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;
81 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
84 ltd_foreach_tgt_safe(ltd, tgt_desc, tmp) {
86 if (!tgt_desc->ltd_reap)
89 list_add(&tgt_desc->ltd_kill, &kill);
90 lod_tgt_pool_remove(<d->ltd_tgt_pool,
92 ltd_del_tgt(ltd, tgt_desc);
95 mutex_unlock(<d->ltd_mutex);
96 up_read(<d->ltd_rw_sem);
98 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
101 list_del(&tgt_desc->ltd_kill);
102 rc = obd_disconnect(tgt_desc->ltd_exp);
104 CERROR("%s: failed to disconnect %s: rc = %d\n",
105 lod2obd(lod)->obd_name,
106 obd_uuid2str(&tgt_desc->ltd_uuid), rc);
107 OBD_FREE_PTR(tgt_desc);
110 mutex_unlock(<d->ltd_mutex);
111 up_read(<d->ltd_rw_sem);
116 * Connect LOD to a new OSP and add it to the target table.
118 * Connect to the OSP device passed, initialize all the internal
119 * structures related to the device and add it to the target table.
121 * \param[in] env execution environment for this thread
122 * \param[in] lod LOD device to be connected to the new OSP
123 * \param[in] osp name of OSP device name to be added
124 * \param[in] index index of the new target
125 * \param[in] gen target's generation number
126 * \param[in] tgt_index OSP's group
127 * \param[in] type type of device (mdc or osc)
128 * \param[in] active state of OSP: 0 - inactive, 1 - active
130 * \retval 0 if added successfully
131 * \retval negative error number on failure
133 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
134 char *osp, unsigned index, unsigned gen, int tgt_index,
135 char *type, int active)
137 struct obd_connect_data *data = NULL;
138 struct obd_export *exp = NULL;
139 struct obd_device *obd;
140 struct lu_device *lu_dev;
141 struct dt_device *dt_dev;
143 struct lod_tgt_desc *tgt_desc;
144 struct lod_tgt_descs *ltd;
145 struct lustre_cfg *lcfg;
146 struct obd_uuid obd_uuid;
148 bool connected = false;
151 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
154 CERROR("request to add OBD %s with invalid generation: %d\n",
159 obd_str2uuid(&obd_uuid, osp);
161 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
162 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
164 CERROR("can't find %s device\n", osp);
168 LASSERT(obd->obd_lu_dev != NULL);
169 LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
171 lu_dev = obd->obd_lu_dev;
172 dt_dev = lu2dt_dev(lu_dev);
176 GOTO(out_cleanup, rc = -ENOMEM);
178 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
179 data->ocd_version = LUSTRE_VERSION_CODE;
180 data->ocd_index = index;
182 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
184 data->ocd_connect_flags |= OBD_CONNECT_AT |
187 #ifdef HAVE_LRU_RESIZE_SUPPORT
188 OBD_CONNECT_LRU_RESIZE |
191 OBD_CONNECT_REQPORTAL |
192 OBD_CONNECT_SKIP_ORPHAN |
194 OBD_CONNECT_LVB_TYPE |
195 OBD_CONNECT_VERSION |
196 OBD_CONNECT_PINGLESS |
198 OBD_CONNECT_BULK_MBITS;
200 data->ocd_group = tgt_index;
201 ltd = &lod->lod_ost_descs;
203 struct obd_import *imp = obd->u.cli.cl_import;
206 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
207 data->ocd_connect_flags |= OBD_CONNECT_ACL |
209 OBD_CONNECT_MDS_MDS |
214 OBD_CONNECT_BULK_MBITS;
215 spin_lock(&imp->imp_lock);
216 imp->imp_server_timeout = 1;
217 spin_unlock(&imp->imp_lock);
218 imp->imp_client->cli_request_portal = OUT_PORTAL;
219 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
221 ltd = &lod->lod_mdt_descs;
224 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
227 CERROR("%s: cannot connect to next dev %s (%d)\n",
228 obd->obd_name, osp, rc);
229 GOTO(out_cleanup, rc);
233 /* Allocate ost descriptor and fill it */
234 OBD_ALLOC_PTR(tgt_desc);
236 GOTO(out_cleanup, rc = -ENOMEM);
238 tgt_desc->ltd_tgt = dt_dev;
239 tgt_desc->ltd_exp = exp;
240 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
241 tgt_desc->ltd_gen = gen;
242 tgt_desc->ltd_index = index;
243 tgt_desc->ltd_active = active;
245 down_write(<d->ltd_rw_sem);
246 mutex_lock(<d->ltd_mutex);
247 rc = ltd_add_tgt(ltd, tgt_desc);
251 rc = lu_qos_add_tgt(<d->ltd_qos, tgt_desc);
253 GOTO(out_del_tgt, rc);
255 rc = lod_tgt_pool_add(<d->ltd_tgt_pool, index,
256 ltd->ltd_lov_desc.ld_tgt_count);
258 CERROR("%s: can't set up pool, failed with %d\n",
260 GOTO(out_del_tgt, rc);
263 mutex_unlock(<d->ltd_mutex);
264 up_write(<d->ltd_rw_sem);
266 if (lod->lod_recovery_completed)
267 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
269 if (!for_ost && lod->lod_initialized) {
270 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
272 CERROR("%s: cannot start llog on %s:rc = %d\n",
273 lod2obd(lod)->obd_name, osp, rc);
278 rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
280 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
281 "index = %u, rc = %d\n", osp, type, index, rc);
282 GOTO(out_fini_llog, rc);
286 lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
287 tgt_desc->ltd_recovery_thread);
289 down_write(<d->ltd_rw_sem);
290 mutex_lock(<d->ltd_mutex);
291 if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
292 struct ptlrpc_thread *thread;
294 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
295 OBD_FREE_PTR(thread);
297 lod_tgt_pool_remove(<d->ltd_tgt_pool, index);
299 ltd_del_tgt(ltd, tgt_desc);
301 mutex_unlock(<d->ltd_mutex);
302 up_write(<d->ltd_rw_sem);
303 OBD_FREE_PTR(tgt_desc);
305 /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
306 * objects from the MDT stack. See LU-7184. */
307 lcfg = &lod_env_info(env)->lti_lustre_cfg;
308 memset(lcfg, 0, sizeof(*lcfg));
309 lcfg->lcfg_version = LUSTRE_CFG_VERSION;
310 lcfg->lcfg_command = LCFG_CLEANUP;
311 lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
320 * Schedule target removal from the target table.
322 * Mark the device as dead. The device is not removed here because it may
323 * still be in use. The device will be removed in lod_putref() when the
324 * last reference is released.
326 * \param[in] env execution environment for this thread
327 * \param[in] lod LOD device the target table belongs to
328 * \param[in] ltd target table
329 * \param[in] tgt target
331 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
332 struct lod_tgt_descs *ltd, struct lu_tgt_desc *tgt)
334 lfsck_del_target(env, lod->lod_child, tgt->ltd_tgt, tgt->ltd_index,
337 if (ltd->ltd_is_mdt && tgt->ltd_recovery_thread)
338 OBD_FREE_PTR(tgt->ltd_recovery_thread);
340 if (!tgt->ltd_reap) {
342 ltd->ltd_death_row++;
347 * Schedule removal of all the targets from the given target table.
349 * See more details in the description for __lod_del_device()
351 * \param[in] env execution environment for this thread
352 * \param[in] lod LOD device the target table belongs to
353 * \param[in] ltd target table
357 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
358 struct lod_tgt_descs *ltd)
360 struct lu_tgt_desc *tgt;
362 if (ltd->ltd_tgts_size <= 0)
366 mutex_lock(<d->ltd_mutex);
367 ltd_foreach_tgt(ltd, tgt)
368 __lod_del_device(env, lod, ltd, tgt);
369 mutex_unlock(<d->ltd_mutex);
370 lod_putref(lod, ltd);
372 lu_tgt_descs_fini(ltd);
378 * Remove device by name.
380 * Remove a device identified by \a osp from the target table. Given
381 * the device can be in use, the real deletion happens in lod_putref().
383 * \param[in] env execution environment for this thread
384 * \param[in] lod LOD device to be connected to the new OSP
385 * \param[in] ltd target table
386 * \param[in] osp name of OSP device to be removed
387 * \param[in] idx index of the target
388 * \param[in] gen generation number, not used currently
390 * \retval 0 if the device was scheduled for removal
391 * \retval -EINVAL if no device was found
393 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
394 struct lod_tgt_descs *ltd, char *osp, unsigned int idx,
397 struct obd_device *obd;
398 struct lu_tgt_desc *tgt;
399 struct obd_uuid uuid;
404 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
406 obd_str2uuid(&uuid, osp);
408 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
409 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
411 CERROR("can't find %s device\n", osp);
416 CERROR("%s: request to remove OBD %s with invalid generation %d"
417 "\n", obd->obd_name, osp, gen);
421 obd_str2uuid(&uuid, osp);
424 mutex_lock(<d->ltd_mutex);
425 tgt = LTD_TGT(ltd, idx);
426 /* check that the index is allocated in the bitmap */
427 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) || !tgt) {
428 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
429 GOTO(out, rc = -EINVAL);
432 /* check that the UUID matches */
433 if (!obd_uuid_equals(&uuid, &tgt->ltd_uuid)) {
434 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
435 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), idx, osp);
436 GOTO(out, rc = -EINVAL);
439 __lod_del_device(env, lod, ltd, tgt);
442 mutex_unlock(<d->ltd_mutex);
443 lod_putref(lod, ltd);
448 * Resize per-thread storage to hold specified size.
450 * A helper function to resize per-thread temporary storage. This storage
451 * is used to process LOV/LVM EAs and may be quite large. We do not want to
452 * allocate/release it every time, so instead we put it into the env and
453 * reallocate on demand. The memory is released when the correspondent thread
456 * \param[in] info LOD-specific storage in the environment
457 * \param[in] size new size to grow the buffer to
459 * \retval 0 on success, -ENOMEM if reallocation failed
461 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
463 __u32 round = size_roundup_power2(size);
465 if (info->lti_ea_store) {
466 LASSERT(info->lti_ea_store_size);
467 LASSERT(info->lti_ea_store_size < round);
468 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
469 info->lti_ea_store_size, round);
470 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
471 info->lti_ea_store = NULL;
472 info->lti_ea_store_size = 0;
475 OBD_ALLOC_LARGE(info->lti_ea_store, round);
476 if (info->lti_ea_store == NULL)
478 info->lti_ea_store_size = round;
483 static void lod_free_comp_buffer(struct lod_layout_component *entries,
484 __u16 count, __u32 bufsize)
486 struct lod_layout_component *entry;
489 for (i = 0; i < count; i++) {
491 if (entry->llc_pool != NULL)
492 lod_set_pool(&entry->llc_pool, NULL);
493 if (entry->llc_ostlist.op_array)
494 OBD_FREE(entry->llc_ostlist.op_array,
495 entry->llc_ostlist.op_size);
496 LASSERT(entry->llc_stripe == NULL);
497 LASSERT(entry->llc_stripes_allocated == 0);
501 OBD_FREE_LARGE(entries, bufsize);
504 void lod_free_def_comp_entries(struct lod_default_striping *lds)
506 lod_free_comp_buffer(lds->lds_def_comp_entries,
507 lds->lds_def_comp_size_cnt,
509 sizeof(*lds->lds_def_comp_entries) *
510 lds->lds_def_comp_size_cnt));
511 lds->lds_def_comp_entries = NULL;
512 lds->lds_def_comp_cnt = 0;
513 lds->lds_def_striping_is_composite = 0;
514 lds->lds_def_comp_size_cnt = 0;
518 * Resize per-thread storage to hold default striping component entries
520 * A helper function to resize per-thread temporary storage. This storage
521 * is used to hold default LOV/LVM EAs and may be quite large. We do not want
522 * to allocate/release it every time, so instead we put it into the env and
523 * reallocate it on demand. The memory is released when the correspondent
524 * thread is finished.
526 * \param[in,out] lds default striping
527 * \param[in] count new component count to grow the buffer to
529 * \retval 0 on success, -ENOMEM if reallocation failed
531 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
533 struct lod_layout_component *entries;
534 __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
536 __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
537 lds->lds_def_comp_size_cnt);
542 OBD_ALLOC_LARGE(entries, new);
546 if (lds->lds_def_comp_entries != NULL) {
547 CDEBUG(D_INFO, "default striping component size %d is not "
548 "enough, need %d\n", old, new);
549 lod_free_def_comp_entries(lds);
552 lds->lds_def_comp_entries = entries;
553 lds->lds_def_comp_size_cnt = count;
558 void lod_free_comp_entries(struct lod_object *lo)
560 if (lo->ldo_mirrors) {
561 OBD_FREE(lo->ldo_mirrors,
562 sizeof(*lo->ldo_mirrors) * lo->ldo_mirror_count);
563 lo->ldo_mirrors = NULL;
564 lo->ldo_mirror_count = 0;
566 lod_free_comp_buffer(lo->ldo_comp_entries,
568 sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
569 lo->ldo_comp_entries = NULL;
570 lo->ldo_comp_cnt = 0;
571 lo->ldo_is_composite = 0;
574 int lod_alloc_comp_entries(struct lod_object *lo,
575 int mirror_count, int comp_count)
577 LASSERT(comp_count != 0);
578 LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
580 if (mirror_count > 0) {
581 OBD_ALLOC(lo->ldo_mirrors,
582 sizeof(*lo->ldo_mirrors) * mirror_count);
583 if (!lo->ldo_mirrors)
586 lo->ldo_mirror_count = mirror_count;
589 OBD_ALLOC_LARGE(lo->ldo_comp_entries,
590 sizeof(*lo->ldo_comp_entries) * comp_count);
591 if (lo->ldo_comp_entries == NULL) {
592 OBD_FREE(lo->ldo_mirrors,
593 sizeof(*lo->ldo_mirrors) * mirror_count);
594 lo->ldo_mirror_count = 0;
598 lo->ldo_comp_cnt = comp_count;
602 int lod_fill_mirrors(struct lod_object *lo)
604 struct lod_layout_component *lod_comp;
606 __u16 mirror_id = 0xffff;
610 LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
612 if (!lo->ldo_is_composite)
615 lod_comp = &lo->ldo_comp_entries[0];
616 for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
617 int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
618 int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
620 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
621 lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
622 lo->ldo_mirrors[mirror_idx].lme_primary |= preferred;
623 lo->ldo_mirrors[mirror_idx].lme_end = i;
629 if (mirror_idx >= lo->ldo_mirror_count)
632 mirror_id = mirror_id_of(lod_comp->llc_id);
634 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
635 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
636 lo->ldo_mirrors[mirror_idx].lme_primary = preferred;
637 lo->ldo_mirrors[mirror_idx].lme_start = i;
638 lo->ldo_mirrors[mirror_idx].lme_end = i;
640 if (mirror_idx != lo->ldo_mirror_count - 1)
647 * Generate on-disk lov_mds_md structure for each layout component based on
648 * the information in lod_object->ldo_comp_entries[i].
650 * \param[in] env execution environment for this thread
651 * \param[in] lo LOD object
652 * \param[in] comp_idx index of ldo_comp_entries
653 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
654 * \param[in|out] lmm_size buffer size/lmm size
655 * \param[in] is_dir generate lov ea for dir or file? For dir case,
656 * the stripe info is from the default stripe
657 * template, which is collected in lod_ah_init(),
658 * either from parent object or root object; for
659 * file case, it's from the @lo object
661 * \retval 0 if on disk structure is created successfully
662 * \retval negative error number on failure
664 static int lod_gen_component_ea(const struct lu_env *env,
665 struct lod_object *lo, int comp_idx,
666 struct lov_mds_md *lmm, int *lmm_size,
669 struct lod_thread_info *info = lod_env_info(env);
670 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
671 struct lod_device *lod;
672 struct lov_ost_data_v1 *objs;
673 struct lod_layout_component *lod_comp;
682 &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
684 lod_comp = &lo->ldo_comp_entries[comp_idx];
686 magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
687 if (lod_comp->llc_pattern == 0) /* default striping */
688 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
690 lmm->lmm_magic = cpu_to_le32(magic);
691 lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
692 fid_to_lmm_oi(fid, &lmm->lmm_oi);
693 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
694 lmm->lmm_oi.oi.oi_id++;
695 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
697 lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
698 lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
700 * for dir and uninstantiated component, lmm_layout_gen stores
701 * default stripe offset.
703 lmm->lmm_layout_gen =
704 (is_dir || !lod_comp_inited(lod_comp)) ?
705 cpu_to_le16(lod_comp->llc_stripe_offset) :
706 cpu_to_le16(lod_comp->llc_layout_gen);
708 if (magic == LOV_MAGIC_V1) {
709 objs = &lmm->lmm_objects[0];
711 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
712 size_t cplen = strlcpy(v3->lmm_pool_name,
714 sizeof(v3->lmm_pool_name));
715 if (cplen >= sizeof(v3->lmm_pool_name))
717 objs = &v3->lmm_objects[0];
719 stripe_count = lod_comp_entry_stripe_count(lo, lod_comp, is_dir);
720 if (stripe_count == 0 && !is_dir &&
721 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
722 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
725 if (!is_dir && lo->ldo_is_composite)
726 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
728 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
731 /* generate ost_idx of this component stripe */
732 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
733 for (i = 0; i < stripe_count; i++) {
734 struct dt_object *object;
735 __u32 ost_idx = (__u32)-1UL;
736 int type = LU_SEQ_RANGE_OST;
738 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
739 object = lod_comp->llc_stripe[i];
740 /* instantiated component */
741 info->lti_fid = *lu_object_fid(&object->do_lu);
743 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
745 if (cfs_fail_val == 0)
746 cfs_fail_val = info->lti_fid.f_oid;
748 info->lti_fid.f_oid = cfs_fail_val;
751 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
754 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
755 objs[i].l_ost_gen = cpu_to_le32(0);
756 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
759 rc = lod_fld_lookup(env, lod, &info->lti_fid,
762 CERROR("%s: Can not locate "DFID": rc = %d\n",
763 lod2obd(lod)->obd_name,
764 PFID(&info->lti_fid), rc);
767 } else if (lod_comp->llc_ostlist.op_array &&
768 lod_comp->llc_ostlist.op_count) {
769 /* user specified ost list */
770 ost_idx = lod_comp->llc_ostlist.op_array[i];
773 * with un-instantiated or with no specified ost list
774 * component, its l_ost_idx does not matter.
776 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
779 if (lmm_size != NULL)
780 *lmm_size = lov_mds_md_size(stripe_count, magic);
785 * Generate on-disk lov_mds_md structure based on the information in
786 * the lod_object->ldo_comp_entries.
788 * \param[in] env execution environment for this thread
789 * \param[in] lo LOD object
790 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
791 * \param[in|out] lmm_size buffer size/lmm size
792 * \param[in] is_dir generate lov ea for dir or file? For dir case,
793 * the stripe info is from the default stripe
794 * template, which is collected in lod_ah_init(),
795 * either from parent object or root object; for
796 * file case, it's from the @lo object
798 * \retval 0 if on disk structure is created successfully
799 * \retval negative error number on failure
801 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
802 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
804 struct lov_comp_md_entry_v1 *lcme;
805 struct lov_comp_md_v1 *lcm;
806 struct lod_layout_component *comp_entries;
807 __u16 comp_cnt, mirror_cnt;
808 bool is_composite, is_foreign = false;
809 int i, rc = 0, offset;
813 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
814 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
815 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
817 lo->ldo_def_striping->lds_def_striping_is_composite;
819 comp_cnt = lo->ldo_comp_cnt;
820 mirror_cnt = lo->ldo_mirror_count;
821 comp_entries = lo->ldo_comp_entries;
822 is_composite = lo->ldo_is_composite;
823 is_foreign = lo->ldo_is_foreign;
826 LASSERT(lmm_size != NULL);
829 struct lov_foreign_md *lfm;
831 lfm = (struct lov_foreign_md *)lmm;
832 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
833 /* need to store little-endian */
834 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
835 __swab32s(&lfm->lfm_magic);
836 __swab32s(&lfm->lfm_length);
837 __swab32s(&lfm->lfm_type);
838 __swab32s(&lfm->lfm_flags);
840 *lmm_size = lo->ldo_foreign_lov_size;
844 LASSERT(comp_cnt != 0 && comp_entries != NULL);
847 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
851 lcm = (struct lov_comp_md_v1 *)lmm;
852 memset(lcm, 0, sizeof(*lcm));
854 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
855 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
856 lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
857 lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
859 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
860 LASSERT(offset % sizeof(__u64) == 0);
862 for (i = 0; i < comp_cnt; i++) {
863 struct lod_layout_component *lod_comp;
864 struct lov_mds_md *sub_md;
867 lod_comp = &comp_entries[i];
868 lcme = &lcm->lcm_entries[i];
870 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
871 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
873 /* component could be un-inistantiated */
874 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
875 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
876 lcme->lcme_timestamp =
877 cpu_to_le64(lod_comp->llc_timestamp);
878 if (lod_comp->llc_flags & LCME_FL_EXTENSION && !is_dir)
879 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
881 lcme->lcme_extent.e_start =
882 cpu_to_le64(lod_comp->llc_extent.e_start);
883 lcme->lcme_extent.e_end =
884 cpu_to_le64(lod_comp->llc_extent.e_end);
885 lcme->lcme_offset = cpu_to_le32(offset);
887 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
888 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
891 lcme->lcme_size = cpu_to_le32(size);
893 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
894 "offset:%d lmm_size:%d\n", offset, *lmm_size);
896 lcm->lcm_size = cpu_to_le32(offset);
897 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
899 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
910 * Fill lti_ea_store buffer in the environment with a value for the given
911 * EA. The buffer is reallocated if the value doesn't fit.
913 * \param[in,out] env execution environment for this thread
914 * .lti_ea_store buffer is filled with EA's value
915 * \param[in] lo LOD object
916 * \param[in] name name of the EA
918 * \retval > 0 if EA is fetched successfully
919 * \retval 0 if EA is empty
920 * \retval negative error number on failure
922 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
925 struct lod_thread_info *info = lod_env_info(env);
926 struct dt_object *next = dt_object_child(&lo->ldo_obj);
932 if (unlikely(info->lti_ea_store == NULL)) {
933 /* just to enter in allocation block below */
937 info->lti_buf.lb_buf = info->lti_ea_store;
938 info->lti_buf.lb_len = info->lti_ea_store_size;
939 rc = dt_xattr_get(env, next, &info->lti_buf, name);
942 /* if object is not striped or inaccessible */
943 if (rc == -ENODATA || rc == -ENOENT)
947 /* EA doesn't fit, reallocate new buffer */
948 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
949 if (rc == -ENODATA || rc == -ENOENT)
955 rc = lod_ea_store_resize(info, rc);
965 * Verify the target index is present in the current configuration.
967 * \param[in] md LOD device where the target table is stored
968 * \param[in] idx target's index
970 * \retval 0 if the index is present
971 * \retval -EINVAL if not
973 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
975 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
976 !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
977 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
978 md->lod_ost_descs.ltd_tgts_size);
982 if (unlikely(OST_TGT(md, idx) == NULL)) {
983 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
984 lod2obd(md)->obd_name, idx);
988 if (unlikely(OST_TGT(md, idx)->ltd_tgt == NULL)) {
989 CERROR("%s: invalid lod device, for idx: %d\n",
990 lod2obd(md)->obd_name , idx);
998 * Instantiate objects for stripes.
1000 * Allocate and initialize LU-objects representing the stripes. The number
1001 * of the stripes (ldo_stripe_count) must be initialized already. The caller
1002 * must ensure nobody else is calling the function on the object at the same
1003 * time. FLDB service must be running to be able to map a FID to the targets
1004 * and find appropriate device representing that target.
1006 * \param[in] env execution environment for this thread
1007 * \param[in,out] lo LOD object
1008 * \param[in] objs an array of IDs to creates the objects from
1009 * \param[in] comp_idx index of ldo_comp_entries
1011 * \retval 0 if the objects are instantiated successfully
1012 * \retval negative error number on failure
1014 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1015 struct lov_ost_data_v1 *objs, int comp_idx)
1017 struct lod_layout_component *lod_comp;
1018 struct lod_thread_info *info = lod_env_info(env);
1019 struct lod_device *md;
1020 struct lu_object *o, *n;
1021 struct lu_device *nd;
1022 struct dt_object **stripe = NULL;
1023 __u32 *ost_indices = NULL;
1029 LASSERT(lo != NULL);
1030 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1032 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1033 lod_comp = &lo->ldo_comp_entries[comp_idx];
1035 LASSERT(lod_comp->llc_stripe == NULL);
1036 LASSERT(lod_comp->llc_stripe_count > 0);
1037 LASSERT(lod_comp->llc_stripe_size > 0);
1039 stripe_len = lod_comp->llc_stripe_count;
1040 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1043 OBD_ALLOC(ost_indices, sizeof(*ost_indices) * stripe_len);
1045 GOTO(out, rc = -ENOMEM);
1047 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1048 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1051 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1052 idx = le32_to_cpu(objs[i].l_ost_idx);
1053 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1056 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1057 PFID(&info->lti_fid));
1058 lod_getref(&md->lod_ost_descs);
1060 rc = validate_lod_and_idx(md, idx);
1061 if (unlikely(rc != 0)) {
1062 lod_putref(md, &md->lod_ost_descs);
1066 nd = &OST_TGT(md, idx)->ltd_tgt->dd_lu_dev;
1067 lod_putref(md, &md->lod_ost_descs);
1069 /* In the function below, .hs_keycmp resolves to
1070 * u_obj_hop_keycmp() */
1071 /* coverity[overrun-buffer-val] */
1072 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1074 GOTO(out, rc = PTR_ERR(o));
1076 n = lu_object_locate(o->lo_header, nd->ld_type);
1079 stripe[i] = container_of(n, struct dt_object, do_lu);
1080 ost_indices[i] = idx;
1085 for (i = 0; i < stripe_len; i++)
1086 if (stripe[i] != NULL)
1087 dt_object_put(env, stripe[i]);
1089 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1090 lod_comp->llc_stripe_count = 0;
1092 OBD_FREE(ost_indices,
1093 sizeof(*ost_indices) * stripe_len);
1095 lod_comp->llc_stripe = stripe;
1096 lod_comp->llc_ost_indices = ost_indices;
1097 lod_comp->llc_stripes_allocated = stripe_len;
1104 * Instantiate objects for striping.
1106 * Parse striping information in \a buf and instantiate the objects
1107 * representing the stripes.
1109 * \param[in] env execution environment for this thread
1110 * \param[in] lo LOD object
1111 * \param[in] buf buffer storing LOV EA to parse
1113 * \retval 0 if parsing and objects creation succeed
1114 * \retval negative error number on failure
1116 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1117 const struct lu_buf *buf)
1119 struct lov_mds_md_v1 *lmm;
1120 struct lov_comp_md_v1 *comp_v1 = NULL;
1121 struct lov_foreign_md *foreign = NULL;
1122 struct lov_ost_data_v1 *objs;
1123 __u32 magic, pattern;
1124 __u16 mirror_cnt = 0;
1130 LASSERT(buf->lb_buf);
1131 LASSERT(buf->lb_len);
1132 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1134 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1135 magic = le32_to_cpu(lmm->lmm_magic);
1137 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1138 magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN &&
1139 magic != LOV_MAGIC_SEL)
1140 GOTO(out, rc = -EINVAL);
1142 if (lo->ldo_is_foreign)
1143 lod_free_foreign_lov(lo);
1145 lod_free_comp_entries(lo);
1147 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1148 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1149 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1151 GOTO(out, rc = -EINVAL);
1152 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1153 lo->ldo_is_composite = 1;
1154 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1156 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1157 } else if (magic == LOV_MAGIC_FOREIGN) {
1160 foreign = (struct lov_foreign_md *)buf->lb_buf;
1161 length = offsetof(typeof(*foreign), lfm_value);
1162 if (buf->lb_len < length ||
1163 buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1165 "buf len %zu too small for lov_foreign_md\n",
1167 GOTO(out, rc = -EINVAL);
1170 /* just cache foreign LOV EA raw */
1171 rc = lod_alloc_foreign_lov(lo, length);
1174 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1178 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1179 lo->ldo_is_composite = 0;
1182 rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1186 for (i = 0; i < comp_cnt; i++) {
1187 struct lod_layout_component *lod_comp;
1188 struct lu_extent *ext;
1191 lod_comp = &lo->ldo_comp_entries[i];
1192 if (lo->ldo_is_composite) {
1193 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1194 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1196 ext = &comp_v1->lcm_entries[i].lcme_extent;
1197 lod_comp->llc_extent.e_start =
1198 le64_to_cpu(ext->e_start);
1199 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1200 lod_comp->llc_flags =
1201 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1202 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1203 lod_comp->llc_timestamp = le64_to_cpu(
1204 comp_v1->lcm_entries[i].lcme_timestamp);
1206 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1207 if (lod_comp->llc_id == LCME_ID_INVAL)
1208 GOTO(out, rc = -EINVAL);
1210 if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
1211 comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
1212 struct lod_device *d =
1213 lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1215 CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
1216 lod2obd(d)->obd_name,
1217 lod_comp->llc_flags, lod_comp->llc_id, i,
1218 PFID(lod_object_fid(lo)),
1219 le32_to_cpu(comp_v1->lcm_magic));
1222 lod_comp_set_init(lod_comp);
1225 pattern = le32_to_cpu(lmm->lmm_pattern);
1226 if (!lov_pattern_supported(lov_pattern(pattern)))
1227 GOTO(out, rc = -EINVAL);
1229 lod_comp->llc_pattern = pattern;
1230 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1231 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1232 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1234 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1235 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1237 lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1238 objs = &v3->lmm_objects[0];
1240 lod_set_pool(&lod_comp->llc_pool, NULL);
1241 objs = &lmm->lmm_objects[0];
1245 * If uninstantiated template component has valid l_ost_idx,
1246 * then user has specified ost list for this component.
1248 if (!lod_comp_inited(lod_comp)) {
1251 if (objs[0].l_ost_idx != (__u32)-1UL) {
1254 stripe_count = lod_comp_entry_stripe_count(
1255 lo, lod_comp, false);
1256 if (stripe_count == 0 &&
1257 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1258 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1259 GOTO(out, rc = -E2BIG);
1261 * load the user specified ost list, when this
1262 * component is instantiated later, it will be
1263 * used in lod_alloc_ost_list().
1265 lod_comp->llc_ostlist.op_count = stripe_count;
1266 lod_comp->llc_ostlist.op_size =
1267 stripe_count * sizeof(__u32);
1268 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1269 lod_comp->llc_ostlist.op_size);
1270 if (!lod_comp->llc_ostlist.op_array)
1271 GOTO(out, rc = -ENOMEM);
1273 for (j = 0; j < stripe_count; j++)
1274 lod_comp->llc_ostlist.op_array[j] =
1275 le32_to_cpu(objs[j].l_ost_idx);
1278 * this component OST objects starts from the
1279 * first ost_idx, lod_alloc_ost_list() will
1282 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1285 * for uninstantiated component,
1286 * lmm_layout_gen stores default stripe offset.
1288 lod_comp->llc_stripe_offset =
1289 lmm->lmm_layout_gen;
1293 /* skip un-instantiated component object initialization */
1294 if (!lod_comp_inited(lod_comp))
1297 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1298 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1299 rc = lod_initialize_objects(env, lo, objs, i);
1305 rc = lod_fill_mirrors(lo);
1311 lod_striping_free_nolock(env, lo);
1316 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1317 * is already cached.
1319 * \param[in] lo LOD object
1321 * \retval True if the striping is cached, otherwise
1324 static bool lod_striping_loaded(struct lod_object *lo)
1326 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1327 lo->ldo_comp_cached)
1330 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1331 if (lo->ldo_dir_stripe_loaded)
1334 /* Never load LMV stripe for slaves of striped dir */
1335 if (lo->ldo_dir_slave_stripe)
1343 * A generic function to initialize the stripe objects.
1345 * A protected version of lod_striping_load_locked() - load the striping
1346 * information from storage, parse that and instantiate LU objects to
1347 * represent the stripes. The LOD object \a lo supplies a pointer to the
1348 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1349 * return an array of references to the newly instantiated objects.
1351 * \param[in] env execution environment for this thread
1352 * \param[in,out] lo LOD object, where striping is stored and
1353 * which gets an array of references
1355 * \retval 0 if parsing and object creation succeed
1356 * \retval negative error number on failure
1358 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1360 struct lod_thread_info *info = lod_env_info(env);
1361 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1362 struct lu_buf *buf = &info->lti_buf;
1367 if (!dt_object_exists(next))
1370 if (lod_striping_loaded(lo))
1373 mutex_lock(&lo->ldo_layout_mutex);
1374 if (lod_striping_loaded(lo))
1375 GOTO(unlock, rc = 0);
1377 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1378 rc = lod_get_lov_ea(env, lo);
1383 * there is LOV EA (striping information) in this object
1384 * let's parse it and create in-core objects for the stripes
1386 buf->lb_buf = info->lti_ea_store;
1387 buf->lb_len = info->lti_ea_store_size;
1388 rc = lod_parse_striping(env, lo, buf);
1390 lo->ldo_comp_cached = 1;
1391 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1392 rc = lod_get_lmv_ea(env, lo);
1393 if (rc > sizeof(struct lmv_foreign_md)) {
1394 struct lmv_foreign_md *lfm = info->lti_ea_store;
1396 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
1397 lo->ldo_foreign_lmv = info->lti_ea_store;
1398 lo->ldo_foreign_lmv_size =
1399 info->lti_ea_store_size;
1400 info->lti_ea_store = NULL;
1401 info->lti_ea_store_size = 0;
1403 lo->ldo_dir_stripe_loaded = 1;
1404 lo->ldo_dir_is_foreign = 1;
1405 GOTO(unlock, rc = 0);
1409 if (rc < (int)sizeof(struct lmv_mds_md_v1)) {
1410 /* Let's set stripe_loaded to avoid further
1411 * stripe loading especially for non-stripe directory,
1412 * which can hurt performance. (See LU-9840)
1415 lo->ldo_dir_stripe_loaded = 1;
1416 GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1418 buf->lb_buf = info->lti_ea_store;
1419 buf->lb_len = info->lti_ea_store_size;
1420 if (rc == sizeof(struct lmv_mds_md_v1)) {
1421 rc = lod_load_lmv_shards(env, lo, buf, true);
1422 if (buf->lb_buf != info->lti_ea_store) {
1423 OBD_FREE_LARGE(info->lti_ea_store,
1424 info->lti_ea_store_size);
1425 info->lti_ea_store = buf->lb_buf;
1426 info->lti_ea_store_size = buf->lb_len;
1434 * there is LMV EA (striping information) in this object
1435 * let's parse it and create in-core objects for the stripes
1437 rc = lod_parse_dir_striping(env, lo, buf);
1439 lo->ldo_dir_stripe_loaded = 1;
1443 mutex_unlock(&lo->ldo_layout_mutex);
1448 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1449 const struct lu_buf *buf)
1455 mutex_lock(&lo->ldo_layout_mutex);
1456 lod_striping_free_nolock(env, lo);
1457 rc = lod_parse_striping(env, lo, buf);
1458 mutex_unlock(&lo->ldo_layout_mutex);
1464 * Verify lov_user_md_v1/v3 striping.
1466 * Check the validity of all fields including the magic, stripe size,
1467 * stripe count, stripe offset and that the pool is present. Also check
1468 * that each target index points to an existing target. The additional
1469 * \a is_from_disk turns additional checks. In some cases zero fields
1470 * are allowed (like pattern=0).
1472 * \param[in] d LOD device
1473 * \param[in] buf buffer with LOV EA to verify
1474 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1475 * 1 - from disk, do not allow
1477 * \retval 0 if the striping is valid
1478 * \retval -EINVAL if striping is invalid
1480 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1483 struct lov_user_md_v1 *lum;
1484 struct lov_user_md_v3 *lum3;
1485 struct pool_desc *pool = NULL;
1489 __u16 stripe_offset;
1496 if (buf->lb_len < sizeof(*lum)) {
1497 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1499 GOTO(out, rc = -EINVAL);
1502 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1503 if (magic != LOV_USER_MAGIC_V1 &&
1504 magic != LOV_USER_MAGIC_V3 &&
1505 magic != LOV_USER_MAGIC_SPECIFIC) {
1506 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1507 le32_to_cpu(lum->lmm_magic));
1508 GOTO(out, rc = -EINVAL);
1511 /* the user uses "0" for default stripe pattern normally. */
1512 if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1513 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1515 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1516 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1517 le32_to_cpu(lum->lmm_pattern));
1518 GOTO(out, rc = -EINVAL);
1521 /* a released lum comes from creating orphan on hsm release,
1522 * doesn't make sense to verify it. */
1523 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1526 /* 64kB is the largest common page size we see (ia64), and matches the
1528 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1529 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1530 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1531 stripe_size, LOV_MIN_STRIPE_SIZE);
1532 GOTO(out, rc = -EINVAL);
1535 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1536 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1537 lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1538 /* if offset is not within valid range [0, osts_size) */
1539 if (stripe_offset >= d->lod_ost_descs.ltd_tgts_size) {
1540 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1541 stripe_offset, d->lod_ost_descs.ltd_tgts_size);
1542 GOTO(out, rc = -EINVAL);
1545 /* if lmm_stripe_offset is *not* in bitmap */
1546 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1547 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1549 GOTO(out, rc = -EINVAL);
1553 if (magic == LOV_USER_MAGIC_V1)
1554 lum_size = offsetof(struct lov_user_md_v1,
1556 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1557 lum_size = offsetof(struct lov_user_md_v3,
1560 GOTO(out, rc = -EINVAL);
1562 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1563 if (buf->lb_len < lum_size) {
1564 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1565 "magic %#x and stripe_count %u\n",
1566 buf->lb_len, lum_size, magic, stripe_count);
1567 GOTO(out, rc = -EINVAL);
1570 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1574 /* In the function below, .hs_keycmp resolves to
1575 * pool_hashkey_keycmp() */
1576 /* coverity[overrun-buffer-val] */
1577 pool = lod_find_pool(d, lum3->lmm_pool_name);
1581 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1582 rc = lod_check_index_in_pool(stripe_offset, pool);
1584 GOTO(out, rc = -EINVAL);
1587 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1588 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1589 stripe_count, pool_tgt_count(pool));
1590 GOTO(out, rc = -EINVAL);
1595 lod_pool_putref(pool);
1601 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1603 LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1604 LOV_USER_MAGIC_COMP_V1, "Wrong magic %x\n",
1605 le32_to_cpu(comp->lcm_magic));
1606 LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1607 "bad index %d, max = %d\n",
1608 i, le16_to_cpu(comp->lcm_entry_count));
1610 return &comp->lcm_entries[i];
1613 #define for_each_comp_entry_v1(comp, entry) \
1614 for (entry = comp_entry_v1(comp, 0); \
1615 entry <= comp_entry_v1(comp, \
1616 le16_to_cpu(comp->lcm_entry_count) - 1); \
1619 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
1620 struct lov_comp_md_entry_v1 *dom_ent)
1622 struct lov_comp_md_entry_v1 *ent;
1624 __u32 dom_off, dom_size, comp_size;
1625 void *blob_src, *blob_dst;
1626 unsigned int blob_size, blob_shift;
1628 entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1629 /* if file has only DoM stripe return just error */
1633 comp_size = le32_to_cpu(comp_v1->lcm_size);
1634 dom_off = le32_to_cpu(dom_ent->lcme_offset);
1635 dom_size = le32_to_cpu(dom_ent->lcme_size);
1637 /* shift entries array first */
1638 comp_v1->lcm_entry_count = cpu_to_le16(entries);
1639 memmove(dom_ent, dom_ent + 1,
1640 entries * sizeof(struct lov_comp_md_entry_v1));
1642 /* now move blob of layouts */
1643 blob_dst = (void *)comp_v1 + dom_off - sizeof(*dom_ent);
1644 blob_src = (void *)comp_v1 + dom_off + dom_size;
1645 blob_size = (unsigned long)((void *)comp_v1 + comp_size - blob_src);
1646 blob_shift = sizeof(*dom_ent) + dom_size;
1648 memmove(blob_dst, blob_src, blob_size);
1650 for_each_comp_entry_v1(comp_v1, ent) {
1653 off = le32_to_cpu(ent->lcme_offset);
1654 ent->lcme_offset = cpu_to_le32(off - blob_shift);
1657 comp_v1->lcm_size = cpu_to_le32(comp_size - blob_shift);
1659 /* notify a caller to re-check entry */
1663 int lod_fix_dom_stripe(struct lod_device *d, struct lov_comp_md_v1 *comp_v1,
1664 struct lov_comp_md_entry_v1 *dom_ent)
1666 struct lov_comp_md_entry_v1 *ent;
1667 struct lu_extent *dom_ext, *ext;
1668 struct lov_user_md_v1 *lum;
1673 dom_ext = &dom_ent->lcme_extent;
1674 dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1675 stripe_size = d->lod_dom_max_stripesize;
1677 lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1678 CDEBUG(D_LAYOUT, "DoM component size %u was bigger than MDT limit %u, "
1679 "new size is %u\n", le32_to_cpu(lum->lmm_stripe_size),
1680 d->lod_dom_max_stripesize, stripe_size);
1681 lum->lmm_stripe_size = cpu_to_le32(stripe_size);
1683 for_each_comp_entry_v1(comp_v1, ent) {
1687 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1691 ext = &ent->lcme_extent;
1692 if (ext->e_start != dom_ext->e_end)
1695 /* Found next component after the DoM one with the same
1696 * mirror_id and adjust its start with DoM component end.
1698 * NOTE: we are considering here that there can be only one
1699 * DoM component in a file, all replicas are located on OSTs
1700 * always and don't need adjustment since use own layouts.
1702 ext->e_start = cpu_to_le64(stripe_size);
1706 if (stripe_size == 0) {
1707 /* DoM component size is zero due to server setting,
1708 * remove it from the layout */
1709 rc = lod_erase_dom_stripe(comp_v1, dom_ent);
1711 /* Update DoM extent end finally */
1712 dom_ext->e_end = cpu_to_le64(stripe_size);
1719 * Verify LOV striping.
1721 * \param[in] d LOD device
1722 * \param[in] buf buffer with LOV EA to verify
1723 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1724 * 1 - from disk, do not allow
1725 * \param[in] start extent start for composite layout
1727 * \retval 0 if the striping is valid
1728 * \retval -EINVAL if striping is invalid
1730 int lod_verify_striping(struct lod_device *d, struct lod_object *lo,
1731 const struct lu_buf *buf, bool is_from_disk)
1733 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
1734 struct lov_user_md_v1 *lum;
1735 struct lov_comp_md_v1 *comp_v1;
1736 struct lov_comp_md_entry_v1 *ent;
1737 struct lu_extent *ext;
1740 __u32 stripe_size = 0;
1741 __u16 prev_mid = -1, mirror_id = -1;
1747 if (buf->lb_len < sizeof(lum->lmm_magic)) {
1748 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
1754 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1755 /* treat foreign LOV EA/object case first
1756 * XXX is it expected to try setting again a foreign?
1757 * XXX should we care about different current vs new layouts ?
1759 if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
1760 struct lov_foreign_md *lfm = buf->lb_buf;
1762 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
1764 "buf len %zu < min lov_foreign_md size (%zu)\n",
1765 buf->lb_len, offsetof(typeof(*lfm),
1770 if (foreign_size_le(lfm) > buf->lb_len) {
1772 "buf len %zu < this lov_foreign_md size (%zu)\n",
1773 buf->lb_len, foreign_size_le(lfm));
1776 /* Don't do anything with foreign layouts */
1780 /* normal LOV/layout cases */
1782 if (buf->lb_len < sizeof(*lum)) {
1783 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1788 if (magic != LOV_USER_MAGIC_V1 &&
1789 magic != LOV_USER_MAGIC_V3 &&
1790 magic != LOV_USER_MAGIC_SPECIFIC &&
1791 magic != LOV_USER_MAGIC_COMP_V1) {
1792 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1793 le32_to_cpu(lum->lmm_magic));
1797 if (magic != LOV_USER_MAGIC_COMP_V1)
1798 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1800 /* magic == LOV_USER_MAGIC_COMP_V1 */
1801 comp_v1 = buf->lb_buf;
1802 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1803 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1804 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1810 if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1811 CDEBUG(D_LAYOUT, "entry count is zero\n");
1815 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1816 lo->ldo_comp_cnt > 0) {
1817 /* could be called from lustre.lov.add */
1818 __u32 cnt = lo->ldo_comp_cnt;
1820 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
1821 prev_end = ext->e_end;
1826 for_each_comp_entry_v1(comp_v1, ent) {
1827 ext = &ent->lcme_extent;
1829 if (le64_to_cpu(ext->e_start) > le64_to_cpu(ext->e_end)) {
1830 CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
1831 le64_to_cpu(ext->e_start),
1832 le64_to_cpu(ext->e_end));
1837 /* lcme_id contains valid value */
1838 if (le32_to_cpu(ent->lcme_id) == 0 ||
1839 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
1840 CDEBUG(D_LAYOUT, "invalid id %u\n",
1841 le32_to_cpu(ent->lcme_id));
1845 if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
1846 mirror_id = mirror_id_of(
1847 le32_to_cpu(ent->lcme_id));
1849 /* first component must start with 0 */
1850 if (mirror_id != prev_mid &&
1851 le64_to_cpu(ext->e_start) != 0) {
1853 "invalid start:%llu, expect:0\n",
1854 le64_to_cpu(ext->e_start));
1858 prev_mid = mirror_id;
1862 if (le64_to_cpu(ext->e_start) == 0) {
1867 /* the next must be adjacent with the previous one */
1868 if (le64_to_cpu(ext->e_start) != prev_end) {
1870 "invalid start actual:%llu, expect:%llu\n",
1871 le64_to_cpu(ext->e_start), prev_end);
1875 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
1876 tmp.lb_len = le32_to_cpu(ent->lcme_size);
1878 /* Check DoM entry is always the first one */
1880 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
1882 /* DoM component must be the first in a mirror */
1883 if (le64_to_cpu(ext->e_start) > 0) {
1884 CDEBUG(D_LAYOUT, "invalid DoM component "
1885 "with %llu extent start\n",
1886 le64_to_cpu(ext->e_start));
1889 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1890 /* There is just one stripe on MDT and it must
1891 * cover whole component size. */
1892 if (stripe_size != le64_to_cpu(ext->e_end)) {
1893 CDEBUG(D_LAYOUT, "invalid DoM layout "
1894 "stripe size %u != %llu "
1895 "(component size)\n",
1896 stripe_size, prev_end);
1899 /* Check stripe size againts per-MDT limit */
1900 if (stripe_size > d->lod_dom_max_stripesize) {
1901 CDEBUG(D_LAYOUT, "DoM component size "
1902 "%u is bigger than MDT limit %u, check "
1903 "dom_max_stripesize parameter\n",
1904 stripe_size, d->lod_dom_max_stripesize);
1905 rc = lod_fix_dom_stripe(d, comp_v1, ent);
1906 if (rc == -ERESTART) {
1907 /* DoM entry was removed, re-check
1908 * new layout from start */
1914 /* Any stripe count is forbidden on DoM component */
1915 if (lum->lmm_stripe_count) {
1917 "invalid DoM layout stripe count %u, must be 0\n",
1918 le16_to_cpu(lum->lmm_stripe_count));
1922 /* Any pool is forbidden on DoM component */
1923 if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
1924 struct lov_user_md_v3 *v3 = (void *)lum;
1926 if (v3->lmm_pool_name[0] != '\0') {
1928 "DoM component cannot have pool assigned\n");
1934 prev_end = le64_to_cpu(ext->e_end);
1936 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
1940 if (prev_end == LUSTRE_EOF)
1943 /* extent end must be aligned with the stripe_size */
1944 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1945 if (stripe_size == 0)
1946 stripe_size = desc->ld_default_stripe_size;
1947 if (prev_end % stripe_size) {
1948 CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
1949 "stripe_sz: %u, [%llu, %llu)\n",
1950 stripe_size, ext->e_start, prev_end);
1955 /* make sure that the mirror_count is telling the truth */
1956 if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
1963 * set the default stripe size, if unset.
1965 * \param[in,out] val number of bytes per OST stripe
1967 * The minimum stripe size is 64KB to ensure that a single stripe is an
1968 * even multiple of a client PAGE_SIZE (IA64, PPC, etc). Otherwise, it
1969 * is difficult to split dirty pages across OSCs during writes.
1971 void lod_fix_desc_stripe_size(__u64 *val)
1973 if (*val < LOV_MIN_STRIPE_SIZE) {
1975 LCONSOLE_INFO("Increasing default stripe size to "
1976 "minimum value %u\n",
1977 LOV_DESC_STRIPE_SIZE_DEFAULT);
1978 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1979 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1980 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1981 LCONSOLE_WARN("Changing default stripe size to %llu (a "
1982 "multiple of %u)\n",
1983 *val, LOV_MIN_STRIPE_SIZE);
1988 * set the filesystem default number of stripes, if unset.
1990 * \param[in,out] val number of stripes
1992 * A value of "0" means "use the system-wide default stripe count", which
1993 * has either been inherited by now, or falls back to 1 stripe per file.
1994 * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
1995 * and is a valid value, so is left unchanged here.
1997 void lod_fix_desc_stripe_count(__u32 *val)
2004 * set the filesystem default layout pattern
2006 * \param[in,out] val LOV_PATTERN_* layout
2008 * A value of "0" means "use the system-wide default layout type", which
2009 * has either been inherited by now, or falls back to plain RAID0 striping.
2011 void lod_fix_desc_pattern(__u32 *val)
2013 /* from lov_setstripe */
2014 if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
2015 LCONSOLE_WARN("lod: Unknown stripe pattern: %#x\n", *val);
2020 void lod_fix_lmv_desc_pattern(__u32 *val)
2022 if ((*val) && !lmv_is_known_hash_type(*val)) {
2023 LCONSOLE_WARN("lod: Unknown md stripe pattern: %#x\n", *val);
2028 void lod_fix_desc_qos_maxage(__u32 *val)
2030 /* fix qos_maxage */
2032 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2036 * Used to fix insane default striping.
2038 * \param[in] desc striping description
2040 void lod_fix_desc(struct lov_desc *desc)
2042 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2043 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2044 lod_fix_desc_pattern(&desc->ld_pattern);
2045 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2048 static void lod_fix_lmv_desc(struct lmv_desc *desc)
2050 desc->ld_active_tgt_count = 0;
2051 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2052 lod_fix_lmv_desc_pattern(&desc->ld_pattern);
2053 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2057 * Initialize the structures used to store pools and default striping.
2059 * \param[in] lod LOD device
2060 * \param[in] lcfg configuration structure storing default striping.
2062 * \retval 0 if initialization succeeds
2063 * \retval negative error number on failure
2065 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2067 struct obd_device *obd;
2068 struct lov_desc *desc;
2072 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2073 LASSERT(obd != NULL);
2074 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2076 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2077 CERROR("LOD setup requires a descriptor\n");
2081 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2083 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2084 CERROR("descriptor size wrong: %d > %d\n",
2085 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2089 if (desc->ld_magic != LOV_DESC_MAGIC) {
2090 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2091 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2092 obd->obd_name, desc);
2093 lustre_swab_lov_desc(desc);
2095 CERROR("%s: Bad lov desc magic: %#x\n",
2096 obd->obd_name, desc->ld_magic);
2103 desc->ld_active_tgt_count = 0;
2104 lod->lod_ost_descs.ltd_lov_desc = *desc;
2106 /* NB: config doesn't contain lmv_desc, alter it via sysfs. */
2107 lod_fix_lmv_desc(&lod->lod_mdt_descs.ltd_lmv_desc);
2109 lod->lod_sp_me = LUSTRE_SP_CLI;
2111 /* Set up OST pool environment */
2112 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
2113 HASH_POOLS_MAX_BITS,
2114 HASH_POOLS_BKT_BITS, 0,
2117 &pool_hash_operations,
2119 if (lod->lod_pools_hash_body == NULL)
2122 INIT_LIST_HEAD(&lod->lod_pool_list);
2123 lod->lod_pool_count = 0;
2124 rc = lod_tgt_pool_init(&lod->lod_mdt_descs.ltd_tgt_pool, 0);
2128 rc = lod_tgt_pool_init(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool, 0);
2130 GOTO(out_mdt_pool, rc);
2132 rc = lod_tgt_pool_init(&lod->lod_ost_descs.ltd_tgt_pool, 0);
2134 GOTO(out_mdt_rr_pool, rc);
2136 rc = lod_tgt_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
2138 GOTO(out_ost_pool, rc);
2143 lod_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2145 lod_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2147 lod_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
2149 cfs_hash_putref(lod->lod_pools_hash_body);
2155 * Release the structures describing the pools.
2157 * \param[in] lod LOD device from which we release the structures
2161 int lod_pools_fini(struct lod_device *lod)
2163 struct obd_device *obd = lod2obd(lod);
2164 struct pool_desc *pool, *tmp;
2167 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2168 /* free pool structs */
2169 CDEBUG(D_INFO, "delete pool %p\n", pool);
2170 /* In the function below, .hs_keycmp resolves to
2171 * pool_hashkey_keycmp() */
2172 /* coverity[overrun-buffer-val] */
2173 lod_pool_del(obd, pool->pool_name);
2176 cfs_hash_putref(lod->lod_pools_hash_body);
2177 lod_tgt_pool_free(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool);
2178 lod_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2179 lod_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2180 lod_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);