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 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * lustre/lod/lod_lov.c
31 * A set of helpers to maintain Logical Object Volume (LOV)
32 * Extended Attribute (EA) and known OST targets
34 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
37 #define DEBUG_SUBSYSTEM S_MDS
39 #include <obd_class.h>
40 #include <lustre_lfsck.h>
41 #include <lustre_lmv.h>
42 #include <lustre_swab.h>
44 #include "lod_internal.h"
47 * Increase reference count on the target table.
49 * Increase reference count on the target table usage to prevent racing with
50 * addition/deletion. Any function that expects the table to remain
51 * stationary must take a ref.
53 * \param[in] ltd target table (lod_ost_descs or lod_mdt_descs)
55 void lod_getref(struct lod_tgt_descs *ltd)
57 down_read(<d->ltd_rw_sem);
58 mutex_lock(<d->ltd_mutex);
60 mutex_unlock(<d->ltd_mutex);
64 * Decrease reference count on the target table.
66 * Companion of lod_getref() to release a reference on the target table.
67 * If this is the last reference and the OST entry was scheduled for deletion,
68 * the descriptor is removed from the table.
70 * \param[in] lod LOD device from which we release a reference
71 * \param[in] ltd target table (lod_ost_descs or lod_mdt_descs)
73 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
75 mutex_lock(<d->ltd_mutex);
77 if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
78 struct lod_tgt_desc *tgt_desc, *tmp;
79 struct list_head kill;
82 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
85 INIT_LIST_HEAD(&kill);
87 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
88 tgt_desc = LTD_TGT(ltd, idx);
91 if (!tgt_desc->ltd_reap)
94 list_add(&tgt_desc->ltd_kill, &kill);
95 /*FIXME: only support ost pool for now */
96 if (ltd == &lod->lod_ost_descs) {
97 lod_ost_pool_remove(&lod->lod_pool_info, idx);
98 if (tgt_desc->ltd_active)
99 lod->lod_desc.ld_active_tgt_count--;
101 lu_tgt_descs_del(ltd, tgt_desc);
102 ltd->ltd_death_row--;
104 mutex_unlock(<d->ltd_mutex);
105 up_read(<d->ltd_rw_sem);
107 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
109 list_del(&tgt_desc->ltd_kill);
110 if (ltd == &lod->lod_ost_descs) {
111 /* remove from QoS structures */
112 rc = lqos_del_tgt(&lod->lod_qos, tgt_desc);
114 CERROR("%s: qos_del_tgt(%s) failed:"
116 lod2obd(lod)->obd_name,
117 obd_uuid2str(&tgt_desc->ltd_uuid),
120 rc = obd_disconnect(tgt_desc->ltd_exp);
122 CERROR("%s: failed to disconnect %s: rc = %d\n",
123 lod2obd(lod)->obd_name,
124 obd_uuid2str(&tgt_desc->ltd_uuid), rc);
125 OBD_FREE_PTR(tgt_desc);
128 mutex_unlock(<d->ltd_mutex);
129 up_read(<d->ltd_rw_sem);
134 * Connect LOD to a new OSP and add it to the target table.
136 * Connect to the OSP device passed, initialize all the internal
137 * structures related to the device and add it to the target table.
139 * \param[in] env execution environment for this thread
140 * \param[in] lod LOD device to be connected to the new OSP
141 * \param[in] osp name of OSP device name to be added
142 * \param[in] index index of the new target
143 * \param[in] gen target's generation number
144 * \param[in] tgt_index OSP's group
145 * \param[in] type type of device (mdc or osc)
146 * \param[in] active state of OSP: 0 - inactive, 1 - active
148 * \retval 0 if added successfully
149 * \retval negative error number on failure
151 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
152 char *osp, unsigned index, unsigned gen, int tgt_index,
153 char *type, int active)
155 struct obd_connect_data *data = NULL;
156 struct obd_export *exp = NULL;
157 struct obd_device *obd;
158 struct lu_device *lu_dev;
159 struct dt_device *dt_dev;
161 struct lod_tgt_desc *tgt_desc;
162 struct lod_tgt_descs *ltd;
163 struct lustre_cfg *lcfg;
164 struct obd_uuid obd_uuid;
166 bool connected = false;
169 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
172 CERROR("request to add OBD %s with invalid generation: %d\n",
177 obd_str2uuid(&obd_uuid, osp);
179 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
180 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
182 CERROR("can't find %s device\n", osp);
186 LASSERT(obd->obd_lu_dev != NULL);
187 LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
189 lu_dev = obd->obd_lu_dev;
190 dt_dev = lu2dt_dev(lu_dev);
194 GOTO(out_cleanup, rc = -ENOMEM);
196 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
197 data->ocd_version = LUSTRE_VERSION_CODE;
198 data->ocd_index = index;
200 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
202 data->ocd_connect_flags |= OBD_CONNECT_AT |
205 #ifdef HAVE_LRU_RESIZE_SUPPORT
206 OBD_CONNECT_LRU_RESIZE |
209 OBD_CONNECT_REQPORTAL |
210 OBD_CONNECT_SKIP_ORPHAN |
212 OBD_CONNECT_LVB_TYPE |
213 OBD_CONNECT_VERSION |
214 OBD_CONNECT_PINGLESS |
216 OBD_CONNECT_BULK_MBITS;
218 data->ocd_group = tgt_index;
219 ltd = &lod->lod_ost_descs;
221 struct obd_import *imp = obd->u.cli.cl_import;
224 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
225 data->ocd_connect_flags |= OBD_CONNECT_ACL |
227 OBD_CONNECT_MDS_MDS |
232 OBD_CONNECT_BULK_MBITS;
233 spin_lock(&imp->imp_lock);
234 imp->imp_server_timeout = 1;
235 spin_unlock(&imp->imp_lock);
236 imp->imp_client->cli_request_portal = OUT_PORTAL;
237 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
239 ltd = &lod->lod_mdt_descs;
242 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
245 CERROR("%s: cannot connect to next dev %s (%d)\n",
246 obd->obd_name, osp, rc);
247 GOTO(out_cleanup, rc);
251 /* Allocate ost descriptor and fill it */
252 OBD_ALLOC_PTR(tgt_desc);
254 GOTO(out_cleanup, rc = -ENOMEM);
256 tgt_desc->ltd_tgt = dt_dev;
257 tgt_desc->ltd_exp = exp;
258 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
259 tgt_desc->ltd_gen = gen;
260 tgt_desc->ltd_index = index;
261 tgt_desc->ltd_active = active;
263 down_write(<d->ltd_rw_sem);
264 mutex_lock(<d->ltd_mutex);
265 lu_tgt_descs_add(ltd, tgt_desc);
267 /* pool and qos are not supported for MDS stack yet */
268 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
271 CERROR("%s: can't set up pool, failed with %d\n",
276 rc = lqos_add_tgt(&lod->lod_qos, tgt_desc);
278 CERROR("%s: qos_add_tgt failed with %d\n",
283 /* The new OST is now a full citizen */
284 if (index >= lod->lod_desc.ld_tgt_count)
285 lod->lod_desc.ld_tgt_count = index + 1;
287 lod->lod_desc.ld_active_tgt_count++;
289 mutex_unlock(<d->ltd_mutex);
290 up_write(<d->ltd_rw_sem);
292 if (lod->lod_recovery_completed)
293 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
295 if (!for_ost && lod->lod_initialized) {
296 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
298 CERROR("%s: cannot start llog on %s:rc = %d\n",
299 lod2obd(lod)->obd_name, osp, rc);
304 rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
306 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
307 "index = %u, rc = %d\n", osp, type, index, rc);
308 GOTO(out_fini_llog, rc);
312 lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
313 tgt_desc->ltd_recovery_thread);
315 down_write(<d->ltd_rw_sem);
316 mutex_lock(<d->ltd_mutex);
317 if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
318 struct ptlrpc_thread *thread;
320 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
321 OBD_FREE_PTR(thread);
324 lod_ost_pool_remove(&lod->lod_pool_info, index);
326 lu_tgt_descs_del(ltd, tgt_desc);
327 mutex_unlock(<d->ltd_mutex);
328 up_write(<d->ltd_rw_sem);
329 OBD_FREE_PTR(tgt_desc);
331 /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
332 * objects from the MDT stack. See LU-7184. */
333 lcfg = &lod_env_info(env)->lti_lustre_cfg;
334 memset(lcfg, 0, sizeof(*lcfg));
335 lcfg->lcfg_version = LUSTRE_CFG_VERSION;
336 lcfg->lcfg_command = LCFG_CLEANUP;
337 lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
346 * Schedule target removal from the target table.
348 * Mark the device as dead. The device is not removed here because it may
349 * still be in use. The device will be removed in lod_putref() when the
350 * last reference is released.
352 * \param[in] env execution environment for this thread
353 * \param[in] lod LOD device the target table belongs to
354 * \param[in] ltd target table
355 * \param[in] idx index of the target
356 * \param[in] for_ost type of the target: 0 - MDT, 1 - OST
358 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
359 struct lod_tgt_descs *ltd, unsigned idx,
362 LASSERT(LTD_TGT(ltd, idx));
364 lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
367 if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
368 struct ptlrpc_thread *thread;
370 thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
371 OBD_FREE_PTR(thread);
374 if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
375 LTD_TGT(ltd, idx)->ltd_reap = 1;
376 ltd->ltd_death_row++;
381 * Schedule removal of all the targets from the given target table.
383 * See more details in the description for __lod_del_device()
385 * \param[in] env execution environment for this thread
386 * \param[in] lod LOD device the target table belongs to
387 * \param[in] ltd target table
388 * \param[in] for_ost type of the target: MDT or OST
392 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
393 struct lod_tgt_descs *ltd, bool for_ost)
397 if (ltd->ltd_tgts_size <= 0)
401 mutex_lock(<d->ltd_mutex);
402 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
403 __lod_del_device(env, lod, ltd, idx, for_ost);
404 mutex_unlock(<d->ltd_mutex);
405 lod_putref(lod, ltd);
407 lu_tgt_descs_fini(ltd);
413 * Remove device by name.
415 * Remove a device identified by \a osp from the target table. Given
416 * the device can be in use, the real deletion happens in lod_putref().
418 * \param[in] env execution environment for this thread
419 * \param[in] lod LOD device to be connected to the new OSP
420 * \param[in] ltd target table
421 * \param[in] osp name of OSP device to be removed
422 * \param[in] idx index of the target
423 * \param[in] gen generation number, not used currently
424 * \param[in] for_ost type of the target: 0 - MDT, 1 - OST
426 * \retval 0 if the device was scheduled for removal
427 * \retval -EINVAL if no device was found
429 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
430 struct lod_tgt_descs *ltd, char *osp, unsigned idx,
431 unsigned gen, bool for_ost)
433 struct obd_device *obd;
435 struct obd_uuid uuid;
438 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
440 obd_str2uuid(&uuid, osp);
442 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
443 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
445 CERROR("can't find %s device\n", osp);
450 CERROR("%s: request to remove OBD %s with invalid generation %d"
451 "\n", obd->obd_name, osp, gen);
455 obd_str2uuid(&uuid, osp);
458 mutex_lock(<d->ltd_mutex);
459 /* check that the index is allocated in the bitmap */
460 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
461 !LTD_TGT(ltd, idx)) {
462 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
463 GOTO(out, rc = -EINVAL);
466 /* check that the UUID matches */
467 if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
468 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
469 obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
471 GOTO(out, rc = -EINVAL);
474 __lod_del_device(env, lod, ltd, idx, for_ost);
477 mutex_unlock(<d->ltd_mutex);
478 lod_putref(lod, ltd);
483 * Resize per-thread storage to hold specified size.
485 * A helper function to resize per-thread temporary storage. This storage
486 * is used to process LOV/LVM EAs and may be quite large. We do not want to
487 * allocate/release it every time, so instead we put it into the env and
488 * reallocate on demand. The memory is released when the correspondent thread
491 * \param[in] info LOD-specific storage in the environment
492 * \param[in] size new size to grow the buffer to
494 * \retval 0 on success, -ENOMEM if reallocation failed
496 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
498 __u32 round = size_roundup_power2(size);
500 if (info->lti_ea_store) {
501 LASSERT(info->lti_ea_store_size);
502 LASSERT(info->lti_ea_store_size < round);
503 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
504 info->lti_ea_store_size, round);
505 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
506 info->lti_ea_store = NULL;
507 info->lti_ea_store_size = 0;
510 OBD_ALLOC_LARGE(info->lti_ea_store, round);
511 if (info->lti_ea_store == NULL)
513 info->lti_ea_store_size = round;
518 static void lod_free_comp_buffer(struct lod_layout_component *entries,
519 __u16 count, __u32 bufsize)
521 struct lod_layout_component *entry;
524 for (i = 0; i < count; i++) {
526 if (entry->llc_pool != NULL)
527 lod_set_pool(&entry->llc_pool, NULL);
528 if (entry->llc_ostlist.op_array)
529 OBD_FREE(entry->llc_ostlist.op_array,
530 entry->llc_ostlist.op_size);
531 LASSERT(entry->llc_stripe == NULL);
532 LASSERT(entry->llc_stripes_allocated == 0);
536 OBD_FREE_LARGE(entries, bufsize);
539 void lod_free_def_comp_entries(struct lod_default_striping *lds)
541 lod_free_comp_buffer(lds->lds_def_comp_entries,
542 lds->lds_def_comp_size_cnt,
544 sizeof(*lds->lds_def_comp_entries) *
545 lds->lds_def_comp_size_cnt));
546 lds->lds_def_comp_entries = NULL;
547 lds->lds_def_comp_cnt = 0;
548 lds->lds_def_striping_is_composite = 0;
549 lds->lds_def_comp_size_cnt = 0;
553 * Resize per-thread storage to hold default striping component entries
555 * A helper function to resize per-thread temporary storage. This storage
556 * is used to hold default LOV/LVM EAs and may be quite large. We do not want
557 * to allocate/release it every time, so instead we put it into the env and
558 * reallocate it on demand. The memory is released when the correspondent
559 * thread is finished.
561 * \param[in,out] lds default striping
562 * \param[in] count new component count to grow the buffer to
564 * \retval 0 on success, -ENOMEM if reallocation failed
566 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
568 struct lod_layout_component *entries;
569 __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
571 __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
572 lds->lds_def_comp_size_cnt);
577 OBD_ALLOC_LARGE(entries, new);
581 if (lds->lds_def_comp_entries != NULL) {
582 CDEBUG(D_INFO, "default striping component size %d is not "
583 "enough, need %d\n", old, new);
584 lod_free_def_comp_entries(lds);
587 lds->lds_def_comp_entries = entries;
588 lds->lds_def_comp_size_cnt = count;
593 void lod_free_comp_entries(struct lod_object *lo)
595 if (lo->ldo_mirrors) {
596 OBD_FREE(lo->ldo_mirrors,
597 sizeof(*lo->ldo_mirrors) * lo->ldo_mirror_count);
598 lo->ldo_mirrors = NULL;
599 lo->ldo_mirror_count = 0;
601 lod_free_comp_buffer(lo->ldo_comp_entries,
603 sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
604 lo->ldo_comp_entries = NULL;
605 lo->ldo_comp_cnt = 0;
606 lo->ldo_is_composite = 0;
609 int lod_alloc_comp_entries(struct lod_object *lo,
610 int mirror_count, int comp_count)
612 LASSERT(comp_count != 0);
613 LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
615 if (mirror_count > 0) {
616 OBD_ALLOC(lo->ldo_mirrors,
617 sizeof(*lo->ldo_mirrors) * mirror_count);
618 if (!lo->ldo_mirrors)
621 lo->ldo_mirror_count = mirror_count;
624 OBD_ALLOC_LARGE(lo->ldo_comp_entries,
625 sizeof(*lo->ldo_comp_entries) * comp_count);
626 if (lo->ldo_comp_entries == NULL) {
627 OBD_FREE(lo->ldo_mirrors,
628 sizeof(*lo->ldo_mirrors) * mirror_count);
629 lo->ldo_mirror_count = 0;
633 lo->ldo_comp_cnt = comp_count;
637 int lod_fill_mirrors(struct lod_object *lo)
639 struct lod_layout_component *lod_comp;
641 __u16 mirror_id = 0xffff;
645 LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
647 if (!lo->ldo_is_composite)
650 lod_comp = &lo->ldo_comp_entries[0];
651 for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
652 int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
653 int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
655 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
656 lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
657 lo->ldo_mirrors[mirror_idx].lme_primary |= preferred;
658 lo->ldo_mirrors[mirror_idx].lme_end = i;
664 if (mirror_idx >= lo->ldo_mirror_count)
667 mirror_id = mirror_id_of(lod_comp->llc_id);
669 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
670 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
671 lo->ldo_mirrors[mirror_idx].lme_primary = preferred;
672 lo->ldo_mirrors[mirror_idx].lme_start = i;
673 lo->ldo_mirrors[mirror_idx].lme_end = i;
675 if (mirror_idx != lo->ldo_mirror_count - 1)
682 * Generate on-disk lov_mds_md structure for each layout component based on
683 * the information in lod_object->ldo_comp_entries[i].
685 * \param[in] env execution environment for this thread
686 * \param[in] lo LOD object
687 * \param[in] comp_idx index of ldo_comp_entries
688 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
689 * \param[in|out] lmm_size buffer size/lmm size
690 * \param[in] is_dir generate lov ea for dir or file? For dir case,
691 * the stripe info is from the default stripe
692 * template, which is collected in lod_ah_init(),
693 * either from parent object or root object; for
694 * file case, it's from the @lo object
696 * \retval 0 if on disk structure is created successfully
697 * \retval negative error number on failure
699 static int lod_gen_component_ea(const struct lu_env *env,
700 struct lod_object *lo, int comp_idx,
701 struct lov_mds_md *lmm, int *lmm_size,
704 struct lod_thread_info *info = lod_env_info(env);
705 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
706 struct lod_device *lod;
707 struct lov_ost_data_v1 *objs;
708 struct lod_layout_component *lod_comp;
717 &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
719 lod_comp = &lo->ldo_comp_entries[comp_idx];
721 magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
722 if (lod_comp->llc_pattern == 0) /* default striping */
723 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
725 lmm->lmm_magic = cpu_to_le32(magic);
726 lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
727 fid_to_lmm_oi(fid, &lmm->lmm_oi);
728 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
729 lmm->lmm_oi.oi.oi_id++;
730 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
732 lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
733 lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
735 * for dir and uninstantiated component, lmm_layout_gen stores
736 * default stripe offset.
738 lmm->lmm_layout_gen =
739 (is_dir || !lod_comp_inited(lod_comp)) ?
740 cpu_to_le16(lod_comp->llc_stripe_offset) :
741 cpu_to_le16(lod_comp->llc_layout_gen);
743 if (magic == LOV_MAGIC_V1) {
744 objs = &lmm->lmm_objects[0];
746 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
747 size_t cplen = strlcpy(v3->lmm_pool_name,
749 sizeof(v3->lmm_pool_name));
750 if (cplen >= sizeof(v3->lmm_pool_name))
752 objs = &v3->lmm_objects[0];
754 stripe_count = lod_comp_entry_stripe_count(lo, lod_comp, is_dir);
755 if (stripe_count == 0 && !is_dir &&
756 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
757 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
760 if (!is_dir && lo->ldo_is_composite)
761 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
763 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
766 /* generate ost_idx of this component stripe */
767 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
768 for (i = 0; i < stripe_count; i++) {
769 struct dt_object *object;
770 __u32 ost_idx = (__u32)-1UL;
771 int type = LU_SEQ_RANGE_OST;
773 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
774 object = lod_comp->llc_stripe[i];
775 /* instantiated component */
776 info->lti_fid = *lu_object_fid(&object->do_lu);
778 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
780 if (cfs_fail_val == 0)
781 cfs_fail_val = info->lti_fid.f_oid;
783 info->lti_fid.f_oid = cfs_fail_val;
786 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
789 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
790 objs[i].l_ost_gen = cpu_to_le32(0);
791 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
794 rc = lod_fld_lookup(env, lod, &info->lti_fid,
797 CERROR("%s: Can not locate "DFID": rc = %d\n",
798 lod2obd(lod)->obd_name,
799 PFID(&info->lti_fid), rc);
802 } else if (lod_comp->llc_ostlist.op_array &&
803 lod_comp->llc_ostlist.op_count) {
804 /* user specified ost list */
805 ost_idx = lod_comp->llc_ostlist.op_array[i];
808 * with un-instantiated or with no specified ost list
809 * component, its l_ost_idx does not matter.
811 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
814 if (lmm_size != NULL)
815 *lmm_size = lov_mds_md_size(stripe_count, magic);
820 * Generate on-disk lov_mds_md structure based on the information in
821 * the lod_object->ldo_comp_entries.
823 * \param[in] env execution environment for this thread
824 * \param[in] lo LOD object
825 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
826 * \param[in|out] lmm_size buffer size/lmm size
827 * \param[in] is_dir generate lov ea for dir or file? For dir case,
828 * the stripe info is from the default stripe
829 * template, which is collected in lod_ah_init(),
830 * either from parent object or root object; for
831 * file case, it's from the @lo object
833 * \retval 0 if on disk structure is created successfully
834 * \retval negative error number on failure
836 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
837 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
839 struct lov_comp_md_entry_v1 *lcme;
840 struct lov_comp_md_v1 *lcm;
841 struct lod_layout_component *comp_entries;
842 __u16 comp_cnt, mirror_cnt;
843 bool is_composite, is_foreign = false;
844 int i, rc = 0, offset;
848 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
849 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
850 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
852 lo->ldo_def_striping->lds_def_striping_is_composite;
854 comp_cnt = lo->ldo_comp_cnt;
855 mirror_cnt = lo->ldo_mirror_count;
856 comp_entries = lo->ldo_comp_entries;
857 is_composite = lo->ldo_is_composite;
858 is_foreign = lo->ldo_is_foreign;
861 LASSERT(lmm_size != NULL);
864 struct lov_foreign_md *lfm;
866 lfm = (struct lov_foreign_md *)lmm;
867 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
868 /* need to store little-endian */
869 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
870 __swab32s(&lfm->lfm_magic);
871 __swab32s(&lfm->lfm_length);
872 __swab32s(&lfm->lfm_type);
873 __swab32s(&lfm->lfm_flags);
875 *lmm_size = lo->ldo_foreign_lov_size;
879 LASSERT(comp_cnt != 0 && comp_entries != NULL);
882 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
886 lcm = (struct lov_comp_md_v1 *)lmm;
887 memset(lcm, 0, sizeof(*lcm));
889 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
890 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
891 lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
892 lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
894 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
895 LASSERT(offset % sizeof(__u64) == 0);
897 for (i = 0; i < comp_cnt; i++) {
898 struct lod_layout_component *lod_comp;
899 struct lov_mds_md *sub_md;
902 lod_comp = &comp_entries[i];
903 lcme = &lcm->lcm_entries[i];
905 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
906 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
908 /* component could be un-inistantiated */
909 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
910 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
911 lcme->lcme_timestamp =
912 cpu_to_le64(lod_comp->llc_timestamp);
913 if (lod_comp->llc_flags & LCME_FL_EXTENSION && !is_dir)
914 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
916 lcme->lcme_extent.e_start =
917 cpu_to_le64(lod_comp->llc_extent.e_start);
918 lcme->lcme_extent.e_end =
919 cpu_to_le64(lod_comp->llc_extent.e_end);
920 lcme->lcme_offset = cpu_to_le32(offset);
922 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
923 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
926 lcme->lcme_size = cpu_to_le32(size);
928 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
929 "offset:%d lmm_size:%d\n", offset, *lmm_size);
931 lcm->lcm_size = cpu_to_le32(offset);
932 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
934 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
945 * Fill lti_ea_store buffer in the environment with a value for the given
946 * EA. The buffer is reallocated if the value doesn't fit.
948 * \param[in,out] env execution environment for this thread
949 * .lti_ea_store buffer is filled with EA's value
950 * \param[in] lo LOD object
951 * \param[in] name name of the EA
953 * \retval > 0 if EA is fetched successfully
954 * \retval 0 if EA is empty
955 * \retval negative error number on failure
957 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
960 struct lod_thread_info *info = lod_env_info(env);
961 struct dt_object *next = dt_object_child(&lo->ldo_obj);
967 if (unlikely(info->lti_ea_store == NULL)) {
968 /* just to enter in allocation block below */
972 info->lti_buf.lb_buf = info->lti_ea_store;
973 info->lti_buf.lb_len = info->lti_ea_store_size;
974 rc = dt_xattr_get(env, next, &info->lti_buf, name);
977 /* if object is not striped or inaccessible */
978 if (rc == -ENODATA || rc == -ENOENT)
982 /* EA doesn't fit, reallocate new buffer */
983 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
984 if (rc == -ENODATA || rc == -ENOENT)
990 rc = lod_ea_store_resize(info, rc);
1000 * Verify the target index is present in the current configuration.
1002 * \param[in] md LOD device where the target table is stored
1003 * \param[in] idx target's index
1005 * \retval 0 if the index is present
1006 * \retval -EINVAL if not
1008 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1010 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1011 !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
1012 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1013 md->lod_ost_descs.ltd_tgts_size);
1017 if (unlikely(OST_TGT(md, idx) == NULL)) {
1018 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1019 lod2obd(md)->obd_name, idx);
1023 if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
1024 CERROR("%s: invalid lod device, for idx: %d\n",
1025 lod2obd(md)->obd_name , idx);
1033 * Instantiate objects for stripes.
1035 * Allocate and initialize LU-objects representing the stripes. The number
1036 * of the stripes (ldo_stripe_count) must be initialized already. The caller
1037 * must ensure nobody else is calling the function on the object at the same
1038 * time. FLDB service must be running to be able to map a FID to the targets
1039 * and find appropriate device representing that target.
1041 * \param[in] env execution environment for this thread
1042 * \param[in,out] lo LOD object
1043 * \param[in] objs an array of IDs to creates the objects from
1044 * \param[in] comp_idx index of ldo_comp_entries
1046 * \retval 0 if the objects are instantiated successfully
1047 * \retval negative error number on failure
1049 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1050 struct lov_ost_data_v1 *objs, int comp_idx)
1052 struct lod_layout_component *lod_comp;
1053 struct lod_thread_info *info = lod_env_info(env);
1054 struct lod_device *md;
1055 struct lu_object *o, *n;
1056 struct lu_device *nd;
1057 struct dt_object **stripe = NULL;
1058 __u32 *ost_indices = NULL;
1064 LASSERT(lo != NULL);
1065 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1067 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1068 lod_comp = &lo->ldo_comp_entries[comp_idx];
1070 LASSERT(lod_comp->llc_stripe == NULL);
1071 LASSERT(lod_comp->llc_stripe_count > 0);
1072 LASSERT(lod_comp->llc_stripe_size > 0);
1074 stripe_len = lod_comp->llc_stripe_count;
1075 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1078 OBD_ALLOC(ost_indices, sizeof(*ost_indices) * stripe_len);
1080 GOTO(out, rc = -ENOMEM);
1082 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1083 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1086 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1087 idx = le32_to_cpu(objs[i].l_ost_idx);
1088 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1091 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1092 PFID(&info->lti_fid));
1093 lod_getref(&md->lod_ost_descs);
1095 rc = validate_lod_and_idx(md, idx);
1096 if (unlikely(rc != 0)) {
1097 lod_putref(md, &md->lod_ost_descs);
1101 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
1102 lod_putref(md, &md->lod_ost_descs);
1104 /* In the function below, .hs_keycmp resolves to
1105 * u_obj_hop_keycmp() */
1106 /* coverity[overrun-buffer-val] */
1107 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1109 GOTO(out, rc = PTR_ERR(o));
1111 n = lu_object_locate(o->lo_header, nd->ld_type);
1114 stripe[i] = container_of(n, struct dt_object, do_lu);
1115 ost_indices[i] = idx;
1120 for (i = 0; i < stripe_len; i++)
1121 if (stripe[i] != NULL)
1122 dt_object_put(env, stripe[i]);
1124 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1125 lod_comp->llc_stripe_count = 0;
1127 OBD_FREE(ost_indices,
1128 sizeof(*ost_indices) * stripe_len);
1130 lod_comp->llc_stripe = stripe;
1131 lod_comp->llc_ost_indices = ost_indices;
1132 lod_comp->llc_stripes_allocated = stripe_len;
1139 * Instantiate objects for striping.
1141 * Parse striping information in \a buf and instantiate the objects
1142 * representing the stripes.
1144 * \param[in] env execution environment for this thread
1145 * \param[in] lo LOD object
1146 * \param[in] buf buffer storing LOV EA to parse
1148 * \retval 0 if parsing and objects creation succeed
1149 * \retval negative error number on failure
1151 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1152 const struct lu_buf *buf)
1154 struct lov_mds_md_v1 *lmm;
1155 struct lov_comp_md_v1 *comp_v1 = NULL;
1156 struct lov_foreign_md *foreign = NULL;
1157 struct lov_ost_data_v1 *objs;
1158 __u32 magic, pattern;
1159 __u16 mirror_cnt = 0;
1165 LASSERT(buf->lb_buf);
1166 LASSERT(buf->lb_len);
1167 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1169 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1170 magic = le32_to_cpu(lmm->lmm_magic);
1172 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1173 magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN &&
1174 magic != LOV_MAGIC_SEL)
1175 GOTO(out, rc = -EINVAL);
1177 if (lo->ldo_is_foreign)
1178 lod_free_foreign_lov(lo);
1180 lod_free_comp_entries(lo);
1182 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1183 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1184 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1186 GOTO(out, rc = -EINVAL);
1187 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1188 lo->ldo_is_composite = 1;
1189 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1191 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1192 } else if (magic == LOV_MAGIC_FOREIGN) {
1195 foreign = (struct lov_foreign_md *)buf->lb_buf;
1196 length = offsetof(typeof(*foreign), lfm_value);
1197 if (buf->lb_len < length ||
1198 buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1200 "buf len %zu too small for lov_foreign_md\n",
1202 GOTO(out, rc = -EINVAL);
1205 /* just cache foreign LOV EA raw */
1206 rc = lod_alloc_foreign_lov(lo, length);
1209 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1213 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1214 lo->ldo_is_composite = 0;
1217 rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1221 for (i = 0; i < comp_cnt; i++) {
1222 struct lod_layout_component *lod_comp;
1223 struct lu_extent *ext;
1226 lod_comp = &lo->ldo_comp_entries[i];
1227 if (lo->ldo_is_composite) {
1228 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1229 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1231 ext = &comp_v1->lcm_entries[i].lcme_extent;
1232 lod_comp->llc_extent.e_start =
1233 le64_to_cpu(ext->e_start);
1234 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1235 lod_comp->llc_flags =
1236 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1237 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1238 lod_comp->llc_timestamp = le64_to_cpu(
1239 comp_v1->lcm_entries[i].lcme_timestamp);
1241 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1242 if (lod_comp->llc_id == LCME_ID_INVAL)
1243 GOTO(out, rc = -EINVAL);
1245 if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
1246 comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
1247 struct lod_device *d =
1248 lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1250 CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
1251 lod2obd(d)->obd_name,
1252 lod_comp->llc_flags, lod_comp->llc_id, i,
1253 PFID(lod_object_fid(lo)),
1254 le32_to_cpu(comp_v1->lcm_magic));
1257 lod_comp_set_init(lod_comp);
1260 pattern = le32_to_cpu(lmm->lmm_pattern);
1261 if (!lov_pattern_supported(lov_pattern(pattern)))
1262 GOTO(out, rc = -EINVAL);
1264 lod_comp->llc_pattern = pattern;
1265 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1266 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1267 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1269 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1270 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1272 lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1273 objs = &v3->lmm_objects[0];
1275 lod_set_pool(&lod_comp->llc_pool, NULL);
1276 objs = &lmm->lmm_objects[0];
1280 * If uninstantiated template component has valid l_ost_idx,
1281 * then user has specified ost list for this component.
1283 if (!lod_comp_inited(lod_comp)) {
1286 if (objs[0].l_ost_idx != (__u32)-1UL) {
1289 stripe_count = lod_comp_entry_stripe_count(
1290 lo, lod_comp, false);
1291 if (stripe_count == 0 &&
1292 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1293 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1294 GOTO(out, rc = -E2BIG);
1296 * load the user specified ost list, when this
1297 * component is instantiated later, it will be
1298 * used in lod_alloc_ost_list().
1300 lod_comp->llc_ostlist.op_count = stripe_count;
1301 lod_comp->llc_ostlist.op_size =
1302 stripe_count * sizeof(__u32);
1303 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1304 lod_comp->llc_ostlist.op_size);
1305 if (!lod_comp->llc_ostlist.op_array)
1306 GOTO(out, rc = -ENOMEM);
1308 for (j = 0; j < stripe_count; j++)
1309 lod_comp->llc_ostlist.op_array[j] =
1310 le32_to_cpu(objs[j].l_ost_idx);
1313 * this component OST objects starts from the
1314 * first ost_idx, lod_alloc_ost_list() will
1317 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1320 * for uninstantiated component,
1321 * lmm_layout_gen stores default stripe offset.
1323 lod_comp->llc_stripe_offset =
1324 lmm->lmm_layout_gen;
1328 /* skip un-instantiated component object initialization */
1329 if (!lod_comp_inited(lod_comp))
1332 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1333 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1334 rc = lod_initialize_objects(env, lo, objs, i);
1340 rc = lod_fill_mirrors(lo);
1346 lod_striping_free_nolock(env, lo);
1351 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1352 * is already cached.
1354 * \param[in] lo LOD object
1356 * \retval True if the striping is cached, otherwise
1359 static bool lod_striping_loaded(struct lod_object *lo)
1361 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1362 lo->ldo_comp_cached)
1365 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1366 if (lo->ldo_dir_stripe_loaded)
1369 /* Never load LMV stripe for slaves of striped dir */
1370 if (lo->ldo_dir_slave_stripe)
1378 * A generic function to initialize the stripe objects.
1380 * A protected version of lod_striping_load_locked() - load the striping
1381 * information from storage, parse that and instantiate LU objects to
1382 * represent the stripes. The LOD object \a lo supplies a pointer to the
1383 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1384 * return an array of references to the newly instantiated objects.
1386 * \param[in] env execution environment for this thread
1387 * \param[in,out] lo LOD object, where striping is stored and
1388 * which gets an array of references
1390 * \retval 0 if parsing and object creation succeed
1391 * \retval negative error number on failure
1393 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1395 struct lod_thread_info *info = lod_env_info(env);
1396 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1397 struct lu_buf *buf = &info->lti_buf;
1402 if (!dt_object_exists(next))
1405 if (lod_striping_loaded(lo))
1408 mutex_lock(&lo->ldo_layout_mutex);
1409 if (lod_striping_loaded(lo))
1410 GOTO(unlock, rc = 0);
1412 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1413 rc = lod_get_lov_ea(env, lo);
1418 * there is LOV EA (striping information) in this object
1419 * let's parse it and create in-core objects for the stripes
1421 buf->lb_buf = info->lti_ea_store;
1422 buf->lb_len = info->lti_ea_store_size;
1423 rc = lod_parse_striping(env, lo, buf);
1425 lo->ldo_comp_cached = 1;
1426 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1427 rc = lod_get_lmv_ea(env, lo);
1428 if (rc > sizeof(struct lmv_foreign_md)) {
1429 struct lmv_foreign_md *lfm = info->lti_ea_store;
1431 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
1432 lo->ldo_foreign_lmv = info->lti_ea_store;
1433 lo->ldo_foreign_lmv_size =
1434 info->lti_ea_store_size;
1435 info->lti_ea_store = NULL;
1436 info->lti_ea_store_size = 0;
1438 lo->ldo_dir_stripe_loaded = 1;
1439 lo->ldo_dir_is_foreign = 1;
1440 GOTO(unlock, rc = 0);
1444 if (rc < (int)sizeof(struct lmv_mds_md_v1)) {
1445 /* Let's set stripe_loaded to avoid further
1446 * stripe loading especially for non-stripe directory,
1447 * which can hurt performance. (See LU-9840)
1450 lo->ldo_dir_stripe_loaded = 1;
1451 GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1453 buf->lb_buf = info->lti_ea_store;
1454 buf->lb_len = info->lti_ea_store_size;
1455 if (rc == sizeof(struct lmv_mds_md_v1)) {
1456 rc = lod_load_lmv_shards(env, lo, buf, true);
1457 if (buf->lb_buf != info->lti_ea_store) {
1458 OBD_FREE_LARGE(info->lti_ea_store,
1459 info->lti_ea_store_size);
1460 info->lti_ea_store = buf->lb_buf;
1461 info->lti_ea_store_size = buf->lb_len;
1469 * there is LMV EA (striping information) in this object
1470 * let's parse it and create in-core objects for the stripes
1472 rc = lod_parse_dir_striping(env, lo, buf);
1474 lo->ldo_dir_stripe_loaded = 1;
1478 mutex_unlock(&lo->ldo_layout_mutex);
1483 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1484 const struct lu_buf *buf)
1490 mutex_lock(&lo->ldo_layout_mutex);
1491 lod_striping_free_nolock(env, lo);
1492 rc = lod_parse_striping(env, lo, buf);
1493 mutex_unlock(&lo->ldo_layout_mutex);
1499 * Verify lov_user_md_v1/v3 striping.
1501 * Check the validity of all fields including the magic, stripe size,
1502 * stripe count, stripe offset and that the pool is present. Also check
1503 * that each target index points to an existing target. The additional
1504 * \a is_from_disk turns additional checks. In some cases zero fields
1505 * are allowed (like pattern=0).
1507 * \param[in] d LOD device
1508 * \param[in] buf buffer with LOV EA to verify
1509 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1510 * 1 - from disk, do not allow
1512 * \retval 0 if the striping is valid
1513 * \retval -EINVAL if striping is invalid
1515 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1518 struct lov_user_md_v1 *lum;
1519 struct lov_user_md_v3 *lum3;
1520 struct pool_desc *pool = NULL;
1524 __u16 stripe_offset;
1531 if (buf->lb_len < sizeof(*lum)) {
1532 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1534 GOTO(out, rc = -EINVAL);
1537 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1538 if (magic != LOV_USER_MAGIC_V1 &&
1539 magic != LOV_USER_MAGIC_V3 &&
1540 magic != LOV_USER_MAGIC_SPECIFIC) {
1541 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1542 le32_to_cpu(lum->lmm_magic));
1543 GOTO(out, rc = -EINVAL);
1546 /* the user uses "0" for default stripe pattern normally. */
1547 if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1548 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1550 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1551 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1552 le32_to_cpu(lum->lmm_pattern));
1553 GOTO(out, rc = -EINVAL);
1556 /* a released lum comes from creating orphan on hsm release,
1557 * doesn't make sense to verify it. */
1558 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1561 /* 64kB is the largest common page size we see (ia64), and matches the
1563 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1564 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1565 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1566 stripe_size, LOV_MIN_STRIPE_SIZE);
1567 GOTO(out, rc = -EINVAL);
1570 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1571 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1572 lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1573 /* if offset is not within valid range [0, osts_size) */
1574 if (stripe_offset >= d->lod_osts_size) {
1575 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1576 stripe_offset, d->lod_osts_size);
1577 GOTO(out, rc = -EINVAL);
1580 /* if lmm_stripe_offset is *not* in bitmap */
1581 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1582 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1584 GOTO(out, rc = -EINVAL);
1588 if (magic == LOV_USER_MAGIC_V1)
1589 lum_size = offsetof(struct lov_user_md_v1,
1591 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1592 lum_size = offsetof(struct lov_user_md_v3,
1595 GOTO(out, rc = -EINVAL);
1597 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1598 if (buf->lb_len < lum_size) {
1599 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1600 "magic %#x and stripe_count %u\n",
1601 buf->lb_len, lum_size, magic, stripe_count);
1602 GOTO(out, rc = -EINVAL);
1605 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1609 /* In the function below, .hs_keycmp resolves to
1610 * pool_hashkey_keycmp() */
1611 /* coverity[overrun-buffer-val] */
1612 pool = lod_find_pool(d, lum3->lmm_pool_name);
1616 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1617 rc = lod_check_index_in_pool(stripe_offset, pool);
1619 GOTO(out, rc = -EINVAL);
1622 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1623 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1624 stripe_count, pool_tgt_count(pool));
1625 GOTO(out, rc = -EINVAL);
1630 lod_pool_putref(pool);
1636 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1638 LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1639 LOV_USER_MAGIC_COMP_V1, "Wrong magic %x\n",
1640 le32_to_cpu(comp->lcm_magic));
1641 LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1642 "bad index %d, max = %d\n",
1643 i, le16_to_cpu(comp->lcm_entry_count));
1645 return &comp->lcm_entries[i];
1648 #define for_each_comp_entry_v1(comp, entry) \
1649 for (entry = comp_entry_v1(comp, 0); \
1650 entry <= comp_entry_v1(comp, \
1651 le16_to_cpu(comp->lcm_entry_count) - 1); \
1654 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
1655 struct lov_comp_md_entry_v1 *dom_ent)
1657 struct lov_comp_md_entry_v1 *ent;
1659 __u32 dom_off, dom_size, comp_size;
1660 void *blob_src, *blob_dst;
1661 unsigned int blob_size, blob_shift;
1663 entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1664 /* if file has only DoM stripe return just error */
1668 comp_size = le32_to_cpu(comp_v1->lcm_size);
1669 dom_off = le32_to_cpu(dom_ent->lcme_offset);
1670 dom_size = le32_to_cpu(dom_ent->lcme_size);
1672 /* shift entries array first */
1673 comp_v1->lcm_entry_count = cpu_to_le16(entries);
1674 memmove(dom_ent, dom_ent + 1,
1675 entries * sizeof(struct lov_comp_md_entry_v1));
1677 /* now move blob of layouts */
1678 blob_dst = (void *)comp_v1 + dom_off - sizeof(*dom_ent);
1679 blob_src = (void *)comp_v1 + dom_off + dom_size;
1680 blob_size = (unsigned long)((void *)comp_v1 + comp_size - blob_src);
1681 blob_shift = sizeof(*dom_ent) + dom_size;
1683 memmove(blob_dst, blob_src, blob_size);
1685 for_each_comp_entry_v1(comp_v1, ent) {
1688 off = le32_to_cpu(ent->lcme_offset);
1689 ent->lcme_offset = cpu_to_le32(off - blob_shift);
1692 comp_v1->lcm_size = cpu_to_le32(comp_size - blob_shift);
1694 /* notify a caller to re-check entry */
1698 int lod_fix_dom_stripe(struct lod_device *d, struct lov_comp_md_v1 *comp_v1,
1699 struct lov_comp_md_entry_v1 *dom_ent)
1701 struct lov_comp_md_entry_v1 *ent;
1702 struct lu_extent *dom_ext, *ext;
1703 struct lov_user_md_v1 *lum;
1708 dom_ext = &dom_ent->lcme_extent;
1709 dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1710 stripe_size = d->lod_dom_max_stripesize;
1712 lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1713 CDEBUG(D_LAYOUT, "DoM component size %u was bigger than MDT limit %u, "
1714 "new size is %u\n", le32_to_cpu(lum->lmm_stripe_size),
1715 d->lod_dom_max_stripesize, stripe_size);
1716 lum->lmm_stripe_size = cpu_to_le32(stripe_size);
1718 for_each_comp_entry_v1(comp_v1, ent) {
1722 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1726 ext = &ent->lcme_extent;
1727 if (ext->e_start != dom_ext->e_end)
1730 /* Found next component after the DoM one with the same
1731 * mirror_id and adjust its start with DoM component end.
1733 * NOTE: we are considering here that there can be only one
1734 * DoM component in a file, all replicas are located on OSTs
1735 * always and don't need adjustment since use own layouts.
1737 ext->e_start = cpu_to_le64(stripe_size);
1741 if (stripe_size == 0) {
1742 /* DoM component size is zero due to server setting,
1743 * remove it from the layout */
1744 rc = lod_erase_dom_stripe(comp_v1, dom_ent);
1746 /* Update DoM extent end finally */
1747 dom_ext->e_end = cpu_to_le64(stripe_size);
1754 * Verify LOV striping.
1756 * \param[in] d LOD device
1757 * \param[in] buf buffer with LOV EA to verify
1758 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1759 * 1 - from disk, do not allow
1760 * \param[in] start extent start for composite layout
1762 * \retval 0 if the striping is valid
1763 * \retval -EINVAL if striping is invalid
1765 int lod_verify_striping(struct lod_device *d, struct lod_object *lo,
1766 const struct lu_buf *buf, bool is_from_disk)
1768 struct lov_desc *desc = &d->lod_desc;
1769 struct lov_user_md_v1 *lum;
1770 struct lov_comp_md_v1 *comp_v1;
1771 struct lov_comp_md_entry_v1 *ent;
1772 struct lu_extent *ext;
1775 __u32 stripe_size = 0;
1776 __u16 prev_mid = -1, mirror_id = -1;
1782 if (buf->lb_len < sizeof(lum->lmm_magic)) {
1783 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
1789 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1790 /* treat foreign LOV EA/object case first
1791 * XXX is it expected to try setting again a foreign?
1792 * XXX should we care about different current vs new layouts ?
1794 if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
1795 struct lov_foreign_md *lfm = buf->lb_buf;
1797 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
1799 "buf len %zu < min lov_foreign_md size (%zu)\n",
1800 buf->lb_len, offsetof(typeof(*lfm),
1805 if (foreign_size_le(lfm) > buf->lb_len) {
1807 "buf len %zu < this lov_foreign_md size (%zu)\n",
1808 buf->lb_len, foreign_size_le(lfm));
1811 /* Don't do anything with foreign layouts */
1815 /* normal LOV/layout cases */
1817 if (buf->lb_len < sizeof(*lum)) {
1818 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1823 if (magic != LOV_USER_MAGIC_V1 &&
1824 magic != LOV_USER_MAGIC_V3 &&
1825 magic != LOV_USER_MAGIC_SPECIFIC &&
1826 magic != LOV_USER_MAGIC_COMP_V1) {
1827 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1828 le32_to_cpu(lum->lmm_magic));
1832 if (magic != LOV_USER_MAGIC_COMP_V1)
1833 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1835 /* magic == LOV_USER_MAGIC_COMP_V1 */
1836 comp_v1 = buf->lb_buf;
1837 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1838 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1839 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1845 if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1846 CDEBUG(D_LAYOUT, "entry count is zero\n");
1850 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1851 lo->ldo_comp_cnt > 0) {
1852 /* could be called from lustre.lov.add */
1853 __u32 cnt = lo->ldo_comp_cnt;
1855 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
1856 prev_end = ext->e_end;
1861 for_each_comp_entry_v1(comp_v1, ent) {
1862 ext = &ent->lcme_extent;
1864 if (le64_to_cpu(ext->e_start) > le64_to_cpu(ext->e_end)) {
1865 CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
1866 le64_to_cpu(ext->e_start),
1867 le64_to_cpu(ext->e_end));
1872 /* lcme_id contains valid value */
1873 if (le32_to_cpu(ent->lcme_id) == 0 ||
1874 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
1875 CDEBUG(D_LAYOUT, "invalid id %u\n",
1876 le32_to_cpu(ent->lcme_id));
1880 if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
1881 mirror_id = mirror_id_of(
1882 le32_to_cpu(ent->lcme_id));
1884 /* first component must start with 0 */
1885 if (mirror_id != prev_mid &&
1886 le64_to_cpu(ext->e_start) != 0) {
1888 "invalid start:%llu, expect:0\n",
1889 le64_to_cpu(ext->e_start));
1893 prev_mid = mirror_id;
1897 if (le64_to_cpu(ext->e_start) == 0) {
1902 /* the next must be adjacent with the previous one */
1903 if (le64_to_cpu(ext->e_start) != prev_end) {
1905 "invalid start actual:%llu, expect:%llu\n",
1906 le64_to_cpu(ext->e_start), prev_end);
1910 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
1911 tmp.lb_len = le32_to_cpu(ent->lcme_size);
1913 /* Check DoM entry is always the first one */
1915 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
1917 /* DoM component must be the first in a mirror */
1918 if (le64_to_cpu(ext->e_start) > 0) {
1919 CDEBUG(D_LAYOUT, "invalid DoM component "
1920 "with %llu extent start\n",
1921 le64_to_cpu(ext->e_start));
1924 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1925 /* There is just one stripe on MDT and it must
1926 * cover whole component size. */
1927 if (stripe_size != le64_to_cpu(ext->e_end)) {
1928 CDEBUG(D_LAYOUT, "invalid DoM layout "
1929 "stripe size %u != %llu "
1930 "(component size)\n",
1931 stripe_size, prev_end);
1934 /* Check stripe size againts per-MDT limit */
1935 if (stripe_size > d->lod_dom_max_stripesize) {
1936 CDEBUG(D_LAYOUT, "DoM component size "
1937 "%u is bigger than MDT limit %u, check "
1938 "dom_max_stripesize parameter\n",
1939 stripe_size, d->lod_dom_max_stripesize);
1940 rc = lod_fix_dom_stripe(d, comp_v1, ent);
1941 if (rc == -ERESTART) {
1942 /* DoM entry was removed, re-check
1943 * new layout from start */
1951 prev_end = le64_to_cpu(ext->e_end);
1953 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
1957 if (prev_end == LUSTRE_EOF)
1960 /* extent end must be aligned with the stripe_size */
1961 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1962 if (stripe_size == 0)
1963 stripe_size = desc->ld_default_stripe_size;
1964 if (prev_end % stripe_size) {
1965 CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
1966 "stripe_sz: %u, [%llu, %llu)\n",
1967 stripe_size, ext->e_start, prev_end);
1972 /* make sure that the mirror_count is telling the truth */
1973 if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
1980 * set the default stripe size, if unset.
1982 * \param[in,out] val number of bytes per OST stripe
1984 * The minimum stripe size is 64KB to ensure that a single stripe is an
1985 * even multiple of a client PAGE_SIZE (IA64, PPC, etc). Otherwise, it
1986 * is difficult to split dirty pages across OSCs during writes.
1988 void lod_fix_desc_stripe_size(__u64 *val)
1990 if (*val < LOV_MIN_STRIPE_SIZE) {
1992 LCONSOLE_INFO("Increasing default stripe size to "
1993 "minimum value %u\n",
1994 LOV_DESC_STRIPE_SIZE_DEFAULT);
1995 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1996 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1997 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1998 LCONSOLE_WARN("Changing default stripe size to %llu (a "
1999 "multiple of %u)\n",
2000 *val, LOV_MIN_STRIPE_SIZE);
2005 * set the filesystem default number of stripes, if unset.
2007 * \param[in,out] val number of stripes
2009 * A value of "0" means "use the system-wide default stripe count", which
2010 * has either been inherited by now, or falls back to 1 stripe per file.
2011 * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
2012 * and is a valid value, so is left unchanged here.
2014 void lod_fix_desc_stripe_count(__u32 *val)
2021 * set the filesystem default layout pattern
2023 * \param[in,out] val LOV_PATTERN_* layout
2025 * A value of "0" means "use the system-wide default layout type", which
2026 * has either been inherited by now, or falls back to plain RAID0 striping.
2028 void lod_fix_desc_pattern(__u32 *val)
2030 /* from lov_setstripe */
2031 if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
2032 LCONSOLE_WARN("lod: Unknown stripe pattern: %#x\n", *val);
2037 void lod_fix_desc_qos_maxage(__u32 *val)
2039 /* fix qos_maxage */
2041 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2045 * Used to fix insane default striping.
2047 * \param[in] desc striping description
2049 void lod_fix_desc(struct lov_desc *desc)
2051 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2052 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2053 lod_fix_desc_pattern(&desc->ld_pattern);
2054 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2058 * Initialize the structures used to store pools and default striping.
2060 * \param[in] lod LOD device
2061 * \param[in] lcfg configuration structure storing default striping.
2063 * \retval 0 if initialization succeeds
2064 * \retval negative error number on failure
2066 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2068 struct obd_device *obd;
2069 struct lov_desc *desc;
2073 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2074 LASSERT(obd != NULL);
2075 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2077 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2078 CERROR("LOD setup requires a descriptor\n");
2082 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2084 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2085 CERROR("descriptor size wrong: %d > %d\n",
2086 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2090 if (desc->ld_magic != LOV_DESC_MAGIC) {
2091 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2092 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2093 obd->obd_name, desc);
2094 lustre_swab_lov_desc(desc);
2096 CERROR("%s: Bad lov desc magic: %#x\n",
2097 obd->obd_name, desc->ld_magic);
2104 desc->ld_active_tgt_count = 0;
2105 lod->lod_desc = *desc;
2107 lod->lod_sp_me = LUSTRE_SP_CLI;
2109 /* Set up allocation policy (QoS and RR) */
2110 INIT_LIST_HEAD(&lod->lod_qos.lq_svr_list);
2111 init_rwsem(&lod->lod_qos.lq_rw_sem);
2112 lod->lod_qos.lq_dirty = 1;
2113 lod->lod_qos.lq_reset = 1;
2114 /* Default priority is toward free space balance */
2115 lod->lod_qos.lq_prio_free = 232;
2116 /* Default threshold for rr (roughly 17%) */
2117 lod->lod_qos.lq_threshold_rr = 43;
2119 lu_qos_rr_init(&lod->lod_qos.lq_rr);
2121 /* Set up OST pool environment */
2122 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
2123 HASH_POOLS_MAX_BITS,
2124 HASH_POOLS_BKT_BITS, 0,
2127 &pool_hash_operations,
2129 if (lod->lod_pools_hash_body == NULL)
2132 INIT_LIST_HEAD(&lod->lod_pool_list);
2133 lod->lod_pool_count = 0;
2134 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
2137 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
2139 GOTO(out_pool_info, rc);
2144 lod_ost_pool_free(&lod->lod_pool_info);
2146 cfs_hash_putref(lod->lod_pools_hash_body);
2152 * Release the structures describing the pools.
2154 * \param[in] lod LOD device from which we release the structures
2158 int lod_pools_fini(struct lod_device *lod)
2160 struct obd_device *obd = lod2obd(lod);
2161 struct pool_desc *pool, *tmp;
2164 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2165 /* free pool structs */
2166 CDEBUG(D_INFO, "delete pool %p\n", pool);
2167 /* In the function below, .hs_keycmp resolves to
2168 * pool_hashkey_keycmp() */
2169 /* coverity[overrun-buffer-val] */
2170 lod_pool_del(obd, pool->pool_name);
2173 cfs_hash_putref(lod->lod_pools_hash_body);
2174 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
2175 lod_ost_pool_free(&lod->lod_pool_info);