X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Flod%2Flod_lov.c;h=47eefff982fd18c1a8d43d798b40a7d3265f82cf;hb=0030ed640284e4ed5afa9187c5a907141132b083;hp=a79365966bae673666966dd4bc17c783ea702232;hpb=976cbeb5ff1d32e7cac54478e6ff249cffe45116;p=fs%2Flustre-release.git diff --git a/lustre/lod/lod_lov.c b/lustre/lod/lod_lov.c index a793659..47eefff 100644 --- a/lustre/lod/lod_lov.c +++ b/lustre/lod/lod_lov.c @@ -6,116 +6,146 @@ * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 only, * as published by the Free Software Foundation. - + * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License version 2 for more details. A copy is * included in the COPYING file that accompanied this code. - + * * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * version 2 along with this program; If not, see + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ /* * Copyright 2009 Sun Microsystems, Inc. All rights reserved * Use is subject to license terms. + * + * Copyright (c) 2012, 2014, Intel Corporation. */ /* * lustre/lod/lod_lov.c * - * Author: Alex Zhuravlev + * A set of helpers to maintain Logical Object Volume (LOV) + * Extended Attribute (EA) and known OST targets + * + * Author: Alex Zhuravlev */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_MDS #include -#include +#include +#include #include "lod_internal.h" -/* - * Keep a refcount of lod->lod_osts usage to prevent racing with - * addition/deletion. Any function that expects lov_tgts to remain stationary - * must take a ref. +/** + * Increase reference count on the target table. * - * \param lod - is the lod device from which we want to grab a reference + * Increase reference count on the target table usage to prevent racing with + * addition/deletion. Any function that expects the table to remain + * stationary must take a ref. + * + * \param[in] ltd target table (lod_ost_descs or lod_mdt_descs) */ -void lod_getref(struct lod_device *lod) +void lod_getref(struct lod_tgt_descs *ltd) { - cfs_down_read(&lod->lod_rw_sem); - cfs_mutex_lock(&lod->lod_mutex); - lod->lod_refcount++; - cfs_mutex_unlock(&lod->lod_mutex); + down_read(<d->ltd_rw_sem); + mutex_lock(<d->ltd_mutex); + ltd->ltd_refcount++; + mutex_unlock(<d->ltd_mutex); } -/* - * Companion of lod_getref() to release a reference on the lod table. - * If this is the last reference and the ost entry was scheduled for deletion, - * the descriptor is removed from the array. +/** + * Decrease reference count on the target table. * - * \param lod - is the lod device from which we release a reference + * Companion of lod_getref() to release a reference on the target table. + * If this is the last reference and the OST entry was scheduled for deletion, + * the descriptor is removed from the table. + * + * \param[in] lod LOD device from which we release a reference + * \param[in] ltd target table (lod_ost_descs or lod_mdt_descs) */ -void lod_putref(struct lod_device *lod) +void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd) { - cfs_mutex_lock(&lod->lod_mutex); - lod->lod_refcount--; - if (lod->lod_refcount == 0 && lod->lod_death_row) { - struct lod_ost_desc *ost_desc, *tmp; - int idx; - CFS_LIST_HEAD(kill); + mutex_lock(<d->ltd_mutex); + ltd->ltd_refcount--; + if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) { + struct lod_tgt_desc *tgt_desc, *tmp; + struct list_head kill; + unsigned int idx; - CDEBUG(D_CONFIG, "destroying %d lod desc\n", - lod->lod_death_row); + CDEBUG(D_CONFIG, "destroying %d ltd desc\n", + ltd->ltd_death_row); - cfs_foreach_bit(lod->lod_ost_bitmap, idx) { - ost_desc = OST_TGT(lod, idx); - LASSERT(ost_desc); + INIT_LIST_HEAD(&kill); - if (!ost_desc->ltd_reap) - continue; + cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) { + tgt_desc = LTD_TGT(ltd, idx); + LASSERT(tgt_desc); - cfs_list_add(&ost_desc->ltd_kill, &kill); + if (!tgt_desc->ltd_reap) + continue; - lod_ost_pool_remove(&lod->lod_pool_info, idx); - OST_TGT(lod, idx) = NULL; - lod->lod_ostnr--; - cfs_bitmap_clear(lod->lod_ost_bitmap, idx); - if (ost_desc->ltd_active) - lod->lod_desc.ld_active_tgt_count--; - lod->lod_death_row--; + list_add(&tgt_desc->ltd_kill, &kill); + LTD_TGT(ltd, idx) = NULL; + /*FIXME: only support ost pool for now */ + if (ltd == &lod->lod_ost_descs) { + lod_ost_pool_remove(&lod->lod_pool_info, idx); + if (tgt_desc->ltd_active) + lod->lod_desc.ld_active_tgt_count--; + } + ltd->ltd_tgtnr--; + cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx); + ltd->ltd_death_row--; } - cfs_mutex_unlock(&lod->lod_mutex); - cfs_up_read(&lod->lod_rw_sem); + mutex_unlock(<d->ltd_mutex); + up_read(<d->ltd_rw_sem); - cfs_list_for_each_entry_safe(ost_desc, tmp, &kill, ltd_kill) { + list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) { int rc; - cfs_list_del(&ost_desc->ltd_kill); - /* remove from QoS structures */ - rc = qos_del_tgt(lod, ost_desc); - if (rc) - CERROR("%s: qos_del_tgt(%s) failed: rc = %d\n", - lod2obd(lod)->obd_name, - obd_uuid2str(&ost_desc->ltd_uuid), rc); - rc = obd_disconnect(ost_desc->ltd_exp); + list_del(&tgt_desc->ltd_kill); + if (ltd == &lod->lod_ost_descs) { + /* remove from QoS structures */ + rc = qos_del_tgt(lod, tgt_desc); + if (rc) + CERROR("%s: qos_del_tgt(%s) failed:" + "rc = %d\n", + lod2obd(lod)->obd_name, + obd_uuid2str(&tgt_desc->ltd_uuid), + rc); + } + rc = obd_disconnect(tgt_desc->ltd_exp); if (rc) CERROR("%s: failed to disconnect %s: rc = %d\n", lod2obd(lod)->obd_name, - obd_uuid2str(&ost_desc->ltd_uuid), rc); - OBD_FREE_PTR(ost_desc); + obd_uuid2str(&tgt_desc->ltd_uuid), rc); + OBD_FREE_PTR(tgt_desc); } } else { - cfs_mutex_unlock(&lod->lod_mutex); - cfs_up_read(&lod->lod_rw_sem); + mutex_unlock(<d->ltd_mutex); + up_read(<d->ltd_rw_sem); } } -static int lod_bitmap_resize(struct lod_device *lod, __u32 newsize) +/** + * Expand size of target table. + * + * When the target table is full, we have to extend the table. To do so, + * we allocate new memory with some reserve, move data from the old table + * to the new one and release memory consumed by the old table. + * Notice we take ltd_rw_sem exclusively to ensure atomic switch. + * + * \param[in] ltd target table + * \param[in] newsize new size of the table + * + * \retval 0 on success + * \retval -ENOMEM if reallocation failed + */ +static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize) { cfs_bitmap_t *new_bitmap, *old_bitmap = NULL; int rc = 0; @@ -123,9 +153,9 @@ static int lod_bitmap_resize(struct lod_device *lod, __u32 newsize) /* grab write reference on the lod. Relocating the array requires * exclusive access */ - cfs_down_write(&lod->lod_rw_sem); - if (newsize <= lod->lod_osts_size) + down_write(<d->ltd_rw_sem); + if (newsize <= ltd->ltd_tgts_size) /* someone else has already resize the array */ GOTO(out, rc = 0); @@ -134,38 +164,48 @@ static int lod_bitmap_resize(struct lod_device *lod, __u32 newsize) if (!new_bitmap) GOTO(out, rc = -ENOMEM); - if (lod->lod_osts_size > 0) { + if (ltd->ltd_tgts_size > 0) { /* the bitmap already exists, we need * to copy data from old one */ - cfs_bitmap_copy(new_bitmap, lod->lod_ost_bitmap); - old_bitmap = lod->lod_ost_bitmap; + cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap); + old_bitmap = ltd->ltd_tgt_bitmap; } - lod->lod_osts_size = newsize; - lod->lod_ost_bitmap = new_bitmap; + ltd->ltd_tgts_size = newsize; + ltd->ltd_tgt_bitmap = new_bitmap; if (old_bitmap) CFS_FREE_BITMAP(old_bitmap); - CDEBUG(D_CONFIG, "ost size: %d\n", lod->lod_osts_size); + CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size); EXIT; out: - cfs_up_write(&lod->lod_rw_sem); + up_write(<d->ltd_rw_sem); return rc; } -/* - * Connect LOD to a new OSP and add it to the device table. +/** + * Connect LOD to a new OSP and add it to the target table. + * + * Connect to the OSP device passed, initialize all the internal + * structures related to the device and add it to the target table. * - * \param env - is the environment passed by the caller - * \param lod - is the LOD device to be connected to the new OSP - * \param osp - is the name of OSP device name about to be added - * \param index - is the OSP index - * \param gen - is the generation number + * \param[in] env execution environment for this thread + * \param[in] lod LOD device to be connected to the new OSP + * \param[in] osp name of OSP device name to be added + * \param[in] index index of the new target + * \param[in] gen target's generation number + * \param[in] tgt_index OSP's group + * \param[in] type type of device (mdc or osc) + * \param[in] active state of OSP: 0 - inactive, 1 - active + * + * \retval 0 if added successfully + * \retval negative error number on failure */ int lod_add_device(const struct lu_env *env, struct lod_device *lod, - char *osp, unsigned index, unsigned gen, int active) + char *osp, unsigned index, unsigned gen, int tgt_index, + char *type, int active) { struct obd_connect_data *data = NULL; struct obd_export *exp = NULL; @@ -173,9 +213,10 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod, struct lu_device *ldev; struct dt_device *d; int rc; - struct lod_ost_desc *ost_desc; - struct obd_uuid obd_uuid; - + struct lod_tgt_desc *tgt_desc; + struct lod_tgt_descs *ltd; + struct obd_uuid obd_uuid; + bool for_ost; ENTRY; CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen); @@ -203,6 +244,49 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod, data->ocd_version = LUSTRE_VERSION_CODE; data->ocd_index = index; + if (strcmp(LUSTRE_OSC_NAME, type) == 0) { + for_ost = true; + data->ocd_connect_flags |= OBD_CONNECT_AT | + OBD_CONNECT_FULL20 | + OBD_CONNECT_INDEX | +#ifdef HAVE_LRU_RESIZE_SUPPORT + OBD_CONNECT_LRU_RESIZE | +#endif + OBD_CONNECT_MDS | + OBD_CONNECT_OSS_CAPA | + OBD_CONNECT_REQPORTAL | + OBD_CONNECT_SKIP_ORPHAN | + OBD_CONNECT_FID | + OBD_CONNECT_LVB_TYPE | + OBD_CONNECT_VERSION | + OBD_CONNECT_PINGLESS | + OBD_CONNECT_LFSCK; + + data->ocd_group = tgt_index; + ltd = &lod->lod_ost_descs; + } else { + struct obd_import *imp = obd->u.cli.cl_import; + + for_ost = false; + data->ocd_ibits_known = MDS_INODELOCK_UPDATE; + data->ocd_connect_flags |= OBD_CONNECT_ACL | + OBD_CONNECT_MDS_CAPA | + OBD_CONNECT_OSS_CAPA | + OBD_CONNECT_IBITS | + OBD_CONNECT_MDS_MDS | + OBD_CONNECT_FID | + OBD_CONNECT_AT | + OBD_CONNECT_FULL20 | + OBD_CONNECT_LFSCK; + spin_lock(&imp->imp_lock); + imp->imp_server_timeout = 1; + spin_unlock(&imp->imp_lock); + imp->imp_client->cli_request_portal = OUT_PORTAL; + CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n", + obd->obd_name); + ltd = &lod->lod_mdt_descs; + } + rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL); OBD_FREE_PTR(data); if (rc) { @@ -212,129 +296,193 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod, } LASSERT(obd->obd_lu_dev); - LASSERT(obd->obd_lu_dev->ld_site = lod->lod_dt_dev.dd_lu_dev.ld_site); + LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site); ldev = obd->obd_lu_dev; d = lu2dt_dev(ldev); /* Allocate ost descriptor and fill it */ - OBD_ALLOC_PTR(ost_desc); - if (!ost_desc) + OBD_ALLOC_PTR(tgt_desc); + if (!tgt_desc) GOTO(out_conn, rc = -ENOMEM); - ost_desc->ltd_ost = d; - ost_desc->ltd_exp = exp; - ost_desc->ltd_uuid = obd->u.cli.cl_target_uuid; - ost_desc->ltd_gen = gen; - ost_desc->ltd_index = index; - ost_desc->ltd_active = active; + tgt_desc->ltd_tgt = d; + tgt_desc->ltd_exp = exp; + tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid; + tgt_desc->ltd_gen = gen; + tgt_desc->ltd_index = index; + tgt_desc->ltd_active = active; - lod_getref(lod); - if (index >= lod->lod_osts_size) { + lod_getref(ltd); + if (index >= ltd->ltd_tgts_size) { /* we have to increase the size of the lod_osts array */ __u32 newsize; - newsize = max(lod->lod_osts_size, (__u32)2); + newsize = max(ltd->ltd_tgts_size, (__u32)2); while (newsize < index + 1) newsize = newsize << 1; /* lod_bitmap_resize() needs lod_rw_sem * which we hold with th reference */ - lod_putref(lod); + lod_putref(lod, ltd); - rc = lod_bitmap_resize(lod, newsize); + rc = ltd_bitmap_resize(ltd, newsize); if (rc) GOTO(out_desc, rc); - lod_getref(lod); + lod_getref(ltd); } - cfs_mutex_lock(&lod->lod_mutex); - if (cfs_bitmap_check(lod->lod_ost_bitmap, index)) { + mutex_lock(<d->ltd_mutex); + if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) { CERROR("%s: device %d is registered already\n", obd->obd_name, index); GOTO(out_mutex, rc = -EEXIST); } - if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) { - OBD_ALLOC_PTR(lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK]); - if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) { + if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) { + OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]); + if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) { CERROR("can't allocate index to add %s\n", obd->obd_name); GOTO(out_mutex, rc = -ENOMEM); } } - rc = lod_ost_pool_add(&lod->lod_pool_info, index, lod->lod_osts_size); - if (rc) { - CERROR("%s: can't set up pool, failed with %d\n", - obd->obd_name, rc); - GOTO(out_mutex, rc); - } + if (!strcmp(LUSTRE_OSC_NAME, type)) { + /* pool and qos are not supported for MDS stack yet */ + rc = lod_ost_pool_add(&lod->lod_pool_info, index, + lod->lod_osts_size); + if (rc) { + CERROR("%s: can't set up pool, failed with %d\n", + obd->obd_name, rc); + GOTO(out_mutex, rc); + } - rc = qos_add_tgt(lod, ost_desc); - if (rc) { - CERROR("%s: qos_add_tgt(%s) failed: rc = %d\n", obd->obd_name, - obd_uuid2str(&ost_desc->ltd_uuid), rc); - GOTO(out_pool, rc); - } - - /* The new OST is now a full citizen */ - if (index >= lod->lod_desc.ld_tgt_count) - lod->lod_desc.ld_tgt_count = index + 1; - if (active) - lod->lod_desc.ld_active_tgt_count++; - OST_TGT(lod, index) = ost_desc; - cfs_bitmap_set(lod->lod_ost_bitmap, index); - lod->lod_ostnr++; - cfs_mutex_unlock(&lod->lod_mutex); - lod_putref(lod); + rc = qos_add_tgt(lod, tgt_desc); + if (rc) { + CERROR("%s: qos_add_tgt failed with %d\n", + obd->obd_name, rc); + GOTO(out_pool, rc); + } + + /* The new OST is now a full citizen */ + if (index >= lod->lod_desc.ld_tgt_count) + lod->lod_desc.ld_tgt_count = index + 1; + if (active) + lod->lod_desc.ld_active_tgt_count++; + } + LTD_TGT(ltd, index) = tgt_desc; + cfs_bitmap_set(ltd->ltd_tgt_bitmap, index); + ltd->ltd_tgtnr++; + mutex_unlock(<d->ltd_mutex); + lod_putref(lod, ltd); if (lod->lod_recovery_completed) ldev->ld_ops->ldo_recovery_complete(env, ldev); - RETURN(0); + rc = lfsck_add_target(env, lod->lod_child, d, exp, index, for_ost); + if (rc != 0) + CERROR("Fail to add LFSCK target: name = %s, type = %s, " + "index = %u, rc = %d\n", osp, type, index, rc); + + RETURN(rc); out_pool: lod_ost_pool_remove(&lod->lod_pool_info, index); out_mutex: - cfs_mutex_unlock(&lod->lod_mutex); - lod_putref(lod); + mutex_unlock(<d->ltd_mutex); + lod_putref(lod, ltd); out_desc: - OBD_FREE_PTR(ost_desc); + OBD_FREE_PTR(tgt_desc); out_conn: obd_disconnect(exp); out_free: return rc; } -/* - * helper function to schedule OST removal from the device table +/** + * Schedule target removal from the target table. + * + * Mark the device as dead. The device is not removed here because it may + * still be in use. The device will be removed in lod_putref() when the + * last reference is released. + * + * \param[in] env execution environment for this thread + * \param[in] lod LOD device the target table belongs to + * \param[in] ltd target table + * \param[in] idx index of the target + * \param[in] for_ost type of the target: 0 - MDT, 1 - OST */ -static void __lod_del_device(struct lod_device *lod, unsigned idx) +static void __lod_del_device(const struct lu_env *env, struct lod_device *lod, + struct lod_tgt_descs *ltd, unsigned idx, + bool for_ost) { - LASSERT(OST_TGT(lod,idx)); - if (OST_TGT(lod,idx)->ltd_reap == 0) { - OST_TGT(lod,idx)->ltd_reap = 1; - lod->lod_death_row++; + LASSERT(LTD_TGT(ltd, idx)); + + lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt, + idx, for_ost); + + if (LTD_TGT(ltd, idx)->ltd_reap == 0) { + LTD_TGT(ltd, idx)->ltd_reap = 1; + ltd->ltd_death_row++; } } -/* - * Add support for administratively disabled OST (through the MGS). - * Schedule a target for deletion. Disconnection and real removal from the - * table takes place in lod_putref() once the last table user release its - * reference. - * - * \param env - is the environment passed by the caller - * \param lod - is the lod device currently connected to the OSP about to be - * removed - * \param osp - is the name of OSP device about to be removed - * \param idx - is the OSP index - * \param gen - is the generation number, not used currently +/** + * Schedule removal of all the targets from the given target table. + * + * See more details in the description for __lod_del_device() + * + * \param[in] env execution environment for this thread + * \param[in] lod LOD device the target table belongs to + * \param[in] ltd target table + * \param[in] for_ost type of the target: MDT or OST + * + * \retval 0 always + */ +int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod, + struct lod_tgt_descs *ltd, bool for_ost) +{ + unsigned int idx; + + if (ltd->ltd_tgts_size <= 0) + return 0; + lod_getref(ltd); + mutex_lock(<d->ltd_mutex); + cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) + __lod_del_device(env, lod, ltd, idx, for_ost); + mutex_unlock(<d->ltd_mutex); + lod_putref(lod, ltd); + CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap); + for (idx = 0; idx < TGT_PTRS; idx++) { + if (ltd->ltd_tgt_idx[idx]) + OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]); + } + ltd->ltd_tgts_size = 0; + return 0; +} + +/** + * Remove device by name. + * + * Remove a device identified by \a osp from the target table. Given + * the device can be in use, the real deletion happens in lod_putref(). + * + * \param[in] env execution environment for this thread + * \param[in] lod LOD device to be connected to the new OSP + * \param[in] ltd target table + * \param[in] osp name of OSP device to be removed + * \param[in] idx index of the target + * \param[in] gen generation number, not used currently + * \param[in] for_ost type of the target: 0 - MDT, 1 - OST + * + * \retval 0 if the device was scheduled for removal + * \retval -EINVAL if no device was found */ int lod_del_device(const struct lu_env *env, struct lod_device *lod, - char *osp, unsigned idx, unsigned gen) + struct lod_tgt_descs *ltd, char *osp, unsigned idx, + unsigned gen, bool for_ost) { struct obd_device *obd; int rc = 0; @@ -343,7 +491,10 @@ int lod_del_device(const struct lu_env *env, struct lod_device *lod, CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen); - obd = class_name2obd(osp); + obd_str2uuid(&uuid, osp); + + obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME, + &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid); if (obd == NULL) { CERROR("can't find %s device\n", osp); RETURN(-EINVAL); @@ -357,35 +508,51 @@ int lod_del_device(const struct lu_env *env, struct lod_device *lod, obd_str2uuid(&uuid, osp); - lod_getref(lod); - cfs_mutex_lock(&lod->lod_mutex); + lod_getref(ltd); + mutex_lock(<d->ltd_mutex); /* check that the index is allocated in the bitmap */ - if (!cfs_bitmap_check(lod->lod_ost_bitmap, idx) || !OST_TGT(lod,idx)) { + if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) || + !LTD_TGT(ltd, idx)) { CERROR("%s: device %d is not set up\n", obd->obd_name, idx); GOTO(out, rc = -EINVAL); } /* check that the UUID matches */ - if (!obd_uuid_equals(&uuid, &OST_TGT(lod,idx)->ltd_uuid)) { + if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) { CERROR("%s: LOD target UUID %s at index %d does not match %s\n", - obd->obd_name, obd_uuid2str(&OST_TGT(lod,idx)->ltd_uuid), + obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid), idx, osp); GOTO(out, rc = -EINVAL); } - __lod_del_device(lod, idx); + __lod_del_device(env, lod, ltd, idx, for_ost); EXIT; out: - cfs_mutex_unlock(&lod->lod_mutex); - lod_putref(lod); + mutex_unlock(<d->ltd_mutex); + lod_putref(lod, ltd); return(rc); } -int lod_ea_store_resize(struct lod_thread_info *info, int size) +/** + * Resize per-thread storage to hold specified size. + * + * A helper function to resize per-thread temporary storage. This storage + * is used to process LOV/LVM EAs and may be quite large. We do not want to + * allocate/release it every time, so instead we put it into the env and + * reallocate on demand. The memory is released when the correspondent thread + * is finished. + * + * \param[in] info LOD-specific storage in the environment + * \param[in] size new size to grow the buffer to + + * \retval 0 on success, -ENOMEM if reallocation failed + */ +int lod_ea_store_resize(struct lod_thread_info *info, size_t size) { - int round = size_roundup_power2(size); + __u32 round = size_roundup_power2(size); - LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3)); + LASSERT(round <= + lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3)); if (info->lti_ea_store) { LASSERT(info->lti_ea_store_size); LASSERT(info->lti_ea_store_size < round); @@ -403,34 +570,157 @@ int lod_ea_store_resize(struct lod_thread_info *info, int size) RETURN(0); } -int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo) +/** + * Make LOV EA for striped object. + * + * Generate striping information and store it in the LOV EA of the given + * object. The caller must ensure nobody else is calling the function + * against the object concurrently. The transaction must be started. + * FLDB service must be running as well; it's used to map FID to the target, + * which is stored in LOV EA. + * + * \param[in] env execution environment for this thread + * \param[in] lo LOD object + * \param[in] th transaction handle + * + * \retval 0 if LOV EA is stored successfully + * \retval negative error number on failure + */ +int lod_generate_and_set_lovea(const struct lu_env *env, + struct lod_object *lo, struct thandle *th) { - struct lod_thread_info *info = lod_env_info(env); - struct dt_object *next = dt_object_child(&lo->ldo_obj); + struct lod_thread_info *info = lod_env_info(env); + struct dt_object *next = dt_object_child(&lo->ldo_obj); + const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu); + struct lov_mds_md_v1 *lmm; + struct lov_ost_data_v1 *objs; + __u32 magic; + int i, rc; + size_t lmm_size; + ENTRY; + + LASSERT(lo); + + magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1; + lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic); + if (info->lti_ea_store_size < lmm_size) { + rc = lod_ea_store_resize(info, lmm_size); + if (rc) + RETURN(rc); + } + + if (lo->ldo_pattern == 0) /* default striping */ + lo->ldo_pattern = LOV_PATTERN_RAID0; + + lmm = info->lti_ea_store; + + lmm->lmm_magic = cpu_to_le32(magic); + lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern); + fid_to_lmm_oi(fid, &lmm->lmm_oi); + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI)) + lmm->lmm_oi.oi.oi_id++; + lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi); + lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size); + lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr); + if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED) + lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr); + lmm->lmm_layout_gen = 0; + if (magic == LOV_MAGIC_V1) { + objs = &lmm->lmm_objects[0]; + } else { + struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm; + size_t cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool, + sizeof(v3->lmm_pool_name)); + if (cplen >= sizeof(v3->lmm_pool_name)) + RETURN(-E2BIG); + objs = &v3->lmm_objects[0]; + } + + for (i = 0; i < lo->ldo_stripenr; i++) { + struct lu_fid *fid = &info->lti_fid; + struct lod_device *lod; + __u32 index; + int type = LU_SEQ_RANGE_OST; + + lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev); + LASSERT(lo->ldo_stripe[i]); + + *fid = *lu_object_fid(&lo->ldo_stripe[i]->do_lu); + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) { + if (cfs_fail_val == 0) + cfs_fail_val = fid->f_oid; + else + fid->f_oid = cfs_fail_val; + } + + rc = fid_to_ostid(fid, &info->lti_ostid); + LASSERT(rc == 0); + + ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi); + objs[i].l_ost_gen = cpu_to_le32(0); + rc = lod_fld_lookup(env, lod, fid, &index, &type); + if (rc < 0) { + CERROR("%s: Can not locate "DFID": rc = %d\n", + lod2obd(lod)->obd_name, PFID(fid), rc); + lod_object_free_striping(env, lo); + RETURN(rc); + } + objs[i].l_ost_idx = cpu_to_le32(index); + } + + info->lti_buf.lb_buf = lmm; + info->lti_buf.lb_len = lmm_size; + rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, + th, BYPASS_CAPA); + if (rc < 0) + lod_object_free_striping(env, lo); + + RETURN(rc); +} + +/** + * Get LOV EA. + * + * Fill lti_ea_store buffer in the environment with a value for the given + * EA. The buffer is reallocated if the value doesn't fit. + * + * \param[in,out] env execution environment for this thread + * .lti_ea_store buffer is filled with EA's value + * \param[in] lo LOD object + * \param[in] name name of the EA + * + * \retval 0 if EA is fetched successfully + * \retval negative error number on failure + */ +int lod_get_ea(const struct lu_env *env, struct lod_object *lo, + const char *name) +{ + struct lod_thread_info *info = lod_env_info(env); + struct dt_object *next = dt_object_child(&lo->ldo_obj); int rc; ENTRY; LASSERT(info); - if (unlikely(info->lti_ea_store_size == 0)) { + if (unlikely(info->lti_ea_store == NULL)) { /* just to enter in allocation block below */ rc = -ERANGE; } else { repeat: info->lti_buf.lb_buf = info->lti_ea_store; info->lti_buf.lb_len = info->lti_ea_store_size; - rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV, - BYPASS_CAPA); + rc = dt_xattr_get(env, next, &info->lti_buf, name, BYPASS_CAPA); } + /* if object is not striped or inaccessible */ - if (rc == -ENODATA) + if (rc == -ENODATA || rc == -ENOENT) RETURN(0); if (rc == -ERANGE) { /* EA doesn't fit, reallocate new buffer */ - rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV, + rc = dt_xattr_get(env, next, &LU_BUF_NULL, name, BYPASS_CAPA); - if (rc == -ENODATA) + if (rc == -ENODATA || rc == -ENOENT) RETURN(0); else if (rc < 0) RETURN(rc); @@ -445,52 +735,104 @@ repeat: RETURN(rc); } -/* - * allocate array of objects pointers, find/create objects - * stripenr and other fields should be initialized by this moment +/** + * Verify the target index is present in the current configuration. + * + * \param[in] md LOD device where the target table is stored + * \param[in] idx target's index + * + * \retval 0 if the index is present + * \retval -EINVAL if not + */ +static int validate_lod_and_idx(struct lod_device *md, __u32 idx) +{ + if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size || + !cfs_bitmap_check(md->lod_ost_bitmap, idx))) { + CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx, + md->lod_ost_descs.ltd_tgts_size); + return -EINVAL; + } + + if (unlikely(OST_TGT(md, idx) == NULL)) { + CERROR("%s: bad lod_tgt_desc for idx: %d\n", + lod2obd(md)->obd_name, idx); + return -EINVAL; + } + + if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) { + CERROR("%s: invalid lod device, for idx: %d\n", + lod2obd(md)->obd_name , idx); + return -EINVAL; + } + + return 0; +} + +/** + * Instantiate objects for stripes. + * + * Allocate and initialize LU-objects representing the stripes. The number + * of the stripes (ldo_stripenr) must be initialized already. The caller + * must ensure nobody else is calling the function on the object at the same + * time. FLDB service must be running to be able to map a FID to the targets + * and find appropriate device representing that target. + * + * \param[in] env execution environment for this thread + * \param[in,out] lo LOD object + * \param[in] objs an array of IDs to creates the objects from + * + * \retval 0 if the objects are instantiated successfully + * \retval negative error number on failure */ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo, struct lov_ost_data_v1 *objs) { struct lod_thread_info *info = lod_env_info(env); - struct lod_device *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev); + struct lod_device *md; struct lu_object *o, *n; struct lu_device *nd; - int i, idx, rc = 0; + struct dt_object **stripe; + int stripe_len; + int i, rc = 0; + __u32 idx; ENTRY; - LASSERT(lo); + LASSERT(lo != NULL); + md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev); LASSERT(lo->ldo_stripe == NULL); LASSERT(lo->ldo_stripenr > 0); LASSERT(lo->ldo_stripe_size > 0); - i = sizeof(struct dt_object *) * lo->ldo_stripenr; - OBD_ALLOC(lo->ldo_stripe, i); - if (lo->ldo_stripe == NULL) - GOTO(out, rc = -ENOMEM); - lo->ldo_stripes_allocated = lo->ldo_stripenr; + stripe_len = lo->ldo_stripenr; + OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len); + if (stripe == NULL) + RETURN(-ENOMEM); for (i = 0; i < lo->ldo_stripenr; i++) { + if (unlikely(lovea_slot_is_dummy(&objs[i]))) + continue; + + ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid); + idx = le32_to_cpu(objs[i].l_ost_idx); + rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx); + if (rc != 0) + GOTO(out, rc); + LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n", + PFID(&info->lti_fid)); + lod_getref(&md->lod_ost_descs); + + rc = validate_lod_and_idx(md, idx); + if (unlikely(rc != 0)) { + lod_putref(md, &md->lod_ost_descs); + GOTO(out, rc); + } - info->lti_ostid.oi_id = le64_to_cpu(objs[i].l_object_id); - /* XXX: support for DNE? */ - info->lti_ostid.oi_seq = le64_to_cpu(objs[i].l_object_seq); - idx = le64_to_cpu(objs[i].l_ost_idx); - fid_ostid_unpack(&info->lti_fid, &info->lti_ostid, idx); - - /* - * XXX: assertion is left for testing, to make - * sure we never process requests till configuration - * is completed. to be changed to -EINVAL - */ - - lod_getref(md); - LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx)); - LASSERT(OST_TGT(md,idx)); - LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx); nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev; - lod_putref(md); + lod_putref(md, &md->lod_ost_descs); + /* In the function below, .hs_keycmp resolves to + * u_obj_hop_keycmp() */ + /* coverity[overrun-buffer-val] */ o = lu_object_find_at(env, nd, &info->lti_fid, NULL); if (IS_ERR(o)) GOTO(out, rc = PTR_ERR(o)); @@ -498,15 +840,37 @@ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo, n = lu_object_locate(o->lo_header, nd->ld_type); LASSERT(n); - lo->ldo_stripe[i] = container_of(n, struct dt_object, do_lu); + stripe[i] = container_of(n, struct dt_object, do_lu); } out: + if (rc != 0) { + for (i = 0; i < stripe_len; i++) + if (stripe[i] != NULL) + lu_object_put(env, &stripe[i]->do_lu); + + OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len); + lo->ldo_stripenr = 0; + } else { + lo->ldo_stripe = stripe; + lo->ldo_stripes_allocated = stripe_len; + } + RETURN(rc); } -/* - * Parse striping information stored in lti_ea_store +/** + * Instantiate objects for striping. + * + * Parse striping information in \a buf and instantiate the objects + * representing the stripes. + * + * \param[in] env execution environment for this thread + * \param[in] lo LOD object + * \param[in] buf buffer storing LOV EA to parse + * + * \retval 0 if parsing and objects creation succeed + * \retval negative error number on failure */ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo, const struct lu_buf *buf) @@ -514,6 +878,7 @@ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo, struct lov_mds_md_v1 *lmm; struct lov_ost_data_v1 *objs; __u32 magic; + __u32 pattern; int rc = 0; ENTRY; @@ -523,15 +888,20 @@ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo, lmm = (struct lov_mds_md_v1 *) buf->lb_buf; magic = le32_to_cpu(lmm->lmm_magic); + pattern = le32_to_cpu(lmm->lmm_pattern); if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) GOTO(out, rc = -EINVAL); - if (le32_to_cpu(lmm->lmm_pattern) != LOV_PATTERN_RAID0) + if (lov_pattern(pattern) != LOV_PATTERN_RAID0) GOTO(out, rc = -EINVAL); + lo->ldo_pattern = pattern; lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size); - lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count); lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen); + lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count); + /* released file stripenr fixup. */ + if (pattern & LOV_PATTERN_F_RELEASED) + lo->ldo_stripenr = 0; LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic)); @@ -543,68 +913,263 @@ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo, objs = &lmm->lmm_objects[0]; } - rc = lod_initialize_objects(env, lo, objs); + if (lo->ldo_stripenr > 0) + rc = lod_initialize_objects(env, lo, objs); out: RETURN(rc); } -/* - * Load and parse striping information, create in-core representation for the - * stripes +/** + * Initialize the object representing the stripes. + * + * Unless the stripes are initialized already, fetch LOV (for regular + * objects) or LMV (for directory objects) EA and call lod_parse_striping() + * to instantiate the objects representing the stripes. + * + * \param[in] env execution environment for this thread + * \param[in,out] lo LOD object + * + * \retval 0 if parsing and object creation succeed + * \retval negative error number on failure */ -int lod_load_striping(const struct lu_env *env, struct lod_object *lo) +int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo) { struct lod_thread_info *info = lod_env_info(env); + struct lu_buf *buf = &info->lti_buf; struct dt_object *next = dt_object_child(&lo->ldo_obj); - int rc; + int rc = 0; ENTRY; - /* - * currently this code is supposed to be called from declaration - * phase only, thus the object is not expected to be locked by caller - */ - dt_write_lock(env, next, 0); /* already initialized? */ - if (lo->ldo_stripe) { - int i; - /* check validity */ - for (i = 0; i < lo->ldo_stripenr; i++) - LASSERTF(lo->ldo_stripe[i], "stripe %d is NULL\n", i); + if (lo->ldo_stripe != NULL) GOTO(out, rc = 0); - } if (!dt_object_exists(next)) GOTO(out, rc = 0); - /* only regular files can be striped */ - if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG)) + /* Do not load stripe for slaves of striped dir */ + if (lo->ldo_dir_slave_stripe) GOTO(out, rc = 0); - LASSERT(lo->ldo_stripenr == 0); - - rc = lod_get_lov_ea(env, lo); - if (rc <= 0) - GOTO(out, rc); + if (S_ISREG(lu_object_attr(lod2lu_obj(lo)))) { + rc = lod_get_lov_ea(env, lo); + if (rc <= 0) + GOTO(out, rc); + /* + * there is LOV EA (striping information) in this object + * let's parse it and create in-core objects for the stripes + */ + buf->lb_buf = info->lti_ea_store; + buf->lb_len = info->lti_ea_store_size; + rc = lod_parse_striping(env, lo, buf); + } else if (S_ISDIR(lu_object_attr(lod2lu_obj(lo)))) { + rc = lod_get_lmv_ea(env, lo); + if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1)) + GOTO(out, rc = rc > 0 ? -EINVAL : rc); + + buf->lb_buf = info->lti_ea_store; + buf->lb_len = info->lti_ea_store_size; + if (rc == sizeof(struct lmv_mds_md_v1)) { + rc = lod_load_lmv_shards(env, lo, buf, true); + if (buf->lb_buf != info->lti_ea_store) { + OBD_FREE_LARGE(info->lti_ea_store, + info->lti_ea_store_size); + info->lti_ea_store = buf->lb_buf; + info->lti_ea_store_size = buf->lb_len; + } + + if (rc < 0) + GOTO(out, rc); + } - /* - * there is LOV EA (striping information) in this object - * let's parse it and create in-core objects for the stripes - */ - info->lti_buf.lb_buf = info->lti_ea_store; - info->lti_buf.lb_len = info->lti_ea_store_size; - rc = lod_parse_striping(env, lo, &info->lti_buf); + /* + * there is LOV EA (striping information) in this object + * let's parse it and create in-core objects for the stripes + */ + rc = lod_parse_dir_striping(env, lo, buf); + } out: + RETURN(rc); +} + +/** + * A generic function to initialize the stripe objects. + * + * A protected version of lod_load_striping_locked() - load the striping + * information from storage, parse that and instantiate LU objects to + * represent the stripes. The LOD object \a lo supplies a pointer to the + * next sub-object in the LU stack so we can lock it. Also use \a lo to + * return an array of references to the newly instantiated objects. + * + * \param[in] env execution environment for this thread + * \param[in,out] lo LOD object, where striping is stored and + * which gets an array of references + * + * \retval 0 if parsing and object creation succeed + * \retval negative error number on failure + **/ +int lod_load_striping(const struct lu_env *env, struct lod_object *lo) +{ + struct dt_object *next = dt_object_child(&lo->ldo_obj); + int rc = 0; + + /* currently this code is supposed to be called from declaration + * phase only, thus the object is not expected to be locked by caller */ + dt_write_lock(env, next, 0); + rc = lod_load_striping_locked(env, lo); dt_write_unlock(env, next); + return rc; +} + +/** + * Verify striping. + * + * Check the validity of all fields including the magic, stripe size, + * stripe count, stripe offset and that the pool is present. Also check + * that each target index points to an existing target. The additional + * \a is_from_disk turns additional checks. In some cases zero fields + * are allowed (like pattern=0). + * + * \param[in] d LOD device + * \param[in] buf buffer with LOV EA to verify + * \param[in] is_from_disk 0 - from user, allow some fields to be 0 + * 1 - from disk, do not allow + * + * \retval 0 if the striping is valid + * \retval -EINVAL if striping is invalid + */ +int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf, + bool is_from_disk) +{ + struct lov_user_md_v1 *lum; + struct lov_user_md_v3 *lum3; + struct pool_desc *pool = NULL; + __u32 magic; + __u32 stripe_size; + __u16 stripe_count; + __u16 stripe_offset; + size_t lum_size; + int rc = 0; + ENTRY; + + lum = buf->lb_buf; + + LASSERT(sizeof(*lum) < sizeof(*lum3)); + + if (buf->lb_len < sizeof(*lum)) { + CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md\n", + buf->lb_len); + GOTO(out, rc = -EINVAL); + } + + magic = le32_to_cpu(lum->lmm_magic); + if (magic != LOV_USER_MAGIC_V1 && + magic != LOV_USER_MAGIC_V3 && + magic != LOV_MAGIC_V1_DEF && + magic != LOV_MAGIC_V3_DEF) { + CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic); + GOTO(out, rc = -EINVAL); + } + + /* the user uses "0" for default stripe pattern normally. */ + if (!is_from_disk && lum->lmm_pattern == 0) + lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0); + + if (le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) { + CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n", + le32_to_cpu(lum->lmm_pattern)); + GOTO(out, rc = -EINVAL); + } + + /* 64kB is the largest common page size we see (ia64), and matches the + * check in lfs */ + stripe_size = le32_to_cpu(lum->lmm_stripe_size); + if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) { + CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n", + stripe_size, LOV_MIN_STRIPE_SIZE); + GOTO(out, rc = -EINVAL); + } + + stripe_offset = le16_to_cpu(lum->lmm_stripe_offset); + if (stripe_offset != LOV_OFFSET_DEFAULT) { + /* if offset is not within valid range [0, osts_size) */ + if (stripe_offset >= d->lod_osts_size) { + CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n", + stripe_offset, d->lod_osts_size); + GOTO(out, rc = -EINVAL); + } + + /* if lmm_stripe_offset is *not* in bitmap */ + if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) { + CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n", + stripe_offset); + GOTO(out, rc = -EINVAL); + } + } + + if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF) + lum_size = offsetof(struct lov_user_md_v1, + lmm_objects[0]); + else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF) + lum_size = offsetof(struct lov_user_md_v3, + lmm_objects[0]); + else + GOTO(out, rc = -EINVAL); + + stripe_count = le16_to_cpu(lum->lmm_stripe_count); + if (buf->lb_len != lum_size) { + CDEBUG(D_IOCTL, "invalid buf len %zu for lov_user_md with " + "magic %#x and stripe_count %u\n", + buf->lb_len, magic, stripe_count); + GOTO(out, rc = -EINVAL); + } + + if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)) + goto out; + + lum3 = buf->lb_buf; + if (buf->lb_len < sizeof(*lum3)) { + CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md_v3\n", + buf->lb_len); + GOTO(out, rc = -EINVAL); + } + + /* In the function below, .hs_keycmp resolves to + * pool_hashkey_keycmp() */ + /* coverity[overrun-buffer-val] */ + pool = lod_find_pool(d, lum3->lmm_pool_name); + if (pool == NULL) + goto out; + + if (stripe_offset != LOV_OFFSET_DEFAULT) { + rc = lod_check_index_in_pool(stripe_offset, pool); + if (rc < 0) + GOTO(out, rc = -EINVAL); + } + + if (is_from_disk && stripe_count > pool_tgt_count(pool)) { + CDEBUG(D_IOCTL, + "stripe count %u > # OSTs %u in the pool\n", + stripe_count, pool_tgt_count(pool)); + GOTO(out, rc = -EINVAL); + } + +out: + if (pool != NULL) + lod_pool_putref(pool); + RETURN(rc); } void lod_fix_desc_stripe_size(__u64 *val) { - if (*val < PTLRPC_MAX_BRW_SIZE) { - LCONSOLE_WARN("Increasing default stripe size to min %u\n", - PTLRPC_MAX_BRW_SIZE); - *val = PTLRPC_MAX_BRW_SIZE; + if (*val < LOV_MIN_STRIPE_SIZE) { + if (*val != 0) + LCONSOLE_INFO("Increasing default stripe size to " + "minimum value %u\n", + LOV_DESC_STRIPE_SIZE_DEFAULT); + *val = LOV_DESC_STRIPE_SIZE_DEFAULT; } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) { *val &= ~(LOV_MIN_STRIPE_SIZE - 1); LCONSOLE_WARN("Changing default stripe size to "LPU64" (a " @@ -632,9 +1197,14 @@ void lod_fix_desc_qos_maxage(__u32 *val) { /* fix qos_maxage */ if (*val == 0) - *val = QOS_DEFAULT_MAXAGE; + *val = LOV_DESC_QOS_MAXAGE_DEFAULT; } +/** + * Used to fix insane default striping. + * + * \param[in] desc striping description + */ void lod_fix_desc(struct lov_desc *desc) { lod_fix_desc_stripe_size(&desc->ld_default_stripe_size); @@ -643,9 +1213,17 @@ void lod_fix_desc(struct lov_desc *desc) lod_fix_desc_qos_maxage(&desc->ld_qos_maxage); } +/** + * Initialize the structures used to store pools and default striping. + * + * \param[in] lod LOD device + * \param[in] lcfg configuration structure storing default striping. + * + * \retval 0 if initialization succeeds + * \retval negative error number on failure + */ int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg) { - struct lprocfs_static_vars lvars = { 0 }; struct obd_device *obd; struct lov_desc *desc; int rc; @@ -688,8 +1266,8 @@ int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg) lod->lod_sp_me = LUSTRE_SP_CLI; /* Set up allocation policy (QoS and RR) */ - CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list); - cfs_init_rwsem(&lod->lod_qos.lq_rw_sem); + INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list); + init_rwsem(&lod->lod_qos.lq_rw_sem); lod->lod_qos.lq_dirty = 1; lod->lod_qos.lq_rr.lqr_dirty = 1; lod->lod_qos.lq_reset = 1; @@ -697,11 +1275,6 @@ int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg) lod->lod_qos.lq_prio_free = 232; /* Default threshold for rr (roughly 17%) */ lod->lod_qos.lq_threshold_rr = 43; - /* Init statfs fields */ - OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data); - if (NULL == lod->lod_qos.lq_statfs_data) - RETURN(-ENOMEM); - cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq); /* Set up OST pool environment */ lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS, @@ -711,9 +1284,10 @@ int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg) CFS_HASH_MAX_THETA, &pool_hash_operations, CFS_HASH_DEFAULT); - if (!lod->lod_pools_hash_body) - GOTO(out_statfs, rc = -ENOMEM); - CFS_INIT_LIST_HEAD(&lod->lod_pool_list); + if (lod->lod_pools_hash_body == NULL) + RETURN(-ENOMEM); + + INIT_LIST_HEAD(&lod->lod_pool_list); lod->lod_pool_count = 0; rc = lod_ost_pool_init(&lod->lod_pool_info, 0); if (rc) @@ -722,96 +1296,41 @@ int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg) if (rc) GOTO(out_pool_info, rc); - /* the OST array and bitmap are allocated/grown dynamically as OSTs are - * added to the LOD, see lod_add_device() */ - lod->lod_ost_bitmap = NULL; - lod->lod_osts_size = 0; - lod->lod_ostnr = 0; - - lod->lod_death_row = 0; - lod->lod_refcount = 0; - - lprocfs_lod_init_vars(&lvars); - lprocfs_obd_setup(obd, lvars.obd_vars); - -#ifdef LPROCFS - rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd", - 0444, &lod_proc_target_fops, obd); - if (rc) { - CWARN("%s: Error adding the target_obd file %d\n", - obd->obd_name, rc); - GOTO(out_lproc, rc); - } - lod->lod_pool_proc_entry = lprocfs_register("pools", - obd->obd_proc_entry, - NULL, NULL); - if (IS_ERR(lod->lod_pool_proc_entry)) { - int ret = PTR_ERR(lod->lod_pool_proc_entry); - lod->lod_pool_proc_entry = NULL; - CWARN("%s: Failed to create pool proc file %d\n", - obd->obd_name, ret); - rc = lod_pools_fini(lod); - RETURN(ret); - } -#endif - RETURN(0); -out_lproc: - lprocfs_obd_cleanup(obd); - lod_ost_pool_free(&lod->lod_qos.lq_rr.lqr_pool); out_pool_info: lod_ost_pool_free(&lod->lod_pool_info); out_hash: cfs_hash_putref(lod->lod_pools_hash_body); -out_statfs: - OBD_FREE_PTR(lod->lod_qos.lq_statfs_data); + return rc; } +/** + * Release the structures describing the pools. + * + * \param[in] lod LOD device from which we release the structures + * + * \retval 0 always + */ int lod_pools_fini(struct lod_device *lod) { struct obd_device *obd = lod2obd(lod); - cfs_list_t *pos, *tmp; - struct pool_desc *pool; + struct pool_desc *pool, *tmp; ENTRY; - cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) { - pool = cfs_list_entry(pos, struct pool_desc, pool_list); + list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) { /* free pool structs */ CDEBUG(D_INFO, "delete pool %p\n", pool); + /* In the function below, .hs_keycmp resolves to + * pool_hashkey_keycmp() */ + /* coverity[overrun-buffer-val] */ lod_pool_del(obd, pool->pool_name); } - if (lod->lod_osts_size > 0) { - int idx; - lod_getref(lod); - cfs_mutex_lock(&lod->lod_mutex); - cfs_foreach_bit(lod->lod_ost_bitmap, idx) - __lod_del_device(lod, idx); - cfs_mutex_unlock(&lod->lod_mutex); - lod_putref(lod); - CFS_FREE_BITMAP(lod->lod_ost_bitmap); - for (idx = 0; idx < OST_PTRS; idx++) { - if (lod->lod_ost_idx[idx]) - OBD_FREE_PTR(lod->lod_ost_idx[idx]); - } - lod->lod_osts_size = 0; - } - cfs_hash_putref(lod->lod_pools_hash_body); lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool)); lod_ost_pool_free(&lod->lod_pool_info); - /* clear pools parent proc entry only after all pools are killed */ - if (lod->lod_pool_proc_entry) { - lprocfs_remove(&lod->lod_pool_proc_entry); - lod->lod_pool_proc_entry = NULL; - } - - lprocfs_obd_cleanup(obd); - - OBD_FREE_PTR(lod->lod_qos.lq_statfs_data); RETURN(0); } -