO(out, rc = -ENXIO); } else if (KEY_IS(KEY_LAST_ID)) { struct obd_id_info *info = val; - int size = sizeof(obd_id); + __u32 size = sizeof(obd_id); struct lov_tgt_desc *tgt; LASSERT(*vallen == sizeof(struct obd_id_info)); @@ -2378,26 +2529,22 @@ static int lov_get_info(struct obd_export *exp, __u32 keylen, if (!tgt || !tgt->ltd_active) GOTO(out, rc = -ESRCH); - rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data); + rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL); GOTO(out, rc = 0); } else if (KEY_IS(KEY_LOVDESC)) { struct lov_desc *desc_ret = val; *desc_ret = lov->desc; GOTO(out, rc = 0); - } else if (KEY_IS(KEY_LOV_IDX)) { - struct lov_tgt_desc *tgt; - - for(i = 0; i < lov->desc.ld_tgt_count; i++) { - tgt = lov->lov_tgts[i]; - if (obd_uuid_equals(val, &tgt->ltd_uuid)) - GOTO(out, rc = i); - } + } else if (KEY_IS(KEY_FIEMAP)) { + rc = lov_fiemap(lov, keylen, key, vallen, val, lsm); + GOTO(out, rc); } rc = -EINVAL; + out: - lov_putref(obddev); + obd_putref(obddev); RETURN(rc); } @@ -2412,7 +2559,7 @@ static int lov_set_info_async(struct obd_export *exp, obd_count keylen, struct lov_tgt_desc *tgt; unsigned incr, check_uuid, do_inactive, no_set; - unsigned next_id = 0, mds_con = 0; + unsigned next_id = 0, mds_con = 0, capa = 0; ENTRY; incr = check_uuid = do_inactive = no_set = 0; @@ -2423,7 +2570,7 @@ static int lov_set_info_async(struct obd_export *exp, obd_count keylen, RETURN(-ENOMEM); } - lov_getref(obddev); + obd_getref(obddev); count = lov->desc.ld_tgt_count; if (KEY_IS(KEY_NEXT_ID)) { @@ -2432,14 +2579,14 @@ static int lov_set_info_async(struct obd_export *exp, obd_count keylen, incr = sizeof(struct obd_id_info); do_inactive = 1; next_id = 1; - } else if (KEY_IS("checksum")) { + } else if (KEY_IS(KEY_CHECKSUM)) { do_inactive = 1; - } else if (KEY_IS("unlinked")) { - check_uuid = val ? 1 : 0; - } else if (KEY_IS("evict_by_nid")) { + } else if (KEY_IS(KEY_EVICT_BY_NID)) { /* use defaults: do_inactive = incr = 0; */ } else if (KEY_IS(KEY_MDS_CONN)) { mds_con = 1; + } else if (KEY_IS(KEY_CAPA_KEY)) { + capa = 1; } for (i = 0; i < count; i++, val = (char *)val + incr) { @@ -2474,7 +2621,21 @@ static int lov_set_info_async(struct obd_export *exp, obd_count keylen, err = obd_set_info_async(tgt->ltd_exp, keylen, key, vallen, ((struct obd_id_info*)val)->data, set); - } else { + } else if (capa) { + struct mds_capa_info *info = (struct mds_capa_info*)val; + + LASSERT(vallen == sizeof(*info)); + + /* Only want a specific OSC */ + if (info->uuid && + !obd_uuid_equals(info->uuid, &tgt->ltd_uuid)) + continue; + + err = obd_set_info_async(tgt->ltd_exp, keylen, key, + sizeof(*info->capa), + info->capa, set); + + } else { /* Only want a specific OSC */ if (check_uuid && !obd_uuid_equals(val, &tgt->ltd_uuid)) @@ -2488,7 +2649,7 @@ static int lov_set_info_async(struct obd_export *exp, obd_count keylen, rc = err; } - lov_putref(obddev); + obd_putref(obddev); if (no_set) { err = ptlrpc_set_wait(set); if (!rc) @@ -2498,21 +2659,6 @@ static int lov_set_info_async(struct obd_export *exp, obd_count keylen, RETURN(rc); } -static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp, - struct lov_stripe_md *lsm) -{ - int rc; - ENTRY; - - if (!lsm) - RETURN(0); - LASSERT(md_exp); - LASSERT(lsm_op_find(lsm->lsm_magic) != NULL); - rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd); - - RETURN(rc); -} - int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm) { int i, rc = 0; @@ -2553,125 +2699,36 @@ static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm, RETURN(0); } - -#if 0 -struct lov_multi_wait { - struct ldlm_lock *lock; - wait_queue_t wait; - int completed; - int generation; -}; - -int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm, - struct lustre_handle *lockh) -{ - struct lov_lock_handles *lov_lockh = NULL; - struct lustre_handle *lov_lockhp; - struct lov_obd *lov; - struct lov_oinfo *loi; - struct lov_multi_wait *queues; - int rc = 0, i; - ENTRY; - - ASSERT_LSM_MAGIC(lsm); - - if (!exp || !exp->exp_obd) - RETURN(-ENODEV); - - LASSERT(lockh != NULL); - if (lsm->lsm_stripe_count > 1) { - lov_lockh = lov_handle2llh(lockh); - if (lov_lockh == NULL) { - CERROR("LOV: invalid lov lock handle %p\n", lockh); - RETURN(-EINVAL); - } - - lov_lockhp = lov_lockh->llh_handles; - } else { - lov_lockhp = lockh; - } - - OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues)); - if (queues == NULL) - GOTO(out, rc = -ENOMEM); - - lov = &exp->exp_obd->u.lov; - for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; - i++, loi++, lov_lockhp++) { - struct ldlm_lock *lock; - struct obd_device *obd; - - lock = ldlm_handle2lock(lov_lockhp); - if (lock == NULL) { - CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n", - loi->loi_ost_idx, loi->loi_id); - queues[i].completed = 1; - continue; - } - - queues[i].lock = lock; - init_waitqueue_entry(&(queues[i].wait), current); - add_wait_queue(lock->l_waitq, &(queues[i].wait)); - - obd = class_exp2obd(lock->l_conn_export); - if (obd != NULL) - imp = obd->u.cli.cl_import; - if (imp != NULL) { - spin_lock(&imp->imp_lock); - queues[i].generation = imp->imp_generation; - spin_unlock(&imp->imp_lock); - } - } - - lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait, - interrupted_completion_wait, &lwd); - rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi); - - for (i = 0; i < lsm->lsm_stripe_count; i++) - remove_wait_queue(lock->l_waitq, &(queues[i].wait)); - - if (rc == -EINTR || rc == -ETIMEDOUT) { - - - } - - out: - if (lov_lockh != NULL) - lov_llh_put(lov_lockh); - RETURN(rc); -} -#endif - void lov_stripe_lock(struct lov_stripe_md *md) { - LASSERT(md->lsm_lock_owner != cfs_current()); - spin_lock(&md->lsm_lock); - LASSERT(md->lsm_lock_owner == NULL); - md->lsm_lock_owner = cfs_current(); + LASSERT(md->lsm_lock_owner != cfs_curproc_pid()); + cfs_spin_lock(&md->lsm_lock); + LASSERT(md->lsm_lock_owner == 0); + md->lsm_lock_owner = cfs_curproc_pid(); } EXPORT_SYMBOL(lov_stripe_lock); void lov_stripe_unlock(struct lov_stripe_md *md) { - LASSERT(md->lsm_lock_owner == cfs_current()); - md->lsm_lock_owner = NULL; - spin_unlock(&md->lsm_lock); + LASSERT(md->lsm_lock_owner == cfs_curproc_pid()); + md->lsm_lock_owner = 0; + cfs_spin_unlock(&md->lsm_lock); } EXPORT_SYMBOL(lov_stripe_unlock); + struct obd_ops lov_obd_ops = { .o_owner = THIS_MODULE, .o_setup = lov_setup, .o_precleanup = lov_precleanup, .o_cleanup = lov_cleanup, - .o_process_config = lov_process_config, + //.o_process_config = lov_process_config, .o_connect = lov_connect, .o_disconnect = lov_disconnect, .o_statfs = lov_statfs, .o_statfs_async = lov_statfs_async, .o_packmd = lov_packmd, .o_unpackmd = lov_unpackmd, - .o_checkmd = lov_checkmd, .o_create = lov_create, .o_destroy = lov_destroy, .o_getattr = lov_getattr, @@ -2679,23 +2736,14 @@ struct obd_ops lov_obd_ops = { .o_setattr = lov_setattr, .o_setattr_async = lov_setattr_async, .o_brw = lov_brw, - .o_brw_async = lov_brw_async, - .o_prep_async_page = lov_prep_async_page, - .o_queue_async_io = lov_queue_async_io, - .o_set_async_flags = lov_set_async_flags, - .o_queue_group_io = lov_queue_group_io, - .o_trigger_group_io = lov_trigger_group_io, - .o_teardown_async_page = lov_teardown_async_page, .o_merge_lvb = lov_merge_lvb, .o_adjust_kms = lov_adjust_kms, .o_punch = lov_punch, .o_sync = lov_sync, .o_enqueue = lov_enqueue, - .o_match = lov_match, .o_change_cbdata = lov_change_cbdata, .o_cancel = lov_cancel, .o_cancel_unused = lov_cancel_unused, - .o_join_lru = lov_join_lru, .o_iocontrol = lov_iocontrol, .o_get_info = lov_get_info, .o_set_info_async = lov_set_info_async, @@ -2703,6 +2751,12 @@ struct obd_ops lov_obd_ops = { .o_llog_init = lov_llog_init, .o_llog_finish = lov_llog_finish, .o_notify = lov_notify, + .o_pool_new = lov_pool_new, + .o_pool_rem = lov_pool_remove, + .o_pool_add = lov_pool_add, + .o_pool_del = lov_pool_del, + .o_getref = lov_getref, + .o_putref = lov_putref, }; static quota_interface_t *quota_interface; @@ -2710,30 +2764,45 @@ extern quota_interface_t lov_quota_interface; cfs_mem_cache_t *lov_oinfo_slab; +extern struct lu_kmem_descr lov_caches[]; + int __init lov_init(void) { struct lprocfs_static_vars lvars = { 0 }; int rc, rc2; ENTRY; + /* print an address of _any_ initialized kernel symbol from this + * module, to allow debugging with gdb that doesn't support data + * symbols from modules.*/ + CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches); + + rc = lu_kmem_init(lov_caches); + if (rc) + return rc; + lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo", - sizeof(struct lov_oinfo), - 0, SLAB_HWCACHE_ALIGN); - if (lov_oinfo_slab == NULL) + sizeof(struct lov_oinfo), + 0, CFS_SLAB_HWCACHE_ALIGN); + if (lov_oinfo_slab == NULL) { + lu_kmem_fini(lov_caches); return -ENOMEM; + } lprocfs_lov_init_vars(&lvars); - request_module("lquota"); + cfs_request_module("lquota"); quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface); init_obd_quota_ops(quota_interface, &lov_obd_ops); rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars, - LUSTRE_LOV_NAME, NULL); + LUSTRE_LOV_NAME, &lov_device_type); + if (rc) { if (quota_interface) PORTAL_SYMBOL_PUT(lov_quota_interface); rc2 = cfs_mem_cache_destroy(lov_oinfo_slab); LASSERT(rc2 == 0); + lu_kmem_fini(lov_caches); } RETURN(rc); @@ -2743,7 +2812,10 @@ int __init lov_init(void) static void /*__exit*/ lov_exit(void) { int rc; - + + lu_device_type_fini(&lov_device_type); + lu_kmem_fini(lov_caches); + if (quota_interface) PORTAL_SYMBOL_PUT(lov_quota_interface); @@ -2752,7 +2824,7 @@ static void /*__exit*/ lov_exit(void) LASSERT(rc == 0); } -MODULE_AUTHOR("Cluster File Systems, Inc. "); +MODULE_AUTHOR("Sun Microsystems, Inc. "); MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver"); MODULE_LICENSE("GPL");