X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Flov%2Flov_obd.c;h=6753cded0d6569d2044f562954f53d727224f6ea;hp=e7cf3cb71460b4bb96cce4d81af82c5028188798;hb=dd7363438f01fa117c4984c72a6f90c4a80a4588;hpb=4b28597565a5e732e1bb2b6aecf25e71bfeacccc diff --git a/lustre/lov/lov_obd.c b/lustre/lov/lov_obd.c index e7cf3cb..6753cde 100644 --- a/lustre/lov/lov_obd.c +++ b/lustre/lov/lov_obd.c @@ -1,277 +1,343 @@ - /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- +/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * lov/lov.c - * - * Copyright (C) 2002 Cluster File Systems, Inc. - * Author: Phil Schwan + * Copyright (C) 2002-2006 Cluster File Systems, Inc. + * Author: Phil Schwan * Peter Braam + * Mike Shaver + * Nathan Rutman + * + * This file is part of the Lustre file system, http://www.lustre.org + * Lustre is a trademark of Cluster File Systems, Inc. + * + * 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. + * + * 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. * - * This code is issued under the GNU General Public License. - * See the file COPYING in this distribution + * 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. */ -#define EXPORT_SYMTAB +#ifndef EXPORT_SYMTAB +# define EXPORT_SYMTAB +#endif #define DEBUG_SUBSYSTEM S_LOV +#ifdef __KERNEL__ +#include +#else +#include +#endif -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include - -extern struct lprocfs_vars status_var_nm_1[]; -extern struct lprocfs_vars status_class_var[]; - -static kmem_cache_t *lov_file_cache; - -struct lov_file_handles { - struct list_head lfh_list; - __u64 lfh_cookie; - int lfh_count; - struct lustre_handle *lfh_handles; -}; - -extern int lov_packmd(struct lustre_handle *conn, struct lov_mds_md **lmm, - struct lov_stripe_md *lsm); -extern int lov_unpackmd(struct lustre_handle *conn, struct lov_stripe_md **lsm, - struct lov_mds_md *lmm); - -/* obd methods */ -int lov_attach(struct obd_device *dev, obd_count len, void *data) +#include +#include +#include +#include +#include +#include +#include +#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) { - return lprocfs_reg_obd(dev, status_var_nm_1, dev); + 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); + return; } -int lov_detach(struct obd_device *dev) +static void __lov_del_obd(struct obd_device *obd, __u32 index); + +void lov_putref(struct obd_device *obd) { - return lprocfs_dereg_obd(dev); + struct lov_obd *lov = &obd->u.lov; + 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) { + int i; + 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) + continue; + /* Disconnect and delete from list */ + __lov_del_obd(obd, i); + lov->lov_death_row--; + } + } + mutex_up(&lov->lov_lock); } -static int lov_connect(struct lustre_handle *conn, struct obd_device *obd, - obd_uuid_t cluuid, struct recovd_obd *recovd, - ptlrpc_recovery_cb_t recover) +#define MAX_STRING_SIZE 128 +static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate, + struct obd_connect_data *data) { - struct ptlrpc_request *req = NULL; struct lov_obd *lov = &obd->u.lov; - struct client_obd *mdc = &lov->mdcobd->u.cli; - struct lov_desc *desc = &lov->desc; - struct obd_export *exp; - struct lustre_handle mdc_conn; - obd_uuid_t *uuidarray; - int rc, rc2, i; + struct obd_uuid tgt_uuid = lov->lov_tgts[index]->ltd_uuid; + struct obd_device *tgt_obd; + struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" }; + struct lustre_handle conn = {0, }; + struct obd_import *imp; +#ifdef __KERNEL__ + cfs_proc_dir_entry_t *lov_proc_dir; +#endif + int rc; ENTRY; - MOD_INC_USE_COUNT; - rc = class_connect(conn, obd, cluuid); - if (rc) - GOTO(out_dec, rc); - - /* We don't want to actually do the underlying connections more than - * once, so keep track. */ - lov->refcount++; - if (lov->refcount > 1) - RETURN(0); + if (!lov->lov_tgts[index]) + RETURN(-EINVAL); - exp = class_conn2export(conn); - INIT_LIST_HEAD(&exp->exp_lov_data.led_open_head); + tgt_obd = class_find_client_obd(&tgt_uuid, LUSTRE_OSC_NAME, + &obd->obd_uuid); - /* retrieve LOV metadata from MDS */ - rc = obd_connect(&mdc_conn, lov->mdcobd, NULL, recovd, recover); - if (rc) { - CERROR("cannot connect to mdc: rc = %d\n", rc); - GOTO(out_conn, rc); + if (!tgt_obd) { + CERROR("Target %s not attached\n", obd_uuid2str(&tgt_uuid)); + RETURN(-EINVAL); } - - rc = mdc_getlovinfo(obd, &mdc_conn, &req); - rc2 = obd_disconnect(&mdc_conn); - if (rc) { - CERROR("cannot get lov info %d\n", rc); - GOTO(out_conn, rc); + + if (!tgt_obd->obd_set_up) { + CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid)); + RETURN(-EINVAL); } - if (rc2) { - CERROR("error disconnecting from MDS %d\n", rc2); - GOTO(out_conn, rc = rc2); - } + if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX)) + data->ocd_index = index; - /* sanity... */ - if (req->rq_repmsg->bufcount < 2 || - req->rq_repmsg->buflens[0] < sizeof(*desc)) { - CERROR("LOV desc: invalid descriptor returned\n"); - GOTO(out_conn, rc = -EINVAL); - } + /* + * Divine LOV knows that OBDs under it are OSCs. + */ + imp = tgt_obd->u.cli.cl_import; - memcpy(desc, lustre_msg_buf(req->rq_repmsg, 0), sizeof(*desc)); - lov_unpackdesc(desc); + if (activate) { + tgt_obd->obd_no_recov = 0; + /* FIXME this is probably supposed to be + ptlrpc_set_import_active. Horrible naming. */ + ptlrpc_activate_import(imp); + } - if (req->rq_repmsg->buflens[1] < sizeof(*uuidarray)*desc->ld_tgt_count){ - CERROR("LOV desc: invalid uuid array returned\n"); - GOTO(out_conn, rc = -EINVAL); + 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); } - if (memcmp(obd->obd_uuid, desc->ld_uuid, sizeof(desc->ld_uuid))) { - CERROR("LOV desc: uuid %s not on mds device (%s)\n", - obd->obd_uuid, desc->ld_uuid); - GOTO(out_conn, rc = -EINVAL); + rc = obd_connect(&conn, tgt_obd, &lov_osc_uuid, data); + if (rc) { + 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); } - if (desc->ld_tgt_count > 1000) { - CERROR("LOV desc: target count > 1000 (%d)\n", - desc->ld_tgt_count); - GOTO(out_conn, rc = -EINVAL); + 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); } - /* 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. - */ - if ((desc->ld_default_stripe_count ? - desc->ld_default_stripe_count : desc->ld_tgt_count) * - desc->ld_default_stripe_size > ~0UL) { - CERROR("LOV: stripe width "LPU64"x%u > %lu on 32-bit system\n", - desc->ld_default_stripe_size, - desc->ld_default_stripe_count ? - desc->ld_default_stripe_count : desc->ld_tgt_count,~0UL); - GOTO(out_conn, rc = -EINVAL); - } - - lov->bufsize = sizeof(struct lov_tgt_desc) * desc->ld_tgt_count; - OBD_ALLOC(lov->tgts, lov->bufsize); - if (!lov->tgts) { - CERROR("Out of memory\n"); - GOTO(out_conn, rc = -ENOMEM); - } - - uuidarray = lustre_msg_buf(req->rq_repmsg, 1); - for (i = 0; i < desc->ld_tgt_count; i++) - memcpy(lov->tgts[i].uuid, uuidarray[i], sizeof(*uuidarray)); - - for (i = 0; i < desc->ld_tgt_count; i++) { - struct obd_device *tgt = class_uuid2obd(uuidarray[i]); - int rc2; - - if (!tgt) { - CERROR("Target %s not attached\n", uuidarray[i]); - GOTO(out_disc, rc = -EINVAL); + lov->lov_tgts[index]->ltd_reap = 0; + if (activate) { + lov->lov_tgts[index]->ltd_active = 1; + lov->desc.ld_active_tgt_count++; + } + 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); + cfs_proc_dir_entry_t *osc_symlink; + char name[MAX_STRING_SIZE]; + + LASSERT(osc_obd != NULL); + LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC); + LASSERT(osc_obd->obd_type->typ_name != NULL); + 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); + if (osc_symlink == NULL) { + CERROR("could not register LOV target " + "/proc/fs/lustre/%s/%s/target_obds/%s.", + obd->obd_type->typ_name, obd->obd_name, + osc_obd->obd_name); + lprocfs_remove(&lov_proc_dir); } + } +#endif - if (!(tgt->obd_flags & OBD_SET_UP)) { - CERROR("Target %s not set up\n", uuidarray[i]); - GOTO(out_disc, rc = -EINVAL); - } + rc = qos_add_tgt(obd, index); + if (rc) + CERROR("qos_add_tgt failed %d\n", rc); - rc = obd_connect(&lov->tgts[i].conn, tgt, NULL, recovd, - recover); + RETURN(0); +} - /* Register even if connect failed, so that we get reactivation - * notices. - */ - rc2 = obd_iocontrol(IOC_OSC_REGISTER_LOV, &lov->tgts[i].conn, - sizeof(struct obd_device *), obd, NULL); - if (rc2) { - CERROR("Target %s REGISTER_LOV error %d\n", - uuidarray[i], rc2); - GOTO(out_disc, rc2); - } +static int lov_connect(struct lustre_handle *conn, struct obd_device *obd, + struct obd_uuid *cluuid, struct obd_connect_data *data) +{ + struct lov_obd *lov = &obd->u.lov; + struct lov_tgt_desc *tgt; + int i, rc; + ENTRY; + + CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects); + + rc = class_connect(conn, obd, cluuid); + if (rc) + RETURN(rc); + + /* Why should there ever be more than 1 connect? */ + lov->lov_connects++; + LASSERT(lov->lov_connects == 1); + + memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd)); + if (data) + lov->lov_ocd = *data; - /* But mark failed-connect OSCs as inactive! */ + lov_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)) + continue; + /* Flags will be lowest common denominator */ + rc = lov_connect_obd(obd, i, lov->lov_tgts[i]->ltd_activate, + &lov->lov_ocd); if (rc) { - CDEBUG(D_INFO, "Target %s connect error %d\n", - uuidarray[i], rc); - LASSERT(lov->tgts[i].active == 0); - rc = 0; + CERROR("%s: lov connect tgt %d failed: %d\n", + obd->obd_name, i, rc); continue; } - - desc->ld_active_tgt_count++; - lov->tgts[i].active = 1; } + lov_putref(obd); + + RETURN(0); +} + +static int lov_disconnect_obd(struct obd_device *obd, __u32 index) +{ + 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); + int rc; + ENTRY; - mdc->cl_max_mds_easize = obd_size_wiremd(conn, NULL); + CDEBUG(D_CONFIG, "%s: disconnecting target %s\n", + obd->obd_name, osc_obd->obd_name); - out: - ptlrpc_req_finished(req); - RETURN(rc); + if (lov->lov_tgts[index]->ltd_active) { + lov->lov_tgts[index]->ltd_active = 0; + lov->desc.ld_active_tgt_count--; + } + + lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds"); + if (lov_proc_dir) { + cfs_proc_dir_entry_t *osc_symlink; - out_disc: - while (i-- > 0) { - desc->ld_active_tgt_count--; - lov->tgts[i].active = 0; - rc2 = obd_disconnect(&lov->tgts[i].conn); - if (rc2) - CERROR("LOV Target %s disconnect error: rc = %d\n", - uuidarray[i], rc2); + osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name); + if (osc_symlink) { + lprocfs_remove(&osc_symlink); + } else { + CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.", + obd->obd_type->typ_name, obd->obd_name, + osc_obd->obd_name); + } + } + + if (obd->obd_no_recov) { + /* Pass it on to our clients. + * XXX This should be an argument to disconnect, + * XXX not a back-door flag on the OBD. Ah well. + */ + if (osc_obd) + osc_obd->obd_no_recov = 1; + } + + obd_register_observer(osc_obd, NULL); + + rc = obd_disconnect(lov->lov_tgts[index]->ltd_exp); + if (rc) { + CERROR("Target %s disconnect error %d\n", + lov_uuid2str(lov, index), rc); + rc = 0; } - OBD_FREE(lov->tgts, lov->bufsize); - out_conn: - class_disconnect(conn); - out_dec: - MOD_DEC_USE_COUNT; - goto out; + + qos_del_tgt(obd, index); + + lov->lov_tgts[index]->ltd_exp = NULL; + RETURN(0); } -static int lov_disconnect(struct lustre_handle *conn) +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_conn2obd(conn); + struct obd_device *obd = class_exp2obd(exp); struct lov_obd *lov = &obd->u.lov; - struct obd_export *exp; - struct list_head *p, *n; - int rc, i; + int i, rc; + ENTRY; - if (!lov->tgts) - goto out_local; + if (!lov->lov_tgts) + goto out; /* Only disconnect the underlying layers on the final disconnect. */ - lov->refcount--; - if (lov->refcount != 0) - goto out_local; + lov->lov_connects--; + if (lov->lov_connects != 0) { + /* why should there be more than 1 connect? */ + CERROR("disconnect #%d\n", lov->lov_connects); + goto out; + } + /* Let's hold another reference so lov_del_obd doesn't spin through + putref every time */ + lov_getref(obd); for (i = 0; i < lov->desc.ld_tgt_count; i++) { - rc = obd_disconnect(&lov->tgts[i].conn); - if (rc) { - if (lov->tgts[i].active) { - CERROR("Target %s disconnect error %d\n", - lov->tgts[i].uuid, rc); - } - rc = 0; - } - if (lov->tgts[i].active) { - lov->desc.ld_active_tgt_count--; - lov->tgts[i].active = 0; + 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); } } - OBD_FREE(lov->tgts, lov->bufsize); - lov->bufsize = 0; - lov->tgts = NULL; - - exp = class_conn2export(conn); - list_for_each_safe(p, n, &exp->exp_lov_data.led_open_head) { - /* XXX close these, instead of just discarding them? */ - struct lov_file_handles *lfh; - lfh = list_entry(p, typeof(*lfh), lfh_list); - CERROR("discarding open LOV handle %p:"LPX64"\n", - lfh, lfh->lfh_cookie); - list_del(&lfh->lfh_list); - OBD_FREE(lfh->lfh_handles, - lfh->lfh_count * sizeof(*lfh->lfh_handles)); - kmem_cache_free(lov_file_cache, lfh); - } - - out_local: - rc = class_disconnect(conn); - if (!rc) - MOD_DEC_USE_COUNT; - return rc; + lov_putref(obd); + +out: + rc = class_disconnect(exp); /* bz 9811 */ + RETURN(rc); } /* Error codes: @@ -279,1008 +345,1698 @@ static int lov_disconnect(struct lustre_handle *conn) * -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 (!) - * -EALREADY: The OSC is already marked (in)active */ -static int lov_set_osc_active(struct lov_obd *lov, obd_uuid_t uuid, +static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid, int activate) { - struct obd_device *obd; + struct lov_obd *lov = &obd->u.lov; + struct lov_tgt_desc *tgt; int i, rc = 0; ENTRY; CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n", - lov, uuid, activate); + lov, uuid->uuid, activate); + + lov_getref(obd); + for (i = 0; i < lov->desc.ld_tgt_count; i++) { + tgt = lov->lov_tgts[i]; + if (!tgt || !tgt->ltd_exp) + continue; - spin_lock(&lov->lov_lock); - for (i = 0; i < lov->desc.ld_tgt_count; i++) - if (strncmp(uuid, lov->tgts[i].uuid, - sizeof(lov->tgts[i].uuid)) == 0) + CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n", + i, 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); - obd = class_conn2obd(&lov->tgts[i].conn); - if (obd == NULL) { - LBUG(); - GOTO(out, rc = -ENOTCONN); - } - - CDEBUG(D_INFO, "Found OBD %p type %s\n", obd, obd->obd_type->typ_name); - if (strcmp(obd->obd_type->typ_name, "osc") != 0) { - LBUG(); - GOTO(out, rc = -EBADF); - } - - if (lov->tgts[i].active == activate) { - CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd, + if (lov->lov_tgts[i]->ltd_active == activate) { + CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid, activate ? "" : "in"); - GOTO(out, rc = -EALREADY); + GOTO(out, rc); } - CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd, activate ? "" : "in"); + CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid), + activate ? "" : "in"); - lov->tgts[i].active = activate; - if (activate) { - /* - * foreach(export) - * foreach(open_file) - * if (file_handle uses this_osc) - * if (has_no_filehandle) - * open(file_handle, this_osc); - */ - /* XXX reconnect? */ + lov->lov_tgts[i]->ltd_active = activate; + if (activate) lov->desc.ld_active_tgt_count++; - } else { - /* - * Should I invalidate filehandles that refer to this OSC, so - * that I reopen them during reactivation? - */ - /* XXX disconnect from OSC? */ + else lov->desc.ld_active_tgt_count--; - } + /* remove any old qos penalty */ + lov->lov_tgts[i]->ltd_qos.ltq_penalty = 0; - EXIT; out: - spin_unlock(&lov->lov_lock); - return rc; + lov_putref(obd); + RETURN(rc); } -static int lov_setup(struct obd_device *obd, obd_count len, void *buf) +static int lov_notify(struct obd_device *obd, struct obd_device *watched, + enum obd_notify_event ev, void *data) { - struct obd_ioctl_data *data = buf; - struct lov_obd *lov = &obd->u.lov; int rc = 0; ENTRY; - if (data->ioc_inllen1 < 1) { - CERROR("LOV setup requires an MDC UUID\n"); - RETURN(-EINVAL); - } + if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) { + struct obd_uuid *uuid; - if (data->ioc_inllen1 > 37) { - CERROR("mdc UUID must be 36 characters or less\n"); - RETURN(-EINVAL); - } + LASSERT(watched); + + if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) { + CERROR("unexpected notification of %s %s!\n", + watched->obd_type->typ_name, + watched->obd_name); + RETURN(-EINVAL); + } + uuid = &watched->u.cli.cl_target_uuid; - spin_lock_init(&lov->lov_lock); - lov->mdcobd = class_uuid2obd(data->ioc_inlbuf1); - if (!lov->mdcobd) { - CERROR("LOV %s cannot locate MDC %s\n", obd->obd_uuid, - data->ioc_inlbuf1); - rc = -EINVAL; + /* Set OSC as active before notifying the observer, so the + * observer can use the OSC normally. + */ + rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE); + if (rc) { + CERROR("%sactivation of %s failed: %d\n", + (ev == OBD_NOTIFY_ACTIVE) ? "" : "de", + obd_uuid2str(uuid), rc); + RETURN(rc); + } } - RETURN(rc); -} - -static struct lov_file_handles *lov_handle2lfh(struct lustre_handle *handle) -{ - struct lov_file_handles *lfh = NULL; - - if (!handle || !handle->addr) - RETURN(NULL); - lfh = (struct lov_file_handles *)(unsigned long)(handle->addr); - if (!kmem_cache_validate(lov_file_cache, lfh)) - RETURN(NULL); - - if (lfh->lfh_cookie != handle->cookie) - RETURN(NULL); + /* Pass the notification up the chain. */ + if (watched) { + rc = obd_notify_observer(obd, watched, ev, data); + } else { + /* NULL watched means all osc's in the lov (only for syncs) */ + struct lov_obd *lov = &obd->u.lov; + struct obd_device *tgt_obd; + int i; + lov_getref(obd); + for (i = 0; i < lov->desc.ld_tgt_count; i++) { + if (!lov->lov_tgts[i]) + continue; + 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", + obd->obd_name, + obd->obd_observer->obd_name, + tgt_obd->obd_name, rc); + break; + } + } + lov_putref(obd); + } - return lfh; + RETURN(rc); } -/* the LOV expects oa->o_id to be set to the LOV object id */ -static int lov_create(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md **ea) +static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp, + __u32 index, int gen, int active) { - struct obd_export *export = class_conn2export(conn); - struct lov_obd *lov; - struct lov_stripe_md *lsm; - struct lov_oinfo *loi; - struct obdo *tmp; - int ost_count, ost_idx = 1; - int rc = 0, i; + struct lov_obd *lov = &obd->u.lov; + struct lov_tgt_desc *tgt; + int rc; ENTRY; - LASSERT(ea); + CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n", + uuidp->uuid, index, gen, active); - if (!export) + if (gen <= 0) { + CERROR("request to add OBD %s with invalid generation: %d\n", + uuidp->uuid, gen); RETURN(-EINVAL); + } - tmp = obdo_alloc(); - if (!tmp) - RETURN(-ENOMEM); - - lov = &export->exp_obd->u.lov; - - if (!lov->desc.ld_active_tgt_count) - RETURN(-EIO); - - spin_lock(&lov->lov_lock); - ost_count = lov->desc.ld_tgt_count; + mutex_down(&lov->lov_lock); - lsm = *ea; + 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); + RETURN(-EEXIST); + } - /* Can't create more stripes than we have targets (incl inactive). */ - if (lsm && lsm->lsm_stripe_count > lov->desc.ld_tgt_count) - GOTO(out_tmp, rc = -EINVAL); + if (index >= lov->lov_tgt_size) { + /* We need to reallocate the lov target array. */ + struct lov_tgt_desc **newtgts, **old = NULL; + __u32 newsize, oldsize = 0; - /* Free the user lsm if it needs to be changed, to avoid memory leaks */ - if (!lsm || (lsm && - lsm->lsm_stripe_count > lov->desc.ld_active_tgt_count)) { - struct lov_stripe_md *lsm_new = NULL; - rc = obd_alloc_memmd(conn, &lsm_new); - if (rc < 0) { - spin_unlock(&lov->lov_lock); - if (lsm) - obd_free_memmd(conn, &lsm); - GOTO(out_tmp, rc); + newsize = max(lov->lov_tgt_size, (__u32)2); + while (newsize < index + 1) + newsize = newsize << 1; + OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize); + if (newtgts == NULL) { + mutex_up(&lov->lov_lock); + RETURN(-ENOMEM); } - if (lsm) { - LASSERT(lsm->lsm_magic == LOV_MAGIC); - CERROR("replace user LOV MD: stripes %u > %u active\n", - lsm->lsm_stripe_count, - lov->desc.ld_active_tgt_count); - lsm_new->lsm_stripe_offset = lsm->lsm_stripe_offset; - lsm_new->lsm_stripe_size = lsm->lsm_stripe_size; - lsm_new->lsm_stripe_pattern = lsm->lsm_stripe_pattern; - obd_free_memmd(conn, &lsm); + + if (lov->lov_tgt_size) { + memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) * + lov->lov_tgt_size); + old = lov->lov_tgts; + oldsize = lov->lov_tgt_size; } - lsm = lsm_new; - ost_idx = 0; /* if lsm->lsm_stripe_offset is set yet */ - lsm->lsm_magic = LOV_MAGIC; - } - LASSERT(oa->o_valid & OBD_MD_FLID); - lsm->lsm_object_id = oa->o_id; - if (!lsm->lsm_stripe_size) - lsm->lsm_stripe_size = lov->desc.ld_default_stripe_size; + lov->lov_tgts = newtgts; + lov->lov_tgt_size = newsize; +#ifdef __KERNEL__ + smp_rmb(); +#endif + if (old) + OBD_FREE(old, sizeof(*old) * oldsize); - /* 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. - */ - if (lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL) { - CERROR("LOV: stripe width "LPU64"x%u > %lu on 32-bit system\n", - lsm->lsm_stripe_size, lsm->lsm_stripe_count, ~0UL); - spin_unlock(&lov->lov_lock); - GOTO(out_free, rc = -EINVAL); - } + CDEBUG(D_CONFIG, "tgts: %p size: %d\n", + lov->lov_tgts, lov->lov_tgt_size); + } - if (!ost_idx || lsm->lsm_stripe_offset >= ost_count) { - int mult = lsm->lsm_object_id * lsm->lsm_stripe_count; - int stripe_offset = mult % ost_count; - int sub_offset = (mult / ost_count) % lsm->lsm_stripe_count; - lsm->lsm_stripe_offset = stripe_offset + sub_offset; + OBD_ALLOC_PTR(tgt); + if (!tgt) { + mutex_up(&lov->lov_lock); + RETURN(-ENOMEM); } - /* Start with lsm_stripe_offset on an active OSC to avoid confusion */ - while (!lov->tgts[lsm->lsm_stripe_offset].active) - lsm->lsm_stripe_offset = (lsm->lsm_stripe_offset+1) % ost_count; + memset(tgt, 0, sizeof(*tgt)); + tgt->ltd_uuid = *uuidp; + /* XXX - add a sanity check on the generation number. */ + tgt->ltd_gen = gen; + tgt->ltd_index = index; + tgt->ltd_activate = active; + lov->lov_tgts[index] = tgt; + if (index >= lov->desc.ld_tgt_count) + lov->desc.ld_tgt_count = index + 1; + mutex_up(&lov->lov_lock); - /* Pick the OSTs before we release the lock */ - ost_idx = lsm->lsm_stripe_offset; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - CDEBUG(D_INODE, "objid "LPX64"[%d] is ost_idx %d (uuid %s)\n", - lsm->lsm_object_id, i, ost_idx, lov->tgts[ost_idx].uuid); - loi->loi_ost_idx = ost_idx; - do { - ost_idx = (ost_idx + 1) % ost_count; - } while (!lov->tgts[ost_idx].active); + CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n", + index, tgt->ltd_gen, lov->desc.ld_tgt_count); + + if (lov->lov_connects == 0) { + /* lov_connect hasn't been called yet. We'll do the + lov_connect_obd on this target when that fn first runs, + because we don't know the connect flags yet. */ + RETURN(0); } - spin_unlock(&lov->lov_lock); - - CDEBUG(D_INODE, "allocating %d subobjs for objid "LPX64" at idx %d\n", - lsm->lsm_stripe_count,lsm->lsm_object_id,lsm->lsm_stripe_offset); + lov_getref(obd); - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - struct lov_stripe_md obj_md; - struct lov_stripe_md *obj_mdp = &obj_md; + rc = lov_connect_obd(obd, index, active, &lov->lov_ocd); + if (rc) + GOTO(out, rc); - ost_idx = loi->loi_ost_idx; + rc = lov_notify(obd, tgt->ltd_exp->exp_obd, + active ? OBD_NOTIFY_ACTIVE : OBD_NOTIFY_INACTIVE, + (void *)&index); - /* create data objects with "parent" OA */ - memcpy(tmp, oa, sizeof(*tmp)); - /* XXX: LOV STACKING: use real "obj_mdp" sub-data */ - rc = obd_create(&lov->tgts[ost_idx].conn, tmp, &obj_mdp); - if (rc) { - CERROR("error creating objid "LPX64" sub-object on " - "OST idx %d: rc = %d\n", oa->o_id, ost_idx, rc); - GOTO(out_cleanup, rc); - } - loi->loi_id = tmp->o_id; - CDEBUG(D_INODE, "objid "LPX64" has subobj "LPX64" at idx %d\n", - lsm->lsm_object_id, loi->loi_id, ost_idx); +out: + if (rc) { + CERROR("add failed (%d), deleting %s\n", rc, + obd_uuid2str(&tgt->ltd_uuid)); + lov_del_target(obd, index, 0, 0); } - - *ea = lsm; - - out_tmp: - obdo_free(tmp); + lov_putref(obd); RETURN(rc); - - out_cleanup: - while (i-- > 0) { - int err; - - --loi; - /* destroy already created objects here */ - memcpy(tmp, oa, sizeof(*tmp)); - tmp->o_id = loi->loi_id; - err = obd_destroy(&lov->tgts[loi->loi_ost_idx].conn, tmp, NULL); - if (err) - CERROR("Failed to uncreate objid "LPX64" subobj " - LPX64" on OST idx %d: rc = %d\n", - oa->o_id, loi->loi_id, loi->loi_ost_idx, - err); - } - out_free: - if (!*ea) - obd_free_memmd(conn, &lsm); - goto out_tmp; } -static int lov_destroy(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *lsm) +/* Schedule a target for deletion */ +static int lov_del_target(struct obd_device *obd, __u32 index, + struct obd_uuid *uuidp, int gen) { - struct obdo tmp; - struct obd_export *export = class_conn2export(conn); - struct lov_obd *lov; - struct lov_oinfo *loi; - struct lov_file_handles *lfh = NULL; - int rc = 0, i; + struct lov_obd *lov = &obd->u.lov; + int count = lov->desc.ld_tgt_count; + int rc = 0; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea for destruction\n"); + if (index >= count) { + CERROR("LOV target index %d >= number of LOV OBDs %d.\n", + index, count); RETURN(-EINVAL); } - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); - RETURN(-EINVAL); + lov_getref(obd); + + if (!lov->lov_tgts[index]) { + CERROR("LOV target at index %d is not setup.\n", index); + GOTO(out, rc = -EINVAL); } - if (!export || !export->exp_obd) - RETURN(-ENODEV); + if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) { + CERROR("LOV target UUID %s at index %d doesn't match %s.\n", + lov_uuid2str(lov, index), index, + obd_uuid2str(uuidp)); + GOTO(out, rc = -EINVAL); + } - if (oa->o_valid & OBD_MD_FLHANDLE) - lfh = lov_handle2lfh(obdo_handle(oa)); + CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n", + lov_uuid2str(lov, index), index, + lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp, + lov->lov_tgts[index]->ltd_active); - lov = &export->exp_obd->u.lov; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - int err; - if (lov->tgts[loi->loi_ost_idx].active == 0) { - /* Orphan clean up will (someday) fix this up. */ - continue; - } + lov->lov_tgts[index]->ltd_reap = 1; + lov->lov_death_row++; + /* we really delete it from lov_putref */ +out: + lov_putref(obd); - memcpy(&tmp, oa, sizeof(tmp)); - tmp.o_id = loi->loi_id; - if (lfh) - memcpy(obdo_handle(&tmp), &lfh->lfh_handles[i], - sizeof(lfh->lfh_handles[i])); - else - tmp.o_valid &= ~OBD_MD_FLHANDLE; - err = obd_destroy(&lov->tgts[loi->loi_ost_idx].conn, &tmp, - NULL); - if (err && lov->tgts[loi->loi_ost_idx].active) { - CERROR("Error destroying objid "LPX64" subobj " - LPX64" on OST idx %d\n: rc = %d", - oa->o_id, loi->loi_id, loi->loi_ost_idx, err); - if (!rc) - rc = err; - } - } RETURN(rc); } -/* compute object size given "stripeno" and the ost size */ -static obd_size lov_stripe_size(struct lov_stripe_md *lsm, obd_size ost_size, - int stripeno) +/* We are holding lov_lock */ +static void __lov_del_obd(struct obd_device *obd, __u32 index) { - unsigned long ssize = lsm->lsm_stripe_size; - unsigned long swidth = ssize * lsm->lsm_stripe_count; - unsigned long stripe_size; - obd_size lov_size; + struct lov_obd *lov = &obd->u.lov; + struct obd_device *osc_obd; + struct lov_tgt_desc *tgt = lov->lov_tgts[index]; - if (ost_size == 0) - return 0; + LASSERT(tgt); + LASSERT(tgt->ltd_reap); - /* do_div(a, b) returns a % b, and a = a / b */ - stripe_size = do_div(ost_size, ssize); + osc_obd = class_exp2obd(tgt->ltd_exp); - if (stripe_size) - lov_size = ost_size * swidth + stripeno * ssize + stripe_size; - else - lov_size = (ost_size - 1) * swidth + (stripeno + 1) * ssize; + CDEBUG(D_CONFIG, "Removing tgt %s : %s\n", + lov_uuid2str(lov, index), + osc_obd ? osc_obd->obd_name : ""); - return lov_size; -} + if (tgt->ltd_exp) + lov_disconnect_obd(obd, index); -static void lov_merge_attrs(struct obdo *tgt, struct obdo *src, obd_flag valid, - struct lov_stripe_md *lsm, int stripeno, int *new) -{ - if (*new) { - obdo_cpy_md(tgt, src, valid); - if (valid & OBD_MD_FLSIZE) - tgt->o_size = lov_stripe_size(lsm,src->o_size,stripeno); - *new = 0; - } else { - if (valid & OBD_MD_FLSIZE) { - /* this handles sparse files properly */ - obd_size lov_size; + /* 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_size = lov_stripe_size(lsm, src->o_size, stripeno); - if (lov_size > tgt->o_size) - tgt->o_size = lov_size; - } - if (valid & OBD_MD_FLBLOCKS) - tgt->o_blocks += src->o_blocks; - if (valid & OBD_MD_FLCTIME && tgt->o_ctime < src->o_ctime) - tgt->o_ctime = src->o_ctime; - if (valid & OBD_MD_FLMTIME && tgt->o_mtime < src->o_mtime) - tgt->o_mtime = src->o_mtime; + lov->lov_tgts[index] = NULL; + OBD_FREE_PTR(tgt); + + /* Manual cleanup - no cleanup logs to clean up the osc's. We must + do it ourselves. And we can't do it from lov_cleanup, + because we just lost our only reference to it. */ + if (osc_obd) { + /* Use lov's force/fail flags. */ + osc_obd->obd_force = obd->obd_force; + osc_obd->obd_fail = obd->obd_fail; + class_manual_cleanup(osc_obd); } } -static int lov_getattr(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *lsm) +void lov_fix_desc(struct lov_desc *desc) { - struct obdo tmp; - struct obd_export *export = class_conn2export(conn); - struct lov_obd *lov; - struct lov_oinfo *loi; - struct lov_file_handles *lfh = NULL; - int i; - int new = 1; + if (desc->ld_default_stripe_size < 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); + LCONSOLE_WARN("Changing default stripe size to "LPU64" (a " + "multiple of %u)\n", + desc->ld_default_stripe_size,LOV_MIN_STRIPE_SIZE); + } + + if (desc->ld_default_stripe_count == 0) + desc->ld_default_stripe_count = 1; + + /* 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; + } + /* fix qos_maxage */ + if (desc->ld_qos_maxage == 0) + desc->ld_qos_maxage = QOS_DEFAULT_MAXAGE; +} + +static int lov_setup(struct obd_device *obd, obd_count len, void *buf) +{ + struct lprocfs_static_vars lvars; + struct lustre_cfg *lcfg = buf; + struct lov_desc *desc; + struct lov_obd *lov = &obd->u.lov; + int count; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea\n"); + if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) { + CERROR("LOV setup requires a descriptor\n"); RETURN(-EINVAL); } - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); + desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1); + + if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) { + CERROR("descriptor size wrong: %d > %d\n", + (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1)); RETURN(-EINVAL); } - if (!export || !export->exp_obd) - RETURN(-ENODEV); - - lov = &export->exp_obd->u.lov; - - if (oa->o_valid & OBD_MD_FLHANDLE) - lfh = lov_handle2lfh(obdo_handle(oa)); - - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - int err; - - if (loi->loi_id == 0) - continue; + if (desc->ld_magic != LOV_DESC_MAGIC) { + if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) { + CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n", + obd->obd_name, desc); + lustre_swab_lov_desc(desc); + } else { + CERROR("%s: Bad lov desc magic: %#x\n", + obd->obd_name, desc->ld_magic); + RETURN(-EINVAL); + } + } - if (lov->tgts[loi->loi_ost_idx].active == 0) - continue; + lov_fix_desc(desc); - CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx " - "%u\n", oa->o_id, i, loi->loi_id, loi->loi_ost_idx); - /* create data objects with "parent" OA */ - memcpy(&tmp, oa, sizeof(tmp)); - tmp.o_id = loi->loi_id; - if (lfh) - memcpy(obdo_handle(&tmp), &lfh->lfh_handles[i], - sizeof(lfh->lfh_handles[i])); - else - tmp.o_valid &= ~OBD_MD_FLHANDLE; + /* 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); + } - err = obd_getattr(&lov->tgts[loi->loi_ost_idx].conn, &tmp,NULL); - if (err && lov->tgts[loi->loi_ost_idx].active) { - CERROR("Error getattr objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", - oa->o_id, loi->loi_id, loi->loi_ost_idx, err); - RETURN(err); + 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); + lov->lov_qos.lq_dirty = 1; + lov->lov_qos.lq_dirty_rr = 1; + lov->lov_qos.lq_reset = 1; + /* Default priority is toward free space balance */ + lov->lov_qos.lq_prio_free = 232; + + lprocfs_init_vars(lov, &lvars); + lprocfs_obd_setup(obd, lvars.obd_vars); +#ifdef LPROCFS + { + cfs_proc_dir_entry_t *entry; + + entry = create_proc_entry("target_obd", 0444, + obd->obd_proc_entry); + if (entry != NULL) { + entry->proc_fops = &lov_proc_target_fops; + entry->data = obd; } - lov_merge_attrs(oa, &tmp, tmp.o_valid, lsm, i, &new); } +#endif RETURN(0); } -static int lov_setattr(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *lsm) +static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) { - struct obdo *tmp; - struct obd_export *export = class_conn2export(conn); - struct lov_obd *lov; - struct lov_oinfo *loi; - struct lov_file_handles *lfh = NULL; - int rc = 0, i; + int rc = 0; ENTRY; - /* Note that this code is currently unused, hence LBUG(), just - * to know when/if it is ever revived that it needs cleanups. - */ - LBUG(); - - if (!lsm) { - CERROR("LOV requires striping ea\n"); - RETURN(-EINVAL); + 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) + continue; + obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp), + OBD_CLEANUP_EARLY); + } + break; } - - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); - RETURN(-EINVAL); + 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); +} - if (!export || !export->exp_obd) - RETURN(-ENODEV); - - /* size changes should go through punch and not setattr */ - LASSERT(!(oa->o_valid & OBD_MD_FLSIZE)); - - tmp = obdo_alloc(); - if (!tmp) - RETURN(-ENOMEM); +static int lov_cleanup(struct obd_device *obd) +{ + struct lov_obd *lov = &obd->u.lov; - if (oa->o_valid & OBD_MD_FLHANDLE) - lfh = lov_handle2lfh(obdo_handle(oa)); + lprocfs_obd_cleanup(obd); + if (lov->lov_tgts) { + int i; + 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); + } + } + 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); - lov = &export->exp_obd->u.lov; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - int err; + RETURN(0); +} - obdo_cpy_md(tmp, oa, oa->o_valid); +static int lov_process_config(struct obd_device *obd, obd_count len, void *buf) +{ + struct lustre_cfg *lcfg = buf; + struct obd_uuid obd_uuid; + int cmd; + int rc = 0; + ENTRY; - if (lfh) - memcpy(obdo_handle(tmp), &lfh->lfh_handles[i], - sizeof(lfh->lfh_handles[i])); + switch(cmd = lcfg->lcfg_command) { + case LCFG_LOV_ADD_OBD: + case LCFG_LOV_ADD_INA: + case LCFG_LOV_DEL_OBD: { + __u32 index; + int gen; + /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */ + 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); + if (cmd == LCFG_LOV_ADD_OBD) + rc = lov_add_target(obd, &obd_uuid, index, gen, 1); + else if (cmd == LCFG_LOV_ADD_INA) + rc = lov_add_target(obd, &obd_uuid, index, gen, 0); else - tmp->o_valid &= ~OBD_MD_FLHANDLE; - - tmp->o_id = loi->loi_id; + rc = lov_del_target(obd, index, &obd_uuid, gen); + GOTO(out, rc); + } + case LCFG_PARAM: { + struct lprocfs_static_vars lvars; + struct lov_desc *desc = &(obd->u.lov.desc); + + if (!desc) + GOTO(out, rc = -EINVAL); + + lprocfs_init_vars(lov, &lvars); + + rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars, + lcfg, obd); + GOTO(out, rc); + } + default: { + CERROR("Unknown command: %d\n", lcfg->lcfg_command); + GOTO(out, rc = -EINVAL); - err = obd_setattr(&lov->tgts[loi->loi_ost_idx].conn, tmp, NULL); - if (err) { - CERROR("Error setattr objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", - oa->o_id, loi->loi_id, loi->loi_ost_idx, err); - if (!rc) - rc = err; - } } - obdo_free(tmp); + } +out: RETURN(rc); } -static int lov_open(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *lsm) +#ifndef log2 +#define log2(n) ffz(~(n)) +#endif + +static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa, + struct lov_stripe_md **ea, + struct obd_trans_info *oti) { - struct obdo *tmp; /* on the heap here, on the stack in lov_close? */ - struct obd_export *export = class_conn2export(conn); struct lov_obd *lov; - struct lov_oinfo *loi; - struct lov_file_handles *lfh = NULL; - struct lustre_handle *handle; - int new = 1; + struct obdo *tmp_oa; + struct obd_uuid *ost_uuid = NULL; int rc = 0, i; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea for opening\n"); - RETURN(-EINVAL); - } - - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); - RETURN(-EINVAL); - } + LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS && + src_oa->o_flags == OBD_FL_DELORPHAN); - if (!export || !export->exp_obd) - RETURN(-ENODEV); + lov = &export->exp_obd->u.lov; - tmp = obdo_alloc(); - if (!tmp) + tmp_oa = obdo_alloc(); + if (tmp_oa == NULL) RETURN(-ENOMEM); - lfh = kmem_cache_alloc(lov_file_cache, GFP_KERNEL); - if (!lfh) - GOTO(out_tmp, rc = -ENOMEM); - OBD_ALLOC(lfh->lfh_handles, - lsm->lsm_stripe_count * sizeof(*lfh->lfh_handles)); - if (!lfh->lfh_handles) - GOTO(out_lfh, rc = -ENOMEM); + if (src_oa->o_valid & OBD_MD_FLINLINE) { + ost_uuid = (struct obd_uuid *)src_oa->o_inline; + CDEBUG(D_HA, "clearing orphans only for %s\n", + ost_uuid->uuid); + } - lov = &export->exp_obd->u.lov; - oa->o_size = 0; - oa->o_blocks = 0; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { + lov_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; + struct lov_tgt_desc *tgt; + int err; - if (lov->tgts[loi->loi_ost_idx].active == 0) { + tgt = lov->lov_tgts[i]; + if (!tgt) + continue; + + /* if called for a specific target, we don't + care if it is not active. */ + if (!lov->lov_tgts[i]->ltd_active == 0 && ost_uuid == NULL) { + CDEBUG(D_HA, "lov idx %d inactive\n", i); continue; } - /* create data objects with "parent" OA */ - memcpy(tmp, oa, sizeof(*tmp)); - tmp->o_id = loi->loi_id; + if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid)) + continue; - rc = obd_open(&lov->tgts[loi->loi_ost_idx].conn, tmp, NULL); - if (rc && lov->tgts[loi->loi_ost_idx].active) { - CERROR("Error open objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", - oa->o_id, lsm->lsm_oinfo[i].loi_id, - loi->loi_ost_idx, rc); - goto out_handles; - } + CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i, + obd_uuid2str(ost_uuid)); - lov_merge_attrs(oa, tmp, tmp->o_valid, lsm, i, &new); + memcpy(tmp_oa, src_oa, sizeof(*tmp_oa)); - if (tmp->o_valid & OBD_MD_FLHANDLE) - memcpy(&lfh->lfh_handles[i], obdo_handle(tmp), - sizeof(lfh->lfh_handles[i])); - } + LASSERT(lov->lov_tgts[i]->ltd_exp); + /* 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) + /* 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); - handle = obdo_handle(oa); - - lfh->lfh_count = lsm->lsm_stripe_count; - get_random_bytes(&lfh->lfh_cookie, sizeof(lfh->lfh_cookie)); - - handle->addr = (__u64)(unsigned long)lfh; - handle->cookie = lfh->lfh_cookie; - oa->o_valid |= OBD_MD_FLHANDLE; - list_add(&lfh->lfh_list, &export->exp_lov_data.led_open_head); + if (ost_uuid) + break; + } + lov_putref(export->exp_obd); -out_tmp: - obdo_free(tmp); + obdo_free(tmp_oa); RETURN(rc); +} -out_handles: - for (i--, loi = &lsm->lsm_oinfo[i]; i >= 0; i--, loi--) { - int err; +static int lov_recreate(struct obd_export *exp, struct obdo *src_oa, + struct lov_stripe_md **ea, struct obd_trans_info *oti) +{ + struct lov_stripe_md *obj_mdp, *lsm; + struct lov_obd *lov = &exp->exp_obd->u.lov; + unsigned ost_idx; + int rc, i; + ENTRY; - if (lov->tgts[loi->loi_ost_idx].active == 0) - continue; + LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS && + src_oa->o_flags & OBD_FL_RECREATE_OBJS); + + OBD_ALLOC(obj_mdp, sizeof(*obj_mdp)); + if (obj_mdp == NULL) + RETURN(-ENOMEM); - memcpy(tmp, oa, sizeof(*tmp)); - tmp->o_id = loi->loi_id; - memcpy(obdo_handle(tmp), &lfh->lfh_handles[i], - sizeof(lfh->lfh_handles[i])); + ost_idx = src_oa->o_nlink; + lsm = *ea; + if (lsm == NULL) + GOTO(out, rc = -EINVAL); + if (ost_idx >= lov->desc.ld_tgt_count || + !lov->lov_tgts[ost_idx]) + GOTO(out, rc = -EINVAL); - err = obd_close(&lov->tgts[loi->loi_ost_idx].conn, tmp, NULL); - if (err) { - CERROR("Error closing objid "LPX64" subobj "LPX64 - " on OST idx %d after open error: rc = %d\n", - oa->o_id, loi->loi_id, loi->loi_ost_idx, err); + for (i = 0; i < lsm->lsm_stripe_count; i++) { + if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) { + if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id) + GOTO(out, rc = -EINVAL); + break; } } - - OBD_FREE(lfh->lfh_handles, - lsm->lsm_stripe_count * sizeof(*lfh->lfh_handles)); -out_lfh: - lfh->lfh_cookie = DEAD_HANDLE_MAGIC; - kmem_cache_free(lov_file_cache, lfh); - goto out_tmp; + if (i == lsm->lsm_stripe_count) + GOTO(out, rc = -EINVAL); + + rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti); +out: + OBD_FREE(obj_mdp, sizeof(*obj_mdp)); + RETURN(rc); } -static int lov_close(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *lsm) +/* the LOV expects oa->o_id to be set to the LOV object id */ +static int lov_create(struct obd_export *exp, struct obdo *src_oa, + struct lov_stripe_md **ea, struct obd_trans_info *oti) { - struct obdo tmp; - struct obd_export *export = class_conn2export(conn); struct lov_obd *lov; - struct lov_oinfo *loi; - struct lov_file_handles *lfh = NULL; - int rc = 0, i; + struct obd_info oinfo; + struct lov_request_set *set = NULL; + struct lov_request *req; + struct obd_statfs osfs; + __u64 maxage; + int rc = 0; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea\n"); + LASSERT(ea != NULL); + if (exp == NULL) RETURN(-EINVAL); + + if ((src_oa->o_valid & OBD_MD_FLFLAGS) && + src_oa->o_flags == OBD_FL_DELORPHAN) { + rc = lov_clear_orphans(exp, src_oa, ea, oti); + RETURN(rc); } - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); - RETURN(-EINVAL); + lov = &exp->exp_obd->u.lov; + if (!lov->desc.ld_active_tgt_count) + RETURN(-EIO); + + /* 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); + } + + maxage = cfs_time_shift_64(-lov->desc.ld_qos_maxage); + obd_statfs_rqset(exp->exp_obd, &osfs, maxage); + + rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set); + if (rc) + RETURN(rc); + + 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 = lov_fini_create_set(set, ea); + RETURN(rc); +} - if (!export || !export->exp_obd) +#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); \ +} 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 lov_request_set *set; + struct obd_info oinfo; + struct lov_request *req; + struct list_head *pos; + struct lov_obd *lov; + int rc = 0, err; + ENTRY; + + ASSERT_LSM_MAGIC(lsm); + + if (!exp || !exp->exp_obd) RETURN(-ENODEV); - if (oa->o_valid & OBD_MD_FLHANDLE) - lfh = lov_handle2lfh(obdo_handle(oa)); + if (oa->o_valid & OBD_MD_FLCOOKIE) { + LASSERT(oti); + LASSERT(oti->oti_logcookies); + } - lov = &export->exp_obd->u.lov; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { + lov = &exp->exp_obd->u.lov; + rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set); + if (rc) + RETURN(rc); + + list_for_each (pos, &set->set_list) { int err; - - if (lov->tgts[loi->loi_ost_idx].active == 0) - continue; + req = list_entry(pos, struct lov_request, rq_link); - /* create data objects with "parent" OA */ - memcpy(&tmp, oa, sizeof(tmp)); - tmp.o_id = loi->loi_id; - if (lfh) - memcpy(obdo_handle(&tmp), &lfh->lfh_handles[i], - sizeof(lfh->lfh_handles[i])); - else - tmp.o_valid &= ~OBD_MD_FLHANDLE; + if (oa->o_valid & OBD_MD_FLCOOKIE) + oti->oti_logcookies = set->set_cookies + req->rq_stripe; - err = obd_close(&lov->tgts[loi->loi_ost_idx].conn, &tmp, NULL); + err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp, + req->rq_oi.oi_oa, NULL, oti, NULL); + err = lov_update_common_set(set, req, err); if (err) { - CERROR("Error close objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", - oa->o_id, loi->loi_id, loi->loi_ost_idx, err); + CERROR("error: destroying objid "LPX64" subobj " + LPX64" on OST idx %d: rc = %d\n", + oa->o_id, req->rq_oi.oi_oa->o_id, + req->rq_idx, err); if (!rc) rc = err; } } - if (lfh) { - list_del(&lfh->lfh_list); - OBD_FREE(lfh->lfh_handles, - lsm->lsm_stripe_count * sizeof(*lfh->lfh_handles)); - lfh->lfh_cookie = DEAD_HANDLE_MAGIC; - kmem_cache_free(lov_file_cache, lfh); + + if (rc == 0) { + LASSERT(lsm_op_find(lsm->lsm_magic) != NULL); + rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp); } + err = lov_fini_destroy_set(set); + RETURN(rc ? rc : err); +} +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; + struct lov_obd *lov; + int err = 0, rc = 0; + ENTRY; + + LASSERT(oinfo); + ASSERT_LSM_MAGIC(oinfo->oi_md); + + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + lov = &exp->exp_obd->u.lov; + + rc = lov_prep_getattr_set(exp, oinfo, &set); + if (rc) + RETURN(rc); + + list_for_each (pos, &set->set_list) { + req = 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, + req->rq_oi.oi_oa->o_id, req->rq_idx); + + rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp, + &req->rq_oi); + err = lov_update_common_set(set, req, rc); + if (err) { + CERROR("error: getattr objid "LPX64" subobj " + LPX64" on OST idx %d: rc = %d\n", + oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id, + req->rq_idx, err); + break; + } + } + + rc = lov_fini_getattr_set(set); + if (err) + rc = err; RETURN(rc); } -#ifndef log2 -#define log2(n) ffz(~(n)) -#endif +static int lov_getattr_interpret(struct ptlrpc_request_set *rqset, + void *data, int rc) +{ + struct lov_request_set *lovset = (struct lov_request_set *)data; + int err; + ENTRY; -#warning FIXME: merge these two functions now that they are nearly the same + /* don't do attribute merge if this aysnc op failed */ + if (rc) + lovset->set_completes = 0; + err = lov_fini_getattr_set(lovset); + RETURN(rc ? rc : err); +} -/* compute ost offset in stripe "stripeno" corresponding to offset "lov_off" */ -static obd_off lov_stripe_offset(struct lov_stripe_md *lsm, obd_off lov_off, - int stripeno) +static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo, + struct ptlrpc_request_set *rqset) { - unsigned long ssize = lsm->lsm_stripe_size; - unsigned long swidth = ssize * lsm->lsm_stripe_count; - unsigned long stripe_off, this_stripe; + struct lov_request_set *lovset; + struct lov_obd *lov; + struct list_head *pos; + struct lov_request *req; + int rc = 0, err; + ENTRY; - if (lov_off == OBD_OBJECT_EOF || lov_off == 0) - return lov_off; + LASSERT(oinfo); + ASSERT_LSM_MAGIC(oinfo->oi_md); - /* do_div(a, b) returns a % b, and a = a / b */ - stripe_off = do_div(lov_off, swidth); + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + lov = &exp->exp_obd->u.lov; + + rc = lov_prep_getattr_set(exp, oinfo, &lovset); + if (rc) + RETURN(rc); + + CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n", + oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count, + oinfo->oi_md->lsm_stripe_size); - this_stripe = stripeno * ssize; - if (stripe_off <= this_stripe) - stripe_off = 0; - else { - stripe_off -= this_stripe; + list_for_each (pos, &lovset->set_list) { + req = list_entry(pos, struct lov_request, rq_link); - if (stripe_off > ssize) - stripe_off = ssize; + CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx " + "%u\n", oinfo->oi_oa->o_id, req->rq_stripe, + req->rq_oi.oi_oa->o_id, req->rq_idx); + rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp, + &req->rq_oi, rqset); + if (rc) { + CERROR("error: getattr objid "LPX64" subobj " + LPX64" on OST idx %d: rc = %d\n", + oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id, + req->rq_idx, rc); + GOTO(out, rc); + } } + if (!list_empty(&rqset->set_requests)) { + LASSERT(rc == 0); + LASSERT (rqset->set_interpret == NULL); + rqset->set_interpret = lov_getattr_interpret; + rqset->set_arg = (void *)lovset; + RETURN(rc); + } +out: + if (rc) + lovset->set_completes = 0; + err = lov_fini_getattr_set(lovset); + RETURN(rc ? rc : err); +} + +static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo, + struct obd_trans_info *oti) +{ + struct lov_request_set *set; + struct lov_obd *lov; + struct list_head *pos; + struct lov_request *req; + int err = 0, rc = 0; + ENTRY; + + LASSERT(oinfo); + ASSERT_LSM_MAGIC(oinfo->oi_md); + + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + /* for now, we only expect the following updates here */ + LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE | + OBD_MD_FLMODE | OBD_MD_FLATIME | + 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))); + 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); + + rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp, + &req->rq_oi, NULL); + err = lov_update_setattr_set(set, req, rc); + if (err) { + CERROR("error: setattr objid "LPX64" subobj " + LPX64" on OST idx %d: rc = %d\n", + set->set_oi->oi_oa->o_id, + req->rq_oi.oi_oa->o_id, req->rq_idx, err); + if (!rc) + rc = err; + } + } + err = lov_fini_setattr_set(set); + if (!rc) + rc = err; + RETURN(rc); +} + +static int lov_setattr_interpret(struct ptlrpc_request_set *rqset, + void *data, int rc) +{ + struct lov_request_set *lovset = (struct lov_request_set *)data; + int err; + ENTRY; - return lov_off * ssize + stripe_off; + if (rc) + lovset->set_completes = 0; + err = lov_fini_setattr_set(lovset); + RETURN(rc ? rc : err); } -/* compute which stripe number "lov_off" will be written into */ -static int lov_stripe_number(struct lov_stripe_md *lsm, obd_off lov_off) +/* If @oti is given, the request goes from MDS and responses from OSTs are not + needed. Otherwise, a client is waiting for responses. */ +static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo, + struct obd_trans_info *oti, + struct ptlrpc_request_set *rqset) { - unsigned long ssize = lsm->lsm_stripe_size; - unsigned long swidth = ssize * lsm->lsm_stripe_count; - unsigned long stripe_off; + struct lov_request_set *set; + struct lov_request *req; + struct list_head *pos; + struct lov_obd *lov; + int rc = 0; + ENTRY; + + LASSERT(oinfo); + ASSERT_LSM_MAGIC(oinfo->oi_md); + if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) { + LASSERT(oti); + LASSERT(oti->oti_logcookies); + } - stripe_off = do_div(lov_off, swidth); + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + lov = &exp->exp_obd->u.lov; + rc = lov_prep_setattr_set(exp, oinfo, oti, &set); + if (rc) + RETURN(rc); - return stripe_off / ssize; + CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n", + 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); + + if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) + oti->oti_logcookies = set->set_cookies + req->rq_stripe; + + CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx " + "%u\n", oinfo->oi_oa->o_id, req->rq_stripe, + req->rq_oi.oi_oa->o_id, req->rq_idx); + + rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp, + &req->rq_oi, oti, rqset); + if (rc) { + CERROR("error: setattr objid "LPX64" subobj " + LPX64" on OST idx %d: rc = %d\n", + set->set_oi->oi_oa->o_id, + req->rq_oi.oi_oa->o_id, + req->rq_idx, rc); + break; + } + } + + /* If we are not waiting for responses on async requests, return. */ + if (rc || !rqset || list_empty(&rqset->set_requests)) { + int err; + if (rc) + set->set_completes = 0; + err = lov_fini_setattr_set(set); + RETURN(rc ? rc : err); + } + + LASSERT(rqset->set_interpret == NULL); + rqset->set_interpret = lov_setattr_interpret; + rqset->set_arg = (void *)set; + + RETURN(0); } +static int lov_punch_interpret(struct ptlrpc_request_set *rqset, + void *data, int rc) +{ + struct lov_request_set *lovset = (struct lov_request_set *)data; + int err; + ENTRY; + + if (rc) + lovset->set_completes = 0; + err = lov_fini_punch_set(lovset); + RETURN(rc ? rc : err); +} /* FIXME: maybe we'll just make one node the authoritative attribute node, then * we can send this 'punch' to just the authoritative node and the nodes * that the punch will affect. */ -static int lov_punch(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *lsm, - obd_off start, obd_off end) +static int lov_punch(struct obd_export *exp, struct obd_info *oinfo, + struct obd_trans_info *oti, + struct ptlrpc_request_set *rqset) { - struct obdo tmp; - struct obd_export *export = class_conn2export(conn); + struct lov_request_set *set; struct lov_obd *lov; - struct lov_oinfo *loi; - struct lov_file_handles *lfh = NULL; - int rc = 0, i; + struct list_head *pos; + struct lov_request *req; + int rc = 0; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea\n"); - RETURN(-EINVAL); + LASSERT(oinfo); + ASSERT_LSM_MAGIC(oinfo->oi_md); + + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + lov = &exp->exp_obd->u.lov; + rc = lov_prep_punch_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); + + rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp, + &req->rq_oi, NULL, rqset); + if (rc) { + CERROR("error: punch objid "LPX64" subobj "LPX64 + " on OST idx %d: rc = %d\n", + set->set_oi->oi_oa->o_id, + req->rq_oi.oi_oa->o_id, req->rq_idx, rc); + break; + } } - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); - RETURN(-EINVAL); + if (rc || list_empty(&rqset->set_requests)) { + int err; + err = lov_fini_punch_set(set); + RETURN(rc ? rc : err); } - if (!export || !export->exp_obd) - RETURN(-ENODEV); + LASSERT(rqset->set_interpret == NULL); + rqset->set_interpret = lov_punch_interpret; + rqset->set_arg = (void *)set; - if (oa->o_valid & OBD_MD_FLHANDLE) - lfh = lov_handle2lfh(obdo_handle(oa)); + RETURN(0); +} - lov = &export->exp_obd->u.lov; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - obd_off starti = lov_stripe_offset(lsm, start, i); - obd_off endi = lov_stripe_offset(lsm, end, i); - int err; +static int lov_sync(struct obd_export *exp, struct obdo *oa, + struct lov_stripe_md *lsm, obd_off start, obd_off end) +{ + struct lov_request_set *set; + struct obd_info oinfo; + struct lov_obd *lov; + struct list_head *pos; + struct lov_request *req; + int err = 0, rc = 0; + ENTRY; - if (starti == endi) - continue; - /* create data objects with "parent" OA */ - memcpy(&tmp, oa, sizeof(tmp)); - tmp.o_id = loi->loi_id; - if (lfh) - memcpy(obdo_handle(&tmp), &lfh->lfh_handles[i], - sizeof(lfh->lfh_handles[i])); - else - tmp.o_valid &= ~OBD_MD_FLHANDLE; + ASSERT_LSM_MAGIC(lsm); - err = obd_punch(&lov->tgts[loi->loi_ost_idx].conn, &tmp, NULL, - starti, endi); + if (!exp->exp_obd) + RETURN(-ENODEV); + + lov = &exp->exp_obd->u.lov; + rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set); + if (rc) + RETURN(rc); + + list_for_each (pos, &set->set_list) { + req = 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, + req->rq_oi.oi_policy.l_extent.start, + req->rq_oi.oi_policy.l_extent.end); + err = lov_update_common_set(set, req, rc); if (err) { - CERROR("Error punch objid "LPX64" subobj "LPX64 + CERROR("error: fsync objid "LPX64" subobj "LPX64 " on OST idx %d: rc = %d\n", - oa->o_id, loi->loi_id, loi->loi_ost_idx, err); + set->set_oi->oi_oa->o_id, + req->rq_oi.oi_oa->o_id, req->rq_idx, rc); if (!rc) rc = err; } } + err = lov_fini_sync_set(set); + if (!rc) + rc = err; RETURN(rc); } -static inline int lov_brw(int cmd, struct lustre_handle *conn, - struct lov_stripe_md *lsm, obd_count oa_bufs, - struct brw_page *pga, struct obd_brw_set *set) +static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo, + obd_count oa_bufs, struct brw_page *pga) { - struct { - int bufct; - int index; - int subcount; - struct lov_stripe_md lsm; - int ost_idx; - } *stripeinfo, *si, *si_last; - struct obd_export *export = class_conn2export(conn); - struct lov_obd *lov; - struct brw_page *ioarr; - struct lov_oinfo *loi; - int rc = 0, i, *where, stripe_count = lsm->lsm_stripe_count; + struct obd_info oinfo = { { { 0 } } }; + int i, rc = 0; + + oinfo.oi_oa = lov_oinfo->oi_oa; + + /* The caller just wants to know if there's a chance that this + * I/O can succeed */ + for (i = 0; i < oa_bufs; i++) { + int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off); + int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx; + obd_off start, end; + + if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off, + pga[i].off + pga[i].count, + &start, &end)) + continue; + + if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) { + CDEBUG(D_HA, "lov idx %d inactive\n", ost); + return -EIO; + } + + rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo, + 1, &pga[i], NULL); + if (rc) + break; + } + return rc; +} + +static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo, + obd_count oa_bufs, struct brw_page *pga, + struct obd_trans_info *oti) +{ + struct lov_request_set *set; + struct lov_request *req; + struct list_head *pos; + struct lov_obd *lov = &exp->exp_obd->u.lov; + int err, rc = 0; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea\n"); - RETURN(-EINVAL); + ASSERT_LSM_MAGIC(oinfo->oi_md); + + if (cmd == OBD_BRW_CHECK) { + rc = lov_brw_check(lov, oinfo, oa_bufs, pga); + RETURN(rc); } - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); - RETURN(-EINVAL); + rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set); + if (rc) + RETURN(rc); + + 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); + + sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp; + sub_pga = set->set_pga + req->rq_pgaidx; + rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs, + sub_pga, oti); + if (rc) + break; + lov_update_common_set(set, req, rc); } - lov = &export->exp_obd->u.lov; + err = lov_fini_brw_set(set); + if (!rc) + rc = err; + RETURN(rc); +} + +static int lov_brw_interpret(struct ptlrpc_request_set *reqset, 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); + } + + 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) +{ + struct lov_request_set *lovset; + struct lov_request *req; + struct list_head *pos; + struct lov_obd *lov = &exp->exp_obd->u.lov; + int rc = 0; + ENTRY; + + LASSERT(oinfo); + ASSERT_LSM_MAGIC(oinfo->oi_md); + + if (cmd == OBD_BRW_CHECK) { + rc = lov_brw_check(lov, oinfo, oa_bufs, pga); + RETURN(rc); + } + + rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset); + 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); + + 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) + 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); +out: + lov_fini_brw_set(lovset); + 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) +{ + struct lov_async_page *lap = LAP_FROM_COOKIE(data); + + return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data, + cmd); +} + +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_stripe_idx = lap->lap_stripe; +} + +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); + + lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid); +} + +static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc) +{ + struct lov_async_page *lap = LAP_FROM_COOKIE(data); + + /* 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; +} + +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, +}; + +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) +{ + struct lov_obd *lov = &exp->exp_obd->u.lov; + struct lov_async_page *lap; + int 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_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 (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); + RETURN(rc); +} + +static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo, + struct obd_enqueue_info *einfo) +{ + 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); + + /* we should never be asked to replay a lock this way. */ + LASSERT((einfo->ei_flags & LDLM_FL_REPLAY) == 0); - OBD_ALLOC(stripeinfo, stripe_count * sizeof(*stripeinfo)); - if (!stripeinfo) - GOTO(out_cbdata, rc = -ENOMEM); + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); - OBD_ALLOC(where, sizeof(*where) * oa_bufs); - if (!where) - GOTO(out_sinfo, rc = -ENOMEM); + lov = &exp->exp_obd->u.lov; + rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set); + if (rc) + RETURN(rc); - OBD_ALLOC(ioarr, sizeof(*ioarr) * oa_bufs); - if (!ioarr) - GOTO(out_where, rc = -ENOMEM); + list_for_each (pos, &set->set_list) { + req = list_entry(pos, struct lov_request, rq_link); - for (i = 0; i < oa_bufs; i++) { - where[i] = lov_stripe_number(lsm, pga[i].off); - stripeinfo[where[i]].bufct++; + rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp, + &req->rq_oi, einfo); + if (rc != ELDLM_OK) + GOTO(out, rc); } - for (i = 0, loi = lsm->lsm_oinfo, si_last = si = stripeinfo; - i < stripe_count; i++, loi++, si_last = si, si++) { - if (lov->tgts[loi->loi_ost_idx].active == 0) - GOTO(out_ioarr, rc = -EIO); - if (i > 0) - si->index = si_last->index + si_last->bufct; - si->lsm.lsm_object_id = loi->loi_id; - si->ost_idx = loi->loi_ost_idx; + if (einfo->ei_rqset && !list_empty(&einfo->ei_rqset->set_requests)) { + LASSERT(rc == 0); + LASSERT(einfo->ei_rqset->set_interpret == NULL); + einfo->ei_rqset->set_interpret = lov_enqueue_interpret; + einfo->ei_rqset->set_arg = (void *)set; + RETURN(rc); } +out: + rc = lov_fini_enqueue_set(set, einfo->ei_mode, rc); + RETURN(rc); +} - for (i = 0; i < oa_bufs; i++) { - int which = where[i]; - int shift; +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; - shift = stripeinfo[which].index + stripeinfo[which].subcount; - LASSERT(shift < oa_bufs); - ioarr[shift] = pga[i]; - ioarr[shift].off = lov_stripe_offset(lsm, pga[i].off, which); - stripeinfo[which].subcount++; - } + ASSERT_LSM_MAGIC(lsm); - for (i = 0, si = stripeinfo; i < stripe_count; i++, si++) { - int shift = si->index; + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); - if (si->bufct) { - LASSERT(shift < oa_bufs); - rc = obd_brw(cmd, &lov->tgts[si->ost_idx].conn, - &si->lsm, si->bufct, &ioarr[shift], set); - if (rc) - GOTO(out_ioarr, rc); - } + 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 != 1) + break; } - - out_ioarr: - OBD_FREE(ioarr, sizeof(*ioarr) * oa_bufs); - out_where: - OBD_FREE(where, sizeof(*where) * oa_bufs); - out_sinfo: - OBD_FREE(stripeinfo, stripe_count * sizeof(*stripeinfo)); - out_cbdata: + lov_fini_match_set(set, mode, *flags); RETURN(rc); } -static int lov_enqueue(struct lustre_handle *conn, struct lov_stripe_md *lsm, - struct lustre_handle *parent_lock, - __u32 type, void *cookie, int cookielen, __u32 mode, - int *flags, void *cb, void *data, int datalen, - struct lustre_handle *lockhs) +static int lov_change_cbdata(struct obd_export *exp, + struct lov_stripe_md *lsm, ldlm_iterator_t it, + void *data) { - struct obd_export *export = class_conn2export(conn); struct lov_obd *lov; struct lov_oinfo *loi; - struct lov_stripe_md submd; int rc = 0, i; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea\n"); - RETURN(-EINVAL); - } - - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); - RETURN(-EINVAL); - } - - /* we should never be asked to replay a lock. */ + ASSERT_LSM_MAGIC(lsm); - LASSERT((*flags & LDLM_FL_REPLAY) == 0); - - if (!export || !export->exp_obd) + if (!exp || !exp->exp_obd) RETURN(-ENODEV); - memset(lockhs, 0, sizeof(*lockhs) * lsm->lsm_stripe_count); - - lov = &export->exp_obd->u.lov; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { - struct ldlm_extent *extent = (struct ldlm_extent *)cookie; - struct ldlm_extent sub_ext; - - if (lov->tgts[loi->loi_ost_idx].active == 0) - continue; - - *flags = 0; - sub_ext.start = lov_stripe_offset(lsm, extent->start, i); - sub_ext.end = lov_stripe_offset(lsm, extent->end, i); - if (sub_ext.start == sub_ext.end) - continue; + lov = &exp->exp_obd->u.lov; + for (i = 0; i < lsm->lsm_stripe_count; i++) { + struct lov_stripe_md submd; + loi = lsm->lsm_oinfo[i]; submd.lsm_object_id = loi->loi_id; - /* XXX submd should be that from the subobj, it should come - * opaquely from the LOV. - */ submd.lsm_stripe_count = 0; - /* XXX submd is not fully initialized here */ - *flags = 0; - rc = obd_enqueue(&(lov->tgts[loi->loi_ost_idx].conn), &submd, - parent_lock, type, &sub_ext, sizeof(sub_ext), - mode, flags, cb, data, datalen, &(lockhs[i])); - // XXX add a lock debug statement here - if (rc && lov->tgts[loi->loi_ost_idx].active) { - CERROR("Error enqueue objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", lsm->lsm_object_id, - loi->loi_id, loi->loi_ost_idx, rc); - goto out_locks; - } + rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, + &submd, it, data); } + RETURN(rc); +} - RETURN(0); +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; - out_locks: - for (i--, loi = &lsm->lsm_oinfo[i]; i >= 0; i--, loi--) { - int err; - - if (lov->tgts[loi->loi_ost_idx].active == 0) - continue; + ASSERT_LSM_MAGIC(lsm); - submd.lsm_object_id = loi->loi_id; - submd.lsm_stripe_count = 0; - err = obd_cancel(&lov->tgts[loi->loi_ost_idx].conn, &submd, - mode, &lockhs[i]); - if (err) { - CERROR("Error cancelling objid "LPX64" subobj "LPX64 - " on OST idx %d after enqueue error: rc = %d\n", - loi->loi_id, loi->loi_ost_idx, err); + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); + + 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; + + rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp, + req->rq_oi.oi_md, mode, lov_lockhp); + rc = lov_update_common_set(set, req, rc); + if (rc) { + CERROR("error: cancel objid "LPX64" subobj " + LPX64" on OST idx %d: rc = %d\n", + lsm->lsm_object_id, + req->rq_oi.oi_md->lsm_object_id, + req->rq_idx, rc); + err = rc; } + } - RETURN(rc); + lov_fini_cancel_set(set); + RETURN(err); } -static int lov_cancel(struct lustre_handle *conn, struct lov_stripe_md *lsm, - __u32 mode, struct lustre_handle *lockhs) +static int lov_cancel_unused(struct obd_export *exp, + struct lov_stripe_md *lsm, int flags, void *opaque) { - struct obd_export *export = class_conn2export(conn); struct lov_obd *lov; struct lov_oinfo *loi; int rc = 0, i; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea\n"); - RETURN(-EINVAL); - } + if (!exp || !exp->exp_obd) + RETURN(-ENODEV); - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("LOV striping magic bad %#lx != %#lx\n", - lsm->lsm_magic, LOV_MAGIC); - RETURN(-EINVAL); + lov = &exp->exp_obd->u.lov; + if (lsm == NULL) { + for (i = 0; i < lov->desc.ld_tgt_count; i++) { + int err; + if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp) + continue; + + err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL, + flags, opaque); + if (!rc) + rc = err; + } + RETURN(rc); } - if (!export || !export->exp_obd) - RETURN(-ENODEV); + ASSERT_LSM_MAGIC(lsm); - lov = &export->exp_obd->u.lov; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { + for (i = 0; i < lsm->lsm_stripe_count; i++) { struct lov_stripe_md submd; int err; - if (lov->tgts[loi->loi_ost_idx].active == 0) - continue; - - if (lockhs[i].addr == 0) - continue; + 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); submd.lsm_object_id = loi->loi_id; submd.lsm_stripe_count = 0; - err = obd_cancel(&lov->tgts[loi->loi_ost_idx].conn, &submd, - mode, &lockhs[i]); - if (err && lov->tgts[loi->loi_ost_idx].active) { - CERROR("Error cancel objid "LPX64" subobj "LPX64 + err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, + &submd, flags, opaque); + if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) { + CERROR("error: cancel unused objid "LPX64" subobj "LPX64 " on OST idx %d: rc = %d\n", lsm->lsm_object_id, loi->loi_id, loi->loi_ost_idx, err); if (!rc) @@ -1290,117 +2046,152 @@ static int lov_cancel(struct lustre_handle *conn, struct lov_stripe_md *lsm, RETURN(rc); } -static int lov_cancel_unused(struct lustre_handle *conn, - struct lov_stripe_md *lsm, int flags) +static int lov_join_lru(struct obd_export *exp, + struct lov_stripe_md *lsm, int join) { - struct obd_export *export = class_conn2export(conn); struct lov_obd *lov; struct lov_oinfo *loi; - int rc = 0, i, err; + int i, count = 0; ENTRY; - if (!lsm) { - CERROR("LOV requires striping ea for lock cancellation\n"); - RETURN(-EINVAL); - } - - if (!export || !export->exp_obd) + ASSERT_LSM_MAGIC(lsm); + if (!exp || !exp->exp_obd) RETURN(-ENODEV); - lov = &export->exp_obd->u.lov; - for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) { + lov = &exp->exp_obd->u.lov; + for (i = 0; i < lsm->lsm_stripe_count; i++) { struct lov_stripe_md submd; + int rc = 0; + + 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); submd.lsm_object_id = loi->loi_id; submd.lsm_stripe_count = 0; - err = obd_cancel_unused(&lov->tgts[loi->loi_ost_idx].conn, - &submd, flags); - if (err && lov->tgts[loi->loi_ost_idx].active) { - CERROR("Error cancel unused objid "LPX64" subobj "LPX64 - " on OST idx %d: rc = %d\n", lsm->lsm_object_id, - loi->loi_id, loi->loi_ost_idx, err); - if (!rc) - rc = err; + 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); +} - RETURN(rc); +static 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; + ENTRY; + + if (rc) + lovset->set_completes = 0; + + err = lov_fini_statfs_set(lovset); + RETURN(rc ? rc : err); } -static int lov_statfs(struct lustre_handle *conn, struct obd_statfs *osfs) +static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo, + __u64 max_age, struct ptlrpc_request_set *rqset) { - struct obd_export *export = class_conn2export(conn); + struct lov_request_set *set; + struct lov_request *req; + struct list_head *pos; struct lov_obd *lov; - struct obd_statfs lov_sfs; - int set = 0; int rc = 0; - int i; ENTRY; - if (!export || !export->exp_obd) - RETURN(-ENODEV); + LASSERT(oinfo != NULL); + LASSERT(oinfo->oi_osfs != NULL); - lov = &export->exp_obd->u.lov; + lov = &obd->u.lov; + rc = lov_prep_statfs_set(obd, oinfo, &set); + if (rc) + RETURN(rc); - /* We only get block data from the OBD */ - for (i = 0; i < lov->desc.ld_tgt_count; i++) { + list_for_each (pos, &set->set_list) { + struct obd_device *osc_obd; + + req = 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); + if (rc) + break; + } + + if (rc || list_empty(&rqset->set_requests)) { int err; + if (rc) + set->set_completes = 0; + err = lov_fini_statfs_set(set); + RETURN(rc ? rc : err); + } + + LASSERT(rqset->set_interpret == NULL); + rqset->set_interpret = lov_statfs_interpret; + rqset->set_arg = (void *)set; + RETURN(0); +} + +static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs, + __u64 max_age) +{ + struct lov_obd *lov = &obd->u.lov; + struct obd_statfs lov_sfs; + int set = 0; + int rc = 0, err; + int i; + ENTRY; - if (!lov->tgts[i].active) + /* 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(&lov->tgts[i].conn, &lov_sfs); + err = obd_statfs(class_exp2obd(lov->lov_tgts[i]->ltd_exp), + &lov_sfs, max_age); if (err) { - CERROR("Error statfs OSC %s idx %d: err = %d\n", - lov->tgts[i].uuid, i, err); - if (!rc) + if (lov->lov_tgts[i]->ltd_active && !rc) rc = err; - continue; /* XXX or break? - probably OK to continue */ - } - if (!set) { - memcpy(osfs, &lov_sfs, sizeof(lov_sfs)); - set = 1; - } else { - osfs->os_bfree += lov_sfs.os_bfree; - osfs->os_bavail += lov_sfs.os_bavail; - osfs->os_blocks += lov_sfs.os_blocks; - /* XXX not sure about this one - depends on policy. - * - could be minimum if we always stripe on all OBDs - * (but that would be wrong for any other policy, - * if one of the OBDs has no more objects left) - * - could be sum if we stripe whole objects - * - could be average, just to give a nice number - * - we just pick first OST and hope it is enough - sfs->f_ffree += lov_sfs.f_ffree; - */ + continue; } + lov_update_statfs(class_exp2obd(lov->lov_tgts[i]->ltd_exp), + osfs, &lov_sfs, set); + set++; } - RETURN(rc); + + err = lov_fini_statfs(obd, osfs, set); + qos_update(lov); + + RETURN(rc ? rc : err); } -static int lov_iocontrol(unsigned int cmd, struct lustre_handle *conn, int len, +static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len, void *karg, void *uarg) { - struct obd_device *obddev = class_conn2obd(conn); + struct obd_device *obddev = class_exp2obd(exp); struct lov_obd *lov = &obddev->u.lov; - struct obd_ioctl_data *data = karg; - int i, count = lov->desc.ld_tgt_count; - int rc; - + int i, rc, count = lov->desc.ld_tgt_count; + struct obd_uuid *uuidp; ENTRY; switch (cmd) { - case IOC_LOV_SET_OSC_ACTIVE: { - rc = lov_set_osc_active(lov,data->ioc_inlbuf1,data->ioc_offset); - break; - } case OBD_IOC_LOV_GET_CONFIG: { - struct lov_tgt_desc *tgtdesc; + struct obd_ioctl_data *data; struct lov_desc *desc; - obd_uuid_t *uuidp; char *buf = NULL; + __u32 *genp; - buf = NULL; len = 0; if (obd_ioctl_getdata(&buf, &len, (void *)uarg)) RETURN(-EINVAL); @@ -1408,96 +2199,482 @@ static int lov_iocontrol(unsigned int cmd, struct lustre_handle *conn, int len, data = (struct obd_ioctl_data *)buf; if (sizeof(*desc) > data->ioc_inllen1) { - OBD_FREE(buf, len); + obd_ioctl_freedata(buf, len); + RETURN(-EINVAL); + } + + if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) { + obd_ioctl_freedata(buf, len); RETURN(-EINVAL); } - if (sizeof(*uuidp) * count > data->ioc_inllen2) { - OBD_FREE(buf, len); + if (sizeof(__u32) * count > data->ioc_inllen3) { + obd_ioctl_freedata(buf, len); RETURN(-EINVAL); } desc = (struct lov_desc *)data->ioc_inlbuf1; - uuidp = (obd_uuid_t *)data->ioc_inlbuf2; memcpy(desc, &(lov->desc), sizeof(*desc)); - tgtdesc = lov->tgts; - for (i = 0; i < count; i++, uuidp++, tgtdesc++) - memcpy(uuidp, tgtdesc->uuid, sizeof(*uuidp)); + uuidp = (struct obd_uuid *)data->ioc_inlbuf2; + genp = (__u32 *)data->ioc_inlbuf3; + /* the uuid will be empty for deleted OSTs */ + for (i = 0; i < count; i++, uuidp++, genp++) { + if (!lov->lov_tgts[i]) + continue; + *uuidp = lov->lov_tgts[i]->ltd_uuid; + *genp = lov->lov_tgts[i]->ltd_gen; + } rc = copy_to_user((void *)uarg, buf, len); if (rc) rc = -EFAULT; - OBD_FREE(buf, len); + obd_ioctl_freedata(buf, len); break; } - default: + case LL_IOC_LOV_SETSTRIPE: + rc = lov_setstripe(exp, karg, uarg); + break; + case LL_IOC_LOV_GETSTRIPE: + rc = lov_getstripe(exp, karg, uarg); + break; + case LL_IOC_LOV_SETEA: + rc = lov_setea(exp, karg, uarg); + break; + default: { + int set = 0; + if (count == 0) RETURN(-ENOTTY); + rc = 0; for (i = 0; i < count; i++) { - int err = obd_iocontrol(cmd, &lov->tgts[i].conn, - len, karg, uarg); - if (err && !rc) - rc = err; + int err; + + /* OST was disconnected */ + if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp) + continue; + + err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp, + len, karg, uarg); + if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) { + RETURN(err); + } else if (err) { + if (lov->lov_tgts[i]->ltd_active) { + CDEBUG(err == -ENOTTY ? + D_IOCTL : D_WARNING, + "iocontrol OSC %s on OST " + "idx %d cmd %x: err = %d\n", + lov_uuid2str(lov, i), + i, cmd, err); + if (!rc) + rc = err; + } + } else { + set = 1; + } + } + if (!set && !rc) + rc = -EIO; + } + } + + RETURN(rc); +} + +static int lov_get_info(struct obd_export *exp, __u32 keylen, + void *key, __u32 *vallen, void *val) +{ + struct obd_device *obddev = class_exp2obd(exp); + struct lov_obd *lov = &obddev->u.lov; + int i, rc; + ENTRY; + + if (!vallen || !val) + RETURN(-EFAULT); + + lov_getref(obddev); + + if (KEY_IS("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; + __u32 *stripe = val; + + if (*vallen < sizeof(*stripe)) + GOTO(out, rc = -EFAULT); + *vallen = sizeof(*stripe); + + /* XXX This is another one of those bits that will need to + * change if we ever actually support nested LOVs. It uses + * the lock's export to find out which stripe it is. */ + /* 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]; + 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[1]) { + *stripe = i; + GOTO(out, rc = 0); + } + } + LDLM_ERROR(data->lock, "lock on inode without such object"); + dump_lsm(D_ERROR, data->lsm); + GOTO(out, rc = -ENXIO); + } else if (KEY_IS("last_id")) { + obd_id *ids = val; + int size = sizeof(obd_id); + for (i = 0; i < lov->desc.ld_tgt_count; i++) { + if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) + continue; + rc = obd_get_info(lov->lov_tgts[i]->ltd_exp, + keylen, key, &size, ids + i); + if (rc != 0) + GOTO(out, rc); + } + GOTO(out, rc = 0); + } else if (KEY_IS(KEY_LOVDESC)) { + struct lov_desc *desc_ret = val; + *desc_ret = lov->desc; + + GOTO(out, rc = 0); + } + + rc = -EINVAL; +out: + lov_putref(obddev); + RETURN(rc); +} + +static int lov_set_info_async(struct obd_export *exp, obd_count keylen, + void *key, obd_count vallen, void *val, + struct ptlrpc_request_set *set) +{ + struct obd_device *obddev = class_exp2obd(exp); + struct lov_obd *lov = &obddev->u.lov; + int i, rc = 0, err, incr = 0, check_uuid = 0, do_inactive = 0; + int no_set = !set; + ENTRY; + + if (no_set) { + set = ptlrpc_prep_set(); + if (!set) + RETURN(-ENOMEM); + } + + if (KEY_IS(KEY_NEXT_ID)) { + if (vallen != lov->desc.ld_tgt_count * sizeof(obd_id)) + RETURN(-EINVAL); + vallen = sizeof(obd_id); + incr = sizeof(obd_id); + do_inactive = 1; + } else if (KEY_IS("checksum")) { + do_inactive = 1; + } else if (KEY_IS("mds_conn") || KEY_IS("unlinked")) { + check_uuid = val ? 1 : 0; + } else if (KEY_IS("evict_by_nid")) { + /* use defaults: + do_inactive = incr = 0; + */ + } + + lov_getref(obddev); + + for (i = 0; i < lov->desc.ld_tgt_count; i++, val = (char *)val + incr) { + /* OST was disconnected */ + if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp) + continue; + + /* OST is inactive and we don't want inactive OSCs */ + if (!lov->lov_tgts[i]->ltd_active && !do_inactive) + continue; + + /* Only want a specific OSC */ + if (check_uuid && + !obd_uuid_equals(val, &lov->lov_tgts[i]->ltd_uuid)) + continue; + + err = obd_set_info_async(lov->lov_tgts[i]->ltd_exp, + keylen, key, vallen, val, set); + if (!rc) + rc = err; + } + lov_putref(obddev); + if (no_set) { + err = ptlrpc_set_wait(set); + if (!rc) + rc = err; + ptlrpc_set_destroy(set); + } + 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) +{ + struct lov_oinfo *loi; + int i, rc = 0; + ENTRY; + + for (i = 0; i < lsm->lsm_stripe_count; i++) { + loi = lsm->lsm_oinfo[i]; + if (loi->loi_ar.ar_rc && !rc) + rc = loi->loi_ar.ar_rc; + loi->loi_ar.ar_rc = 0; + } + RETURN(rc); +} +EXPORT_SYMBOL(lov_test_and_clear_async_rc); + + +static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm, + int cmd, __u64 *offset) +{ + __u64 start; + __u32 ssize = lsm->lsm_stripe_size; + + start = *offset; + do_div(start, ssize); + start = start * ssize; + + CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64 + ", end "LPU64"\n", *offset, ssize, start, start + ssize - 1); + if (cmd == OBD_CALC_STRIPE_END) { + *offset = start + ssize - 1; + } else if (cmd == OBD_CALC_STRIPE_START) { + *offset = start; + } else { + LBUG(); + } + + 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(); +} +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); +} +EXPORT_SYMBOL(lov_stripe_unlock); struct obd_ops lov_obd_ops = { - o_attach: lov_attach, - o_detach: lov_detach, - o_setup: lov_setup, - o_connect: lov_connect, - o_disconnect: lov_disconnect, - o_statfs: lov_statfs, - o_packmd: lov_packmd, - o_unpackmd: lov_unpackmd, - o_create: lov_create, - o_destroy: lov_destroy, - o_getattr: lov_getattr, - o_setattr: lov_setattr, - o_open: lov_open, - o_close: lov_close, - o_brw: lov_brw, - o_punch: lov_punch, - o_enqueue: lov_enqueue, - o_cancel: lov_cancel, - o_cancel_unused: lov_cancel_unused, - o_iocontrol: lov_iocontrol + .o_owner = THIS_MODULE, + .o_setup = lov_setup, + .o_precleanup = lov_precleanup, + .o_cleanup = lov_cleanup, + .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, + .o_getattr_async = lov_getattr_async, + .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, + .o_extent_calc = lov_extent_calc, + .o_llog_init = lov_llog_init, + .o_llog_finish = lov_llog_finish, + .o_notify = lov_notify, }; +static quota_interface_t *quota_interface; +extern quota_interface_t lov_quota_interface; -#define LOV_VERSION "v0.1" +cfs_mem_cache_t *lov_oinfo_slab; -static int __init lov_init(void) +int __init lov_init(void) { - int rc; - printk(KERN_INFO "Lustre Logical Object Volume driver " LOV_VERSION - ", info@clusterfs.com\n"); - lov_file_cache = kmem_cache_create("ll_lov_file_data", - sizeof(struct lov_file_handles), - 0, 0, NULL, NULL); - if (!lov_file_cache) - RETURN(-ENOMEM); + struct lprocfs_static_vars lvars; + int rc, rc2; + ENTRY; + + lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo", + sizeof(struct lov_oinfo), + 0, SLAB_HWCACHE_ALIGN); + if (lov_oinfo_slab == NULL) + return -ENOMEM; + lprocfs_init_vars(lov, &lvars); + + 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, lvars.module_vars, + LUSTRE_LOV_NAME); + if (rc) { + if (quota_interface) + PORTAL_SYMBOL_PUT(lov_quota_interface); + rc2 = cfs_mem_cache_destroy(lov_oinfo_slab); + LASSERT(rc2 == 0); + } - rc = class_register_type(&lov_obd_ops, status_class_var, - OBD_LOV_DEVICENAME); RETURN(rc); } -static void __exit lov_exit(void) +#ifdef __KERNEL__ +static void /*__exit*/ lov_exit(void) { - if (kmem_cache_destroy(lov_file_cache)) - CERROR("couldn't free LOV open cache\n"); - class_unregister_type(OBD_LOV_DEVICENAME); + int rc; + + if (quota_interface) + PORTAL_SYMBOL_PUT(lov_quota_interface); + + class_unregister_type(LUSTRE_LOV_NAME); + rc = cfs_mem_cache_destroy(lov_oinfo_slab); + LASSERT(rc == 0); } MODULE_AUTHOR("Cluster File Systems, Inc. "); -MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver " LOV_VERSION); +MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver"); MODULE_LICENSE("GPL"); -module_init(lov_init); -module_exit(lov_exit); +cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit); +#endif