4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, Intel Corporation.
29 * lustre/lod/lod_lov.c
31 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
35 # define EXPORT_SYMTAB
37 #define DEBUG_SUBSYSTEM S_MDS
39 #include <obd_class.h>
42 #include "lod_internal.h"
45 * Keep a refcount of lod->ltd_tgts usage to prevent racing with
46 * addition/deletion. Any function that expects lov_tgts to remain stationary
49 * \param lod - is the lod device from which we want to grab a reference
51 void lod_getref(struct lod_tgt_descs *ltd)
53 down_read(<d->ltd_rw_sem);
54 mutex_lock(<d->ltd_mutex);
56 mutex_unlock(<d->ltd_mutex);
60 * Companion of lod_getref() to release a reference on the lod table.
61 * If this is the last reference and the ost entry was scheduled for deletion,
62 * the descriptor is removed from the array.
64 * \param lod - is the lod device from which we release a reference
66 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
68 mutex_lock(<d->ltd_mutex);
70 if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
71 struct lod_tgt_desc *tgt_desc, *tmp;
75 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
78 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
79 tgt_desc = LTD_TGT(ltd, idx);
82 if (!tgt_desc->ltd_reap)
85 cfs_list_add(&tgt_desc->ltd_kill, &kill);
86 LTD_TGT(ltd, idx) = NULL;
87 /*FIXME: only support ost pool for now */
88 if (ltd == &lod->lod_ost_descs) {
89 lod_ost_pool_remove(&lod->lod_pool_info, idx);
90 if (tgt_desc->ltd_active)
91 lod->lod_desc.ld_active_tgt_count--;
94 cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
97 mutex_unlock(<d->ltd_mutex);
98 up_read(<d->ltd_rw_sem);
100 cfs_list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
102 cfs_list_del(&tgt_desc->ltd_kill);
103 if (ltd == &lod->lod_ost_descs) {
104 /* remove from QoS structures */
105 rc = qos_del_tgt(lod, tgt_desc);
107 CERROR("%s: qos_del_tgt(%s) failed:"
109 lod2obd(lod)->obd_name,
110 obd_uuid2str(&tgt_desc->ltd_uuid),
113 rc = obd_disconnect(tgt_desc->ltd_exp);
115 CERROR("%s: failed to disconnect %s: rc = %d\n",
116 lod2obd(lod)->obd_name,
117 obd_uuid2str(&tgt_desc->ltd_uuid), rc);
118 OBD_FREE_PTR(tgt_desc);
121 mutex_unlock(<d->ltd_mutex);
122 up_read(<d->ltd_rw_sem);
126 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
128 cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
132 /* grab write reference on the lod. Relocating the array requires
133 * exclusive access */
135 down_write(<d->ltd_rw_sem);
136 if (newsize <= ltd->ltd_tgts_size)
137 /* someone else has already resize the array */
140 /* allocate new bitmap */
141 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
143 GOTO(out, rc = -ENOMEM);
145 if (ltd->ltd_tgts_size > 0) {
146 /* the bitmap already exists, we need
147 * to copy data from old one */
148 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
149 old_bitmap = ltd->ltd_tgt_bitmap;
152 ltd->ltd_tgts_size = newsize;
153 ltd->ltd_tgt_bitmap = new_bitmap;
156 CFS_FREE_BITMAP(old_bitmap);
158 CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
162 up_write(<d->ltd_rw_sem);
167 * Connect LOD to a new OSP and add it to the device table.
169 * \param env - is the environment passed by the caller
170 * \param lod - is the LOD device to be connected to the new OSP
171 * \param osp - is the name of OSP device name about to be added
172 * \param index - is the OSP index
173 * \param gen - is the generation number
174 * \param tgt_index - is the group of the OSP.
175 * \param type - is the type of device (mdc or osc)
177 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
178 char *osp, unsigned index, unsigned gen, int tgt_index,
179 char *type, int active)
181 struct obd_connect_data *data = NULL;
182 struct obd_export *exp = NULL;
183 struct obd_device *obd;
184 struct lu_device *ldev;
187 struct lod_tgt_desc *tgt_desc;
188 struct lod_tgt_descs *ltd;
189 struct obd_uuid obd_uuid;
192 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
195 CERROR("request to add OBD %s with invalid generation: %d\n",
200 obd_str2uuid(&obd_uuid, osp);
202 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
203 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
205 CERROR("can't find %s device\n", osp);
213 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
214 data->ocd_version = LUSTRE_VERSION_CODE;
215 data->ocd_index = index;
217 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
218 data->ocd_connect_flags |= OBD_CONNECT_AT |
221 #ifdef HAVE_LRU_RESIZE_SUPPORT
222 OBD_CONNECT_LRU_RESIZE |
225 OBD_CONNECT_OSS_CAPA |
226 OBD_CONNECT_REQPORTAL |
227 OBD_CONNECT_SKIP_ORPHAN |
229 OBD_CONNECT_LVB_TYPE |
230 OBD_CONNECT_VERSION |
231 OBD_CONNECT_PINGLESS;
233 data->ocd_group = tgt_index;
234 ltd = &lod->lod_ost_descs;
236 struct obd_import *imp = obd->u.cli.cl_import;
238 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
239 data->ocd_connect_flags |= OBD_CONNECT_ACL |
240 OBD_CONNECT_MDS_CAPA |
241 OBD_CONNECT_OSS_CAPA |
243 OBD_CONNECT_MDS_MDS |
247 OBD_CONNECT_PINGLESS;
248 /* XXX set MDS-MDS flags, remove this when running this
250 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
251 spin_lock(&imp->imp_lock);
252 imp->imp_server_timeout = 1;
253 spin_unlock(&imp->imp_lock);
254 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
255 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
257 ltd = &lod->lod_mdt_descs;
260 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
263 CERROR("%s: cannot connect to next dev %s (%d)\n",
264 obd->obd_name, osp, rc);
268 LASSERT(obd->obd_lu_dev);
269 LASSERT(obd->obd_lu_dev->ld_site = lod->lod_dt_dev.dd_lu_dev.ld_site);
271 ldev = obd->obd_lu_dev;
274 /* Allocate ost descriptor and fill it */
275 OBD_ALLOC_PTR(tgt_desc);
277 GOTO(out_conn, rc = -ENOMEM);
279 tgt_desc->ltd_tgt = d;
280 tgt_desc->ltd_exp = exp;
281 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
282 tgt_desc->ltd_gen = gen;
283 tgt_desc->ltd_index = index;
284 tgt_desc->ltd_active = active;
287 if (index >= ltd->ltd_tgts_size) {
288 /* we have to increase the size of the lod_osts array */
291 newsize = max(ltd->ltd_tgts_size, (__u32)2);
292 while (newsize < index + 1)
293 newsize = newsize << 1;
295 /* lod_bitmap_resize() needs lod_rw_sem
296 * which we hold with th reference */
297 lod_putref(lod, ltd);
299 rc = ltd_bitmap_resize(ltd, newsize);
306 mutex_lock(<d->ltd_mutex);
307 if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
308 CERROR("%s: device %d is registered already\n", obd->obd_name,
310 GOTO(out_mutex, rc = -EEXIST);
313 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
314 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
315 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
316 CERROR("can't allocate index to add %s\n",
318 GOTO(out_mutex, rc = -ENOMEM);
322 if (!strcmp(LUSTRE_OSC_NAME, type)) {
323 /* pool and qos are not supported for MDS stack yet */
324 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
327 CERROR("%s: can't set up pool, failed with %d\n",
332 rc = qos_add_tgt(lod, tgt_desc);
334 CERROR("%s: qos_add_tgt failed with %d\n",
339 /* The new OST is now a full citizen */
340 if (index >= lod->lod_desc.ld_tgt_count)
341 lod->lod_desc.ld_tgt_count = index + 1;
343 lod->lod_desc.ld_active_tgt_count++;
346 LTD_TGT(ltd, index) = tgt_desc;
347 cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
349 mutex_unlock(<d->ltd_mutex);
350 lod_putref(lod, ltd);
351 if (lod->lod_recovery_completed)
352 ldev->ld_ops->ldo_recovery_complete(env, ldev);
357 lod_ost_pool_remove(&lod->lod_pool_info, index);
359 mutex_unlock(<d->ltd_mutex);
360 lod_putref(lod, ltd);
362 OBD_FREE_PTR(tgt_desc);
370 * helper function to schedule OST removal from the device table
372 static void __lod_del_device(struct lod_tgt_descs *ltd,
375 LASSERT(LTD_TGT(ltd, idx));
376 if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
377 LTD_TGT(ltd, idx)->ltd_reap = 1;
378 ltd->ltd_death_row++;
382 int lod_fini_tgt(struct lod_device *lod, struct lod_tgt_descs *ltd)
386 if (ltd->ltd_tgts_size <= 0)
389 mutex_lock(<d->ltd_mutex);
390 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
391 __lod_del_device(ltd, idx);
392 mutex_unlock(<d->ltd_mutex);
393 lod_putref(lod, ltd);
394 CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
395 for (idx = 0; idx < TGT_PTRS; idx++) {
396 if (ltd->ltd_tgt_idx[idx])
397 OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
399 ltd->ltd_tgts_size = 0;
404 * Add support for administratively disabled OST (through the MGS).
405 * Schedule a target for deletion. Disconnection and real removal from the
406 * table takes place in lod_putref() once the last table user release its
409 * \param env - is the environment passed by the caller
410 * \param lod - is the lod device currently connected to the OSP about to be
412 * \param osp - is the name of OSP device about to be removed
413 * \param idx - is the OSP index
414 * \param gen - is the generation number, not used currently
416 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
417 struct lod_tgt_descs *ltd, char *osp, unsigned idx,
420 struct obd_device *obd;
422 struct obd_uuid uuid;
425 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
427 obd_str2uuid(&uuid, osp);
429 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
430 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
432 CERROR("can't find %s device\n", osp);
437 CERROR("%s: request to remove OBD %s with invalid generation %d"
438 "\n", obd->obd_name, osp, gen);
442 obd_str2uuid(&uuid, osp);
445 mutex_lock(<d->ltd_mutex);
446 /* check that the index is allocated in the bitmap */
447 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
448 !LTD_TGT(ltd, idx)) {
449 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
450 GOTO(out, rc = -EINVAL);
453 /* check that the UUID matches */
454 if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
455 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
456 obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
458 GOTO(out, rc = -EINVAL);
461 __lod_del_device(ltd, idx);
464 mutex_unlock(<d->ltd_mutex);
465 lod_putref(lod, ltd);
469 int lod_ea_store_resize(struct lod_thread_info *info, int size)
471 int round = size_roundup_power2(size);
473 LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
474 if (info->lti_ea_store) {
475 LASSERT(info->lti_ea_store_size);
476 LASSERT(info->lti_ea_store_size < round);
477 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
478 info->lti_ea_store_size, round);
479 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
480 info->lti_ea_store = NULL;
481 info->lti_ea_store_size = 0;
484 OBD_ALLOC_LARGE(info->lti_ea_store, round);
485 if (info->lti_ea_store == NULL)
487 info->lti_ea_store_size = round;
492 * generate and write LOV EA for given striped object
494 int lod_generate_and_set_lovea(const struct lu_env *env,
495 struct lod_object *lo, struct thandle *th)
497 struct lod_thread_info *info = lod_env_info(env);
498 struct dt_object *next = dt_object_child(&lo->ldo_obj);
499 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
500 struct lov_mds_md_v1 *lmm;
501 struct lov_ost_data_v1 *objs;
508 LASSERT(lo->ldo_stripenr > 0);
510 magic = lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
511 lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
512 if (info->lti_ea_store_size < lmm_size) {
513 rc = lod_ea_store_resize(info, lmm_size);
518 lmm = info->lti_ea_store;
520 lmm->lmm_magic = cpu_to_le32(magic);
521 lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
522 lmm->lmm_object_id = cpu_to_le64(fid_ver_oid(fid));
523 lmm->lmm_object_seq = cpu_to_le64(fid_seq(fid));
524 lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
525 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
526 lmm->lmm_layout_gen = 0;
527 if (magic == LOV_MAGIC_V1) {
528 objs = &lmm->lmm_objects[0];
530 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
531 cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
532 sizeof(v3->lmm_pool_name));
533 if (cplen >= sizeof(v3->lmm_pool_name))
535 objs = &v3->lmm_objects[0];
538 for (i = 0; i < lo->ldo_stripenr; i++) {
539 const struct lu_fid *fid;
540 struct lod_device *lod;
543 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
544 LASSERT(lo->ldo_stripe[i]);
545 fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
547 rc = fid_ostid_pack(fid, &info->lti_ostid);
550 objs[i].l_object_id = cpu_to_le64(info->lti_ostid.oi_id);
551 objs[i].l_object_seq = cpu_to_le64(info->lti_ostid.oi_seq);
552 objs[i].l_ost_gen = cpu_to_le32(0);
553 rc = lod_fld_lookup(env, lod, fid, &index, LU_SEQ_RANGE_OST);
555 CERROR("%s: Can not locate "DFID": rc = %d\n",
556 lod2obd(lod)->obd_name, PFID(fid), rc);
559 objs[i].l_ost_idx = cpu_to_le32(index);
562 info->lti_buf.lb_buf = lmm;
563 info->lti_buf.lb_len = lmm_size;
564 rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
570 int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
572 struct lod_thread_info *info = lod_env_info(env);
573 struct dt_object *next = dt_object_child(&lo->ldo_obj);
579 if (unlikely(info->lti_ea_store_size == 0)) {
580 /* just to enter in allocation block below */
584 info->lti_buf.lb_buf = info->lti_ea_store;
585 info->lti_buf.lb_len = info->lti_ea_store_size;
586 rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV,
589 /* if object is not striped or inaccessible */
594 /* EA doesn't fit, reallocate new buffer */
595 rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV,
603 rc = lod_ea_store_resize(info, rc);
612 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
615 struct lod_thread_info *info = lod_env_info(env);
616 struct lod_object *lo = lod_dt_obj(dt);
617 struct dt_object *next = dt_object_child(dt);
618 struct lov_user_md_v3 *v3;
623 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
626 * store striping defaults into new directory
627 * used to implement defaults inheritance
630 /* probably nothing to inherite */
631 if (lo->ldo_striping_cached == 0)
634 if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
635 lo->ldo_def_stripe_offset))
638 /* XXX: use thread info */
643 v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
644 v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
645 v3->lmm_object_id = 0;
646 v3->lmm_object_seq = 0;
647 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
648 v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
649 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
651 cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
652 sizeof(v3->lmm_pool_name));
653 if (cplen >= sizeof(v3->lmm_pool_name)) {
659 info->lti_buf.lb_buf = v3;
660 info->lti_buf.lb_len = sizeof(*v3);
661 rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
670 * allocate array of objects pointers, find/create objects
671 * stripenr and other fields should be initialized by this moment
673 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
674 struct lov_ost_data_v1 *objs)
676 struct lod_thread_info *info = lod_env_info(env);
677 struct lod_device *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
678 struct lu_object *o, *n;
679 struct lu_device *nd;
684 LASSERT(lo->ldo_stripe == NULL);
685 LASSERT(lo->ldo_stripenr > 0);
686 LASSERT(lo->ldo_stripe_size > 0);
688 i = sizeof(struct dt_object *) * lo->ldo_stripenr;
689 OBD_ALLOC(lo->ldo_stripe, i);
690 if (lo->ldo_stripe == NULL)
691 GOTO(out, rc = -ENOMEM);
692 lo->ldo_stripes_allocated = lo->ldo_stripenr;
694 for (i = 0; i < lo->ldo_stripenr; i++) {
695 info->lti_ostid.oi_id = le64_to_cpu(objs[i].l_object_id);
696 info->lti_ostid.oi_seq = le64_to_cpu(objs[i].l_object_seq);
697 idx = le64_to_cpu(objs[i].l_ost_idx);
698 fid_ostid_unpack(&info->lti_fid, &info->lti_ostid, idx);
699 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
700 PFID(&info->lti_fid));
702 * XXX: assertion is left for testing, to make
703 * sure we never process requests till configuration
704 * is completed. to be changed to -EINVAL
707 lod_getref(&md->lod_ost_descs);
708 LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx));
709 LASSERT(OST_TGT(md,idx));
710 LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx);
711 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
712 lod_putref(md, &md->lod_ost_descs);
714 /* In the function below, .hs_keycmp resolves to
715 * u_obj_hop_keycmp() */
716 /* coverity[overrun-buffer-val] */
717 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
719 GOTO(out, rc = PTR_ERR(o));
721 n = lu_object_locate(o->lo_header, nd->ld_type);
724 lo->ldo_stripe[i] = container_of(n, struct dt_object, do_lu);
732 * Parse striping information stored in lti_ea_store
734 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
735 const struct lu_buf *buf)
737 struct lov_mds_md_v1 *lmm;
738 struct lov_ost_data_v1 *objs;
744 LASSERT(buf->lb_buf);
745 LASSERT(buf->lb_len);
747 lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
748 magic = le32_to_cpu(lmm->lmm_magic);
750 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
751 GOTO(out, rc = -EINVAL);
752 if (le32_to_cpu(lmm->lmm_pattern) != LOV_PATTERN_RAID0)
753 GOTO(out, rc = -EINVAL);
755 lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
756 lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
757 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
759 LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
761 if (magic == LOV_MAGIC_V3) {
762 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
763 objs = &v3->lmm_objects[0];
764 lod_object_set_pool(lo, v3->lmm_pool_name);
766 objs = &lmm->lmm_objects[0];
769 rc = lod_initialize_objects(env, lo, objs);
776 * Load and parse striping information, create in-core representation for the
779 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
781 struct lod_thread_info *info = lod_env_info(env);
782 struct dt_object *next = dt_object_child(&lo->ldo_obj);
787 * currently this code is supposed to be called from declaration
788 * phase only, thus the object is not expected to be locked by caller
790 dt_write_lock(env, next, 0);
791 /* already initialized? */
792 if (lo->ldo_stripe) {
795 for (i = 0; i < lo->ldo_stripenr; i++)
796 LASSERTF(lo->ldo_stripe[i], "stripe %d is NULL\n", i);
800 if (!dt_object_exists(next))
803 /* only regular files can be striped */
804 if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG))
807 LASSERT(lo->ldo_stripenr == 0);
809 rc = lod_get_lov_ea(env, lo);
814 * there is LOV EA (striping information) in this object
815 * let's parse it and create in-core objects for the stripes
817 info->lti_buf.lb_buf = info->lti_ea_store;
818 info->lti_buf.lb_len = info->lti_ea_store_size;
819 rc = lod_parse_striping(env, lo, &info->lti_buf);
821 dt_write_unlock(env, next);
825 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
828 struct lov_user_md_v1 *lum;
829 struct lov_user_md_v3 *v3 = NULL;
830 struct pool_desc *pool = NULL;
836 if (lum->lmm_magic != LOV_USER_MAGIC_V1 &&
837 lum->lmm_magic != LOV_USER_MAGIC_V3 &&
838 lum->lmm_magic != LOV_MAGIC_V1_DEF &&
839 lum->lmm_magic != LOV_MAGIC_V3_DEF) {
840 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n",
845 if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
846 (specific == 0 && lum->lmm_pattern != 0)) {
847 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
852 /* 64kB is the largest common page size we see (ia64), and matches the
854 if (lum->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
855 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
856 lum->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
860 /* an offset of -1 is treated as a "special" valid offset */
861 if (lum->lmm_stripe_offset != (typeof(lum->lmm_stripe_offset))(-1)) {
862 /* if offset is not within valid range [0, osts_size) */
863 if (lum->lmm_stripe_offset >= d->lod_osts_size) {
864 CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
865 lum->lmm_stripe_offset, d->lod_osts_size);
869 /* if lmm_stripe_offset is *not* in bitmap */
870 if (!cfs_bitmap_check(d->lod_ost_bitmap,
871 lum->lmm_stripe_offset)) {
872 CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
873 lum->lmm_stripe_offset);
878 if (lum->lmm_magic == LOV_USER_MAGIC_V3)
882 /* In the function below, .hs_keycmp resolves to
883 * pool_hashkey_keycmp() */
884 /* coverity[overrun-buffer-val] */
885 pool = lod_find_pool(d, v3->lmm_pool_name);
888 __u16 offs = v3->lmm_stripe_offset;
890 if (offs != (typeof(v3->lmm_stripe_offset))(-1)) {
891 rc = lod_check_index_in_pool(offs, pool);
893 lod_pool_putref(pool);
898 if (specific && lum->lmm_stripe_count > pool_tgt_count(pool)) {
900 "stripe count %u > # OSTs %u in the pool\n",
901 lum->lmm_stripe_count, pool_tgt_count(pool));
902 lod_pool_putref(pool);
906 lod_pool_putref(pool);
912 void lod_fix_desc_stripe_size(__u64 *val)
914 if (*val < LOV_DEFAULT_STRIPE_SIZE) {
915 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
916 LOV_DEFAULT_STRIPE_SIZE);
917 *val = LOV_DEFAULT_STRIPE_SIZE;
918 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
919 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
920 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
922 *val, LOV_MIN_STRIPE_SIZE);
926 void lod_fix_desc_stripe_count(__u32 *val)
932 void lod_fix_desc_pattern(__u32 *val)
934 /* from lov_setstripe */
935 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
936 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
941 void lod_fix_desc_qos_maxage(__u32 *val)
945 *val = QOS_DEFAULT_MAXAGE;
948 void lod_fix_desc(struct lov_desc *desc)
950 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
951 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
952 lod_fix_desc_pattern(&desc->ld_pattern);
953 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
956 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
958 struct obd_device *obd;
959 struct lov_desc *desc;
963 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
964 LASSERT(obd != NULL);
965 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
967 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
968 CERROR("LOD setup requires a descriptor\n");
972 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
974 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
975 CERROR("descriptor size wrong: %d > %d\n",
976 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
980 if (desc->ld_magic != LOV_DESC_MAGIC) {
981 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
982 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
983 obd->obd_name, desc);
984 lustre_swab_lov_desc(desc);
986 CERROR("%s: Bad lov desc magic: %#x\n",
987 obd->obd_name, desc->ld_magic);
994 desc->ld_active_tgt_count = 0;
995 lod->lod_desc = *desc;
997 lod->lod_sp_me = LUSTRE_SP_CLI;
999 /* Set up allocation policy (QoS and RR) */
1000 CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1001 init_rwsem(&lod->lod_qos.lq_rw_sem);
1002 lod->lod_qos.lq_dirty = 1;
1003 lod->lod_qos.lq_rr.lqr_dirty = 1;
1004 lod->lod_qos.lq_reset = 1;
1005 /* Default priority is toward free space balance */
1006 lod->lod_qos.lq_prio_free = 232;
1007 /* Default threshold for rr (roughly 17%) */
1008 lod->lod_qos.lq_threshold_rr = 43;
1009 /* Init statfs fields */
1010 OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data);
1011 if (NULL == lod->lod_qos.lq_statfs_data)
1013 cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq);
1015 /* Set up OST pool environment */
1016 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1017 HASH_POOLS_MAX_BITS,
1018 HASH_POOLS_BKT_BITS, 0,
1021 &pool_hash_operations,
1023 if (!lod->lod_pools_hash_body)
1024 GOTO(out_statfs, rc = -ENOMEM);
1025 CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
1026 lod->lod_pool_count = 0;
1027 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1030 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1032 GOTO(out_pool_info, rc);
1037 lod_ost_pool_free(&lod->lod_pool_info);
1039 cfs_hash_putref(lod->lod_pools_hash_body);
1041 OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
1045 int lod_pools_fini(struct lod_device *lod)
1047 struct obd_device *obd = lod2obd(lod);
1048 cfs_list_t *pos, *tmp;
1049 struct pool_desc *pool;
1052 cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
1053 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
1054 /* free pool structs */
1055 CDEBUG(D_INFO, "delete pool %p\n", pool);
1056 /* In the function below, .hs_keycmp resolves to
1057 * pool_hashkey_keycmp() */
1058 /* coverity[overrun-buffer-val] */
1059 lod_pool_del(obd, pool->pool_name);
1062 cfs_hash_putref(lod->lod_pools_hash_body);
1063 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1064 lod_ost_pool_free(&lod->lod_pool_info);
1065 OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);