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 LTD_TGT(ltd, idx) = NULL;
96 /*FIXME: only support ost pool for now */
97 if (ltd == &lod->lod_ost_descs) {
98 lod_ost_pool_remove(&lod->lod_pool_info, idx);
99 if (tgt_desc->ltd_active)
100 lod->lod_desc.ld_active_tgt_count--;
103 cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
104 ltd->ltd_death_row--;
106 mutex_unlock(<d->ltd_mutex);
107 up_read(<d->ltd_rw_sem);
109 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
111 list_del(&tgt_desc->ltd_kill);
112 if (ltd == &lod->lod_ost_descs) {
113 /* remove from QoS structures */
114 rc = qos_del_tgt(lod, tgt_desc);
116 CERROR("%s: qos_del_tgt(%s) failed:"
118 lod2obd(lod)->obd_name,
119 obd_uuid2str(&tgt_desc->ltd_uuid),
122 rc = obd_disconnect(tgt_desc->ltd_exp);
124 CERROR("%s: failed to disconnect %s: rc = %d\n",
125 lod2obd(lod)->obd_name,
126 obd_uuid2str(&tgt_desc->ltd_uuid), rc);
127 OBD_FREE_PTR(tgt_desc);
130 mutex_unlock(<d->ltd_mutex);
131 up_read(<d->ltd_rw_sem);
136 * Expand size of target table.
138 * When the target table is full, we have to extend the table. To do so,
139 * we allocate new memory with some reserve, move data from the old table
140 * to the new one and release memory consumed by the old table.
141 * Notice we take ltd_rw_sem exclusively to ensure atomic switch.
143 * \param[in] ltd target table
144 * \param[in] newsize new size of the table
146 * \retval 0 on success
147 * \retval -ENOMEM if reallocation failed
149 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
151 struct cfs_bitmap *new_bitmap, *old_bitmap = NULL;
155 /* grab write reference on the lod. Relocating the array requires
156 * exclusive access */
158 down_write(<d->ltd_rw_sem);
159 if (newsize <= ltd->ltd_tgts_size)
160 /* someone else has already resize the array */
163 /* allocate new bitmap */
164 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
166 GOTO(out, rc = -ENOMEM);
168 if (ltd->ltd_tgts_size > 0) {
169 /* the bitmap already exists, we need
170 * to copy data from old one */
171 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
172 old_bitmap = ltd->ltd_tgt_bitmap;
175 ltd->ltd_tgts_size = newsize;
176 ltd->ltd_tgt_bitmap = new_bitmap;
179 CFS_FREE_BITMAP(old_bitmap);
181 CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
185 up_write(<d->ltd_rw_sem);
190 * Connect LOD to a new OSP and add it to the target table.
192 * Connect to the OSP device passed, initialize all the internal
193 * structures related to the device and add it to the target table.
195 * \param[in] env execution environment for this thread
196 * \param[in] lod LOD device to be connected to the new OSP
197 * \param[in] osp name of OSP device name to be added
198 * \param[in] index index of the new target
199 * \param[in] gen target's generation number
200 * \param[in] tgt_index OSP's group
201 * \param[in] type type of device (mdc or osc)
202 * \param[in] active state of OSP: 0 - inactive, 1 - active
204 * \retval 0 if added successfully
205 * \retval negative error number on failure
207 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
208 char *osp, unsigned index, unsigned gen, int tgt_index,
209 char *type, int active)
211 struct obd_connect_data *data = NULL;
212 struct obd_export *exp = NULL;
213 struct obd_device *obd;
214 struct lu_device *lu_dev;
215 struct dt_device *dt_dev;
217 struct lod_tgt_desc *tgt_desc;
218 struct lod_tgt_descs *ltd;
219 struct lustre_cfg *lcfg;
220 struct obd_uuid obd_uuid;
223 bool connected = false;
226 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
229 CERROR("request to add OBD %s with invalid generation: %d\n",
234 obd_str2uuid(&obd_uuid, osp);
236 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
237 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
239 CERROR("can't find %s device\n", osp);
243 LASSERT(obd->obd_lu_dev != NULL);
244 LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
246 lu_dev = obd->obd_lu_dev;
247 dt_dev = lu2dt_dev(lu_dev);
251 GOTO(out_cleanup, rc = -ENOMEM);
253 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
254 data->ocd_version = LUSTRE_VERSION_CODE;
255 data->ocd_index = index;
257 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
259 data->ocd_connect_flags |= OBD_CONNECT_AT |
262 #ifdef HAVE_LRU_RESIZE_SUPPORT
263 OBD_CONNECT_LRU_RESIZE |
266 OBD_CONNECT_REQPORTAL |
267 OBD_CONNECT_SKIP_ORPHAN |
269 OBD_CONNECT_LVB_TYPE |
270 OBD_CONNECT_VERSION |
271 OBD_CONNECT_PINGLESS |
273 OBD_CONNECT_BULK_MBITS;
275 data->ocd_group = tgt_index;
276 ltd = &lod->lod_ost_descs;
278 struct obd_import *imp = obd->u.cli.cl_import;
281 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
282 data->ocd_connect_flags |= OBD_CONNECT_ACL |
284 OBD_CONNECT_MDS_MDS |
289 OBD_CONNECT_BULK_MBITS;
290 spin_lock(&imp->imp_lock);
291 imp->imp_server_timeout = 1;
292 spin_unlock(&imp->imp_lock);
293 imp->imp_client->cli_request_portal = OUT_PORTAL;
294 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
296 ltd = &lod->lod_mdt_descs;
299 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
302 CERROR("%s: cannot connect to next dev %s (%d)\n",
303 obd->obd_name, osp, rc);
304 GOTO(out_cleanup, rc);
308 /* Allocate ost descriptor and fill it */
309 OBD_ALLOC_PTR(tgt_desc);
311 GOTO(out_cleanup, rc = -ENOMEM);
313 tgt_desc->ltd_tgt = dt_dev;
314 tgt_desc->ltd_exp = exp;
315 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
316 tgt_desc->ltd_gen = gen;
317 tgt_desc->ltd_index = index;
318 tgt_desc->ltd_active = active;
321 if (index >= ltd->ltd_tgts_size) {
322 /* we have to increase the size of the lod_osts array */
325 newsize = max(ltd->ltd_tgts_size, (__u32)2);
326 while (newsize < index + 1)
327 newsize = newsize << 1;
329 /* lod_bitmap_resize() needs lod_rw_sem
330 * which we hold with th reference */
331 lod_putref(lod, ltd);
333 rc = ltd_bitmap_resize(ltd, newsize);
340 mutex_lock(<d->ltd_mutex);
342 if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
343 CERROR("%s: device %d is registered already\n", obd->obd_name,
345 GOTO(out_mutex, rc = -EEXIST);
348 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
349 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
350 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
351 CERROR("can't allocate index to add %s\n",
353 GOTO(out_mutex, rc = -ENOMEM);
358 /* pool and qos are not supported for MDS stack yet */
359 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
362 CERROR("%s: can't set up pool, failed with %d\n",
367 rc = qos_add_tgt(lod, tgt_desc);
369 CERROR("%s: qos_add_tgt failed with %d\n",
374 /* The new OST is now a full citizen */
375 if (index >= lod->lod_desc.ld_tgt_count)
376 lod->lod_desc.ld_tgt_count = index + 1;
378 lod->lod_desc.ld_active_tgt_count++;
381 LTD_TGT(ltd, index) = tgt_desc;
382 cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
384 mutex_unlock(<d->ltd_mutex);
385 lod_putref(lod, ltd);
387 if (lod->lod_recovery_completed)
388 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
390 if (!for_ost && lod->lod_initialized) {
391 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
393 CERROR("%s: cannot start llog on %s:rc = %d\n",
394 lod2obd(lod)->obd_name, osp, rc);
399 rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
401 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
402 "index = %u, rc = %d\n", osp, type, index, rc);
403 GOTO(out_fini_llog, rc);
407 lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
408 tgt_desc->ltd_recovery_thread);
411 mutex_lock(<d->ltd_mutex);
413 if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
414 struct ptlrpc_thread *thread;
416 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
417 OBD_FREE_PTR(thread);
420 cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index);
421 LTD_TGT(ltd, index) = NULL;
423 lod_ost_pool_remove(&lod->lod_pool_info, index);
426 mutex_unlock(<d->ltd_mutex);
427 lod_putref(lod, ltd);
430 OBD_FREE_PTR(tgt_desc);
432 /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
433 * objects from the MDT stack. See LU-7184. */
434 lcfg = &lod_env_info(env)->lti_lustre_cfg;
435 memset(lcfg, 0, sizeof(*lcfg));
436 lcfg->lcfg_version = LUSTRE_CFG_VERSION;
437 lcfg->lcfg_command = LCFG_CLEANUP;
438 lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
447 * Schedule target removal from the target table.
449 * Mark the device as dead. The device is not removed here because it may
450 * still be in use. The device will be removed in lod_putref() when the
451 * last reference is released.
453 * \param[in] env execution environment for this thread
454 * \param[in] lod LOD device the target table belongs to
455 * \param[in] ltd target table
456 * \param[in] idx index of the target
457 * \param[in] for_ost type of the target: 0 - MDT, 1 - OST
459 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
460 struct lod_tgt_descs *ltd, unsigned idx,
463 LASSERT(LTD_TGT(ltd, idx));
465 lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
468 if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
469 struct ptlrpc_thread *thread;
471 thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
472 OBD_FREE_PTR(thread);
475 if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
476 LTD_TGT(ltd, idx)->ltd_reap = 1;
477 ltd->ltd_death_row++;
482 * Schedule removal of all the targets from the given target table.
484 * See more details in the description for __lod_del_device()
486 * \param[in] env execution environment for this thread
487 * \param[in] lod LOD device the target table belongs to
488 * \param[in] ltd target table
489 * \param[in] for_ost type of the target: MDT or OST
493 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
494 struct lod_tgt_descs *ltd, bool for_ost)
498 if (ltd->ltd_tgts_size <= 0)
501 mutex_lock(<d->ltd_mutex);
502 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
503 __lod_del_device(env, lod, ltd, idx, for_ost);
504 mutex_unlock(<d->ltd_mutex);
505 lod_putref(lod, ltd);
506 CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
507 for (idx = 0; idx < TGT_PTRS; idx++) {
508 if (ltd->ltd_tgt_idx[idx])
509 OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
511 ltd->ltd_tgts_size = 0;
516 * Remove device by name.
518 * Remove a device identified by \a osp from the target table. Given
519 * the device can be in use, the real deletion happens in lod_putref().
521 * \param[in] env execution environment for this thread
522 * \param[in] lod LOD device to be connected to the new OSP
523 * \param[in] ltd target table
524 * \param[in] osp name of OSP device to be removed
525 * \param[in] idx index of the target
526 * \param[in] gen generation number, not used currently
527 * \param[in] for_ost type of the target: 0 - MDT, 1 - OST
529 * \retval 0 if the device was scheduled for removal
530 * \retval -EINVAL if no device was found
532 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
533 struct lod_tgt_descs *ltd, char *osp, unsigned idx,
534 unsigned gen, bool for_ost)
536 struct obd_device *obd;
538 struct obd_uuid uuid;
541 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
543 obd_str2uuid(&uuid, osp);
545 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
546 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
548 CERROR("can't find %s device\n", osp);
553 CERROR("%s: request to remove OBD %s with invalid generation %d"
554 "\n", obd->obd_name, osp, gen);
558 obd_str2uuid(&uuid, osp);
561 mutex_lock(<d->ltd_mutex);
562 /* check that the index is allocated in the bitmap */
563 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
564 !LTD_TGT(ltd, idx)) {
565 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
566 GOTO(out, rc = -EINVAL);
569 /* check that the UUID matches */
570 if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
571 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
572 obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
574 GOTO(out, rc = -EINVAL);
577 __lod_del_device(env, lod, ltd, idx, for_ost);
580 mutex_unlock(<d->ltd_mutex);
581 lod_putref(lod, ltd);
586 * Resize per-thread storage to hold specified size.
588 * A helper function to resize per-thread temporary storage. This storage
589 * is used to process LOV/LVM EAs and may be quite large. We do not want to
590 * allocate/release it every time, so instead we put it into the env and
591 * reallocate on demand. The memory is released when the correspondent thread
594 * \param[in] info LOD-specific storage in the environment
595 * \param[in] size new size to grow the buffer to
597 * \retval 0 on success, -ENOMEM if reallocation failed
599 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
601 __u32 round = size_roundup_power2(size);
603 if (info->lti_ea_store) {
604 LASSERT(info->lti_ea_store_size);
605 LASSERT(info->lti_ea_store_size < round);
606 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
607 info->lti_ea_store_size, round);
608 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
609 info->lti_ea_store = NULL;
610 info->lti_ea_store_size = 0;
613 OBD_ALLOC_LARGE(info->lti_ea_store, round);
614 if (info->lti_ea_store == NULL)
616 info->lti_ea_store_size = round;
621 static void lod_free_comp_buffer(struct lod_layout_component *entries,
622 __u16 count, __u32 bufsize)
624 struct lod_layout_component *entry;
627 for (i = 0; i < count; i++) {
629 if (entry->llc_pool != NULL)
630 lod_set_pool(&entry->llc_pool, NULL);
631 if (entry->llc_ostlist.op_array)
632 OBD_FREE(entry->llc_ostlist.op_array,
633 entry->llc_ostlist.op_size);
634 LASSERT(entry->llc_stripe == NULL);
635 LASSERT(entry->llc_stripes_allocated == 0);
639 OBD_FREE_LARGE(entries, bufsize);
642 void lod_free_def_comp_entries(struct lod_default_striping *lds)
644 lod_free_comp_buffer(lds->lds_def_comp_entries,
645 lds->lds_def_comp_size_cnt,
647 sizeof(*lds->lds_def_comp_entries) *
648 lds->lds_def_comp_size_cnt));
649 lds->lds_def_comp_entries = NULL;
650 lds->lds_def_comp_cnt = 0;
651 lds->lds_def_striping_is_composite = 0;
652 lds->lds_def_comp_size_cnt = 0;
656 * Resize per-thread storage to hold default striping component entries
658 * A helper function to resize per-thread temporary storage. This storage
659 * is used to hold default LOV/LVM EAs and may be quite large. We do not want
660 * to allocate/release it every time, so instead we put it into the env and
661 * reallocate it on demand. The memory is released when the correspondent
662 * thread is finished.
664 * \param[in,out] lds default striping
665 * \param[in] count new component count to grow the buffer to
667 * \retval 0 on success, -ENOMEM if reallocation failed
669 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
671 struct lod_layout_component *entries;
672 __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
674 __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
675 lds->lds_def_comp_size_cnt);
680 OBD_ALLOC_LARGE(entries, new);
684 if (lds->lds_def_comp_entries != NULL) {
685 CDEBUG(D_INFO, "default striping component size %d is not "
686 "enough, need %d\n", old, new);
687 lod_free_def_comp_entries(lds);
690 lds->lds_def_comp_entries = entries;
691 lds->lds_def_comp_size_cnt = count;
696 void lod_free_comp_entries(struct lod_object *lo)
698 if (lo->ldo_mirrors) {
699 OBD_FREE(lo->ldo_mirrors,
700 sizeof(*lo->ldo_mirrors) * lo->ldo_mirror_count);
701 lo->ldo_mirrors = NULL;
702 lo->ldo_mirror_count = 0;
704 lod_free_comp_buffer(lo->ldo_comp_entries,
706 sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
707 lo->ldo_comp_entries = NULL;
708 lo->ldo_comp_cnt = 0;
709 lo->ldo_is_composite = 0;
712 int lod_alloc_comp_entries(struct lod_object *lo,
713 int mirror_count, int comp_count)
715 LASSERT(comp_count != 0);
716 LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
718 if (mirror_count > 0) {
719 OBD_ALLOC(lo->ldo_mirrors,
720 sizeof(*lo->ldo_mirrors) * mirror_count);
721 if (!lo->ldo_mirrors)
724 lo->ldo_mirror_count = mirror_count;
727 OBD_ALLOC_LARGE(lo->ldo_comp_entries,
728 sizeof(*lo->ldo_comp_entries) * comp_count);
729 if (lo->ldo_comp_entries == NULL) {
730 OBD_FREE(lo->ldo_mirrors,
731 sizeof(*lo->ldo_mirrors) * mirror_count);
732 lo->ldo_mirror_count = 0;
736 lo->ldo_comp_cnt = comp_count;
740 int lod_fill_mirrors(struct lod_object *lo)
742 struct lod_layout_component *lod_comp;
744 __u16 mirror_id = 0xffff;
748 LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
750 if (!lo->ldo_is_composite)
753 lod_comp = &lo->ldo_comp_entries[0];
754 for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
755 int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
756 int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
758 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
759 lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
760 lo->ldo_mirrors[mirror_idx].lme_primary |= preferred;
761 lo->ldo_mirrors[mirror_idx].lme_end = i;
767 if (mirror_idx >= lo->ldo_mirror_count)
770 mirror_id = mirror_id_of(lod_comp->llc_id);
772 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
773 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
774 lo->ldo_mirrors[mirror_idx].lme_primary = preferred;
775 lo->ldo_mirrors[mirror_idx].lme_start = i;
776 lo->ldo_mirrors[mirror_idx].lme_end = i;
778 if (mirror_idx != lo->ldo_mirror_count - 1)
785 * Generate on-disk lov_mds_md structure for each layout component based on
786 * the information in lod_object->ldo_comp_entries[i].
788 * \param[in] env execution environment for this thread
789 * \param[in] lo LOD object
790 * \param[in] comp_idx index of ldo_comp_entries
791 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
792 * \param[in|out] lmm_size buffer size/lmm size
793 * \param[in] is_dir generate lov ea for dir or file? For dir case,
794 * the stripe info is from the default stripe
795 * template, which is collected in lod_ah_init(),
796 * either from parent object or root object; for
797 * file case, it's from the @lo object
799 * \retval 0 if on disk structure is created successfully
800 * \retval negative error number on failure
802 static int lod_gen_component_ea(const struct lu_env *env,
803 struct lod_object *lo, int comp_idx,
804 struct lov_mds_md *lmm, int *lmm_size,
807 struct lod_thread_info *info = lod_env_info(env);
808 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
809 struct lod_device *lod;
810 struct lov_ost_data_v1 *objs;
811 struct lod_layout_component *lod_comp;
820 &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
822 lod_comp = &lo->ldo_comp_entries[comp_idx];
824 magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
825 if (lod_comp->llc_pattern == 0) /* default striping */
826 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
828 lmm->lmm_magic = cpu_to_le32(magic);
829 lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
830 fid_to_lmm_oi(fid, &lmm->lmm_oi);
831 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
832 lmm->lmm_oi.oi.oi_id++;
833 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
835 lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
836 lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
838 * for dir and uninstantiated component, lmm_layout_gen stores
839 * default stripe offset.
841 lmm->lmm_layout_gen =
842 (is_dir || !lod_comp_inited(lod_comp)) ?
843 cpu_to_le16(lod_comp->llc_stripe_offset) :
844 cpu_to_le16(lod_comp->llc_layout_gen);
846 if (magic == LOV_MAGIC_V1) {
847 objs = &lmm->lmm_objects[0];
849 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
850 size_t cplen = strlcpy(v3->lmm_pool_name,
852 sizeof(v3->lmm_pool_name));
853 if (cplen >= sizeof(v3->lmm_pool_name))
855 objs = &v3->lmm_objects[0];
857 stripe_count = lod_comp_entry_stripe_count(lo, lod_comp, is_dir);
858 if (!is_dir && lo->ldo_is_composite)
859 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
861 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
864 /* generate ost_idx of this component stripe */
865 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
866 for (i = 0; i < stripe_count; i++) {
867 struct dt_object *object;
868 __u32 ost_idx = (__u32)-1UL;
869 int type = LU_SEQ_RANGE_OST;
871 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
872 object = lod_comp->llc_stripe[i];
873 /* instantiated component */
874 info->lti_fid = *lu_object_fid(&object->do_lu);
876 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
878 if (cfs_fail_val == 0)
879 cfs_fail_val = info->lti_fid.f_oid;
881 info->lti_fid.f_oid = cfs_fail_val;
884 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
887 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
888 objs[i].l_ost_gen = cpu_to_le32(0);
889 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
892 rc = lod_fld_lookup(env, lod, &info->lti_fid,
895 CERROR("%s: Can not locate "DFID": rc = %d\n",
896 lod2obd(lod)->obd_name,
897 PFID(&info->lti_fid), rc);
900 } else if (lod_comp->llc_ostlist.op_array &&
901 lod_comp->llc_ostlist.op_count) {
902 /* user specified ost list */
903 ost_idx = lod_comp->llc_ostlist.op_array[i];
906 * with un-instantiated or with no specified ost list
907 * component, its l_ost_idx does not matter.
909 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
912 if (lmm_size != NULL)
913 *lmm_size = lov_mds_md_size(stripe_count, magic);
918 * Generate on-disk lov_mds_md structure based on the information in
919 * the lod_object->ldo_comp_entries.
921 * \param[in] env execution environment for this thread
922 * \param[in] lo LOD object
923 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
924 * \param[in|out] lmm_size buffer size/lmm size
925 * \param[in] is_dir generate lov ea for dir or file? For dir case,
926 * the stripe info is from the default stripe
927 * template, which is collected in lod_ah_init(),
928 * either from parent object or root object; for
929 * file case, it's from the @lo object
931 * \retval 0 if on disk structure is created successfully
932 * \retval negative error number on failure
934 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
935 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
937 struct lov_comp_md_entry_v1 *lcme;
938 struct lov_comp_md_v1 *lcm;
939 struct lod_layout_component *comp_entries;
940 __u16 comp_cnt, mirror_cnt;
942 int i, rc = 0, offset;
946 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
947 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
948 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
950 lo->ldo_def_striping->lds_def_striping_is_composite;
952 comp_cnt = lo->ldo_comp_cnt;
953 mirror_cnt = lo->ldo_mirror_count;
954 comp_entries = lo->ldo_comp_entries;
955 is_composite = lo->ldo_is_composite;
958 LASSERT(lmm_size != NULL);
959 LASSERT(comp_cnt != 0 && comp_entries != NULL);
962 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
966 lcm = (struct lov_comp_md_v1 *)lmm;
967 memset(lcm, 0, sizeof(*lcm));
969 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
970 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
971 lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
972 lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
974 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
975 LASSERT(offset % sizeof(__u64) == 0);
977 for (i = 0; i < comp_cnt; i++) {
978 struct lod_layout_component *lod_comp;
979 struct lov_mds_md *sub_md;
982 lod_comp = &comp_entries[i];
983 lcme = &lcm->lcm_entries[i];
985 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
986 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
988 /* component could be un-inistantiated */
989 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
990 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
991 lcme->lcme_timestamp =
992 cpu_to_le64(lod_comp->llc_timestamp);
993 lcme->lcme_extent.e_start =
994 cpu_to_le64(lod_comp->llc_extent.e_start);
995 lcme->lcme_extent.e_end =
996 cpu_to_le64(lod_comp->llc_extent.e_end);
997 lcme->lcme_offset = cpu_to_le32(offset);
999 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
1000 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
1003 lcme->lcme_size = cpu_to_le32(size);
1005 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
1006 "offset:%d lmm_size:%d\n", offset, *lmm_size);
1008 lcm->lcm_size = cpu_to_le32(offset);
1009 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
1011 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
1022 * Fill lti_ea_store buffer in the environment with a value for the given
1023 * EA. The buffer is reallocated if the value doesn't fit.
1025 * \param[in,out] env execution environment for this thread
1026 * .lti_ea_store buffer is filled with EA's value
1027 * \param[in] lo LOD object
1028 * \param[in] name name of the EA
1030 * \retval > 0 if EA is fetched successfully
1031 * \retval 0 if EA is empty
1032 * \retval negative error number on failure
1034 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
1037 struct lod_thread_info *info = lod_env_info(env);
1038 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1044 if (unlikely(info->lti_ea_store == NULL)) {
1045 /* just to enter in allocation block below */
1049 info->lti_buf.lb_buf = info->lti_ea_store;
1050 info->lti_buf.lb_len = info->lti_ea_store_size;
1051 rc = dt_xattr_get(env, next, &info->lti_buf, name);
1054 /* if object is not striped or inaccessible */
1055 if (rc == -ENODATA || rc == -ENOENT)
1058 if (rc == -ERANGE) {
1059 /* EA doesn't fit, reallocate new buffer */
1060 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
1061 if (rc == -ENODATA || rc == -ENOENT)
1067 rc = lod_ea_store_resize(info, rc);
1077 * Verify the target index is present in the current configuration.
1079 * \param[in] md LOD device where the target table is stored
1080 * \param[in] idx target's index
1082 * \retval 0 if the index is present
1083 * \retval -EINVAL if not
1085 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1087 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1088 !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
1089 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1090 md->lod_ost_descs.ltd_tgts_size);
1094 if (unlikely(OST_TGT(md, idx) == NULL)) {
1095 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1096 lod2obd(md)->obd_name, idx);
1100 if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
1101 CERROR("%s: invalid lod device, for idx: %d\n",
1102 lod2obd(md)->obd_name , idx);
1110 * Instantiate objects for stripes.
1112 * Allocate and initialize LU-objects representing the stripes. The number
1113 * of the stripes (ldo_stripe_count) must be initialized already. The caller
1114 * must ensure nobody else is calling the function on the object at the same
1115 * time. FLDB service must be running to be able to map a FID to the targets
1116 * and find appropriate device representing that target.
1118 * \param[in] env execution environment for this thread
1119 * \param[in,out] lo LOD object
1120 * \param[in] objs an array of IDs to creates the objects from
1121 * \param[in] comp_idx index of ldo_comp_entries
1123 * \retval 0 if the objects are instantiated successfully
1124 * \retval negative error number on failure
1126 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1127 struct lov_ost_data_v1 *objs, int comp_idx)
1129 struct lod_layout_component *lod_comp;
1130 struct lod_thread_info *info = lod_env_info(env);
1131 struct lod_device *md;
1132 struct lu_object *o, *n;
1133 struct lu_device *nd;
1134 struct dt_object **stripe = NULL;
1135 __u32 *ost_indices = NULL;
1141 LASSERT(lo != NULL);
1142 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1144 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1145 lod_comp = &lo->ldo_comp_entries[comp_idx];
1147 LASSERT(lod_comp->llc_stripe == NULL);
1148 LASSERT(lod_comp->llc_stripe_count > 0);
1149 LASSERT(lod_comp->llc_stripe_size > 0);
1151 stripe_len = lod_comp->llc_stripe_count;
1152 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1155 OBD_ALLOC(ost_indices, sizeof(*ost_indices) * stripe_len);
1157 GOTO(out, rc = -ENOMEM);
1159 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1160 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1163 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1164 idx = le32_to_cpu(objs[i].l_ost_idx);
1165 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1168 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1169 PFID(&info->lti_fid));
1170 lod_getref(&md->lod_ost_descs);
1172 rc = validate_lod_and_idx(md, idx);
1173 if (unlikely(rc != 0)) {
1174 lod_putref(md, &md->lod_ost_descs);
1178 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
1179 lod_putref(md, &md->lod_ost_descs);
1181 /* In the function below, .hs_keycmp resolves to
1182 * u_obj_hop_keycmp() */
1183 /* coverity[overrun-buffer-val] */
1184 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1186 GOTO(out, rc = PTR_ERR(o));
1188 n = lu_object_locate(o->lo_header, nd->ld_type);
1191 stripe[i] = container_of(n, struct dt_object, do_lu);
1192 ost_indices[i] = idx;
1197 for (i = 0; i < stripe_len; i++)
1198 if (stripe[i] != NULL)
1199 dt_object_put(env, stripe[i]);
1201 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1202 lod_comp->llc_stripe_count = 0;
1204 OBD_FREE(ost_indices,
1205 sizeof(*ost_indices) * stripe_len);
1207 lod_comp->llc_stripe = stripe;
1208 lod_comp->llc_ost_indices = ost_indices;
1209 lod_comp->llc_stripes_allocated = stripe_len;
1216 * Instantiate objects for striping.
1218 * Parse striping information in \a buf and instantiate the objects
1219 * representing the stripes.
1221 * \param[in] env execution environment for this thread
1222 * \param[in] lo LOD object
1223 * \param[in] buf buffer storing LOV EA to parse
1225 * \retval 0 if parsing and objects creation succeed
1226 * \retval negative error number on failure
1228 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1229 const struct lu_buf *buf)
1231 struct lov_mds_md_v1 *lmm;
1232 struct lov_comp_md_v1 *comp_v1 = NULL;
1233 struct lov_ost_data_v1 *objs;
1234 __u32 magic, pattern;
1237 __u16 mirror_cnt = 0;
1241 LASSERT(buf->lb_buf);
1242 LASSERT(buf->lb_len);
1243 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1245 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1246 magic = le32_to_cpu(lmm->lmm_magic);
1248 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1249 magic != LOV_MAGIC_COMP_V1)
1250 GOTO(out, rc = -EINVAL);
1252 lod_free_comp_entries(lo);
1254 if (magic == LOV_MAGIC_COMP_V1) {
1255 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1256 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1258 GOTO(out, rc = -EINVAL);
1259 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1260 lo->ldo_is_composite = 1;
1261 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1263 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1266 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1267 lo->ldo_is_composite = 0;
1270 rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1274 for (i = 0; i < comp_cnt; i++) {
1275 struct lod_layout_component *lod_comp;
1276 struct lu_extent *ext;
1279 lod_comp = &lo->ldo_comp_entries[i];
1280 if (lo->ldo_is_composite) {
1281 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1282 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1283 magic = le32_to_cpu(lmm->lmm_magic);
1285 ext = &comp_v1->lcm_entries[i].lcme_extent;
1286 lod_comp->llc_extent.e_start =
1287 le64_to_cpu(ext->e_start);
1288 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1289 lod_comp->llc_flags =
1290 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1291 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1292 lod_comp->llc_timestamp = le64_to_cpu(
1293 comp_v1->lcm_entries[i].lcme_timestamp);
1295 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1296 if (lod_comp->llc_id == LCME_ID_INVAL)
1297 GOTO(out, rc = -EINVAL);
1299 lod_comp_set_init(lod_comp);
1302 pattern = le32_to_cpu(lmm->lmm_pattern);
1303 if (lov_pattern(pattern) != LOV_PATTERN_RAID0 &&
1304 lov_pattern(pattern) != LOV_PATTERN_MDT)
1305 GOTO(out, rc = -EINVAL);
1307 lod_comp->llc_pattern = pattern;
1308 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1309 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1310 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1312 if (magic == LOV_MAGIC_V3) {
1313 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1314 lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1315 objs = &v3->lmm_objects[0];
1317 lod_set_pool(&lod_comp->llc_pool, NULL);
1318 objs = &lmm->lmm_objects[0];
1322 * If uninstantiated template component has valid l_ost_idx,
1323 * then user has specified ost list for this component.
1325 if (!lod_comp_inited(lod_comp)) {
1328 if (objs[0].l_ost_idx != (__u32)-1UL) {
1329 stripe_count = lod_comp_entry_stripe_count(
1330 lo, lod_comp, false);
1332 * load the user specified ost list, when this
1333 * component is instantiated later, it will be
1334 * used in lod_alloc_ost_list().
1336 lod_comp->llc_ostlist.op_count = stripe_count;
1337 lod_comp->llc_ostlist.op_size =
1338 stripe_count * sizeof(__u32);
1339 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1340 lod_comp->llc_ostlist.op_size);
1341 if (!lod_comp->llc_ostlist.op_array)
1342 GOTO(out, rc = -ENOMEM);
1344 for (j = 0; j < stripe_count; j++)
1345 lod_comp->llc_ostlist.op_array[j] =
1346 le32_to_cpu(objs[j].l_ost_idx);
1349 * this component OST objects starts from the
1350 * first ost_idx, lod_alloc_ost_list() will
1353 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1356 * for uninstantiated component,
1357 * lmm_layout_gen stores default stripe offset.
1359 lod_comp->llc_stripe_offset =
1360 lmm->lmm_layout_gen;
1364 /* skip un-instantiated component object initialization */
1365 if (!lod_comp_inited(lod_comp))
1368 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1369 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1370 rc = lod_initialize_objects(env, lo, objs, i);
1376 rc = lod_fill_mirrors(lo);
1382 lod_striping_free_nolock(env, lo);
1387 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1388 * is already cached.
1390 * \param[in] lo LOD object
1392 * \retval True if the striping is cached, otherwise
1395 static bool lod_striping_loaded(struct lod_object *lo)
1397 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1398 lo->ldo_comp_cached)
1401 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1402 if (lo->ldo_dir_stripe_loaded)
1405 /* Never load LMV stripe for slaves of striped dir */
1406 if (lo->ldo_dir_slave_stripe)
1414 * A generic function to initialize the stripe objects.
1416 * A protected version of lod_striping_load_locked() - load the striping
1417 * information from storage, parse that and instantiate LU objects to
1418 * represent the stripes. The LOD object \a lo supplies a pointer to the
1419 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1420 * return an array of references to the newly instantiated objects.
1422 * \param[in] env execution environment for this thread
1423 * \param[in,out] lo LOD object, where striping is stored and
1424 * which gets an array of references
1426 * \retval 0 if parsing and object creation succeed
1427 * \retval negative error number on failure
1429 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1431 struct lod_thread_info *info = lod_env_info(env);
1432 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1433 struct lu_buf *buf = &info->lti_buf;
1438 if (!dt_object_exists(next))
1441 if (lod_striping_loaded(lo))
1444 mutex_lock(&lo->ldo_layout_mutex);
1445 if (lod_striping_loaded(lo))
1446 GOTO(unlock, rc = 0);
1448 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1449 rc = lod_get_lov_ea(env, lo);
1454 * there is LOV EA (striping information) in this object
1455 * let's parse it and create in-core objects for the stripes
1457 buf->lb_buf = info->lti_ea_store;
1458 buf->lb_len = info->lti_ea_store_size;
1459 rc = lod_parse_striping(env, lo, buf);
1461 lo->ldo_comp_cached = 1;
1462 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1463 rc = lod_get_lmv_ea(env, lo);
1464 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1)) {
1465 /* Let's set stripe_loaded to avoid further
1466 * stripe loading especially for non-stripe directory,
1467 * which can hurt performance. (See LU-9840)
1470 lo->ldo_dir_stripe_loaded = 1;
1471 GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1473 buf->lb_buf = info->lti_ea_store;
1474 buf->lb_len = info->lti_ea_store_size;
1475 if (rc == sizeof(struct lmv_mds_md_v1)) {
1476 rc = lod_load_lmv_shards(env, lo, buf, true);
1477 if (buf->lb_buf != info->lti_ea_store) {
1478 OBD_FREE_LARGE(info->lti_ea_store,
1479 info->lti_ea_store_size);
1480 info->lti_ea_store = buf->lb_buf;
1481 info->lti_ea_store_size = buf->lb_len;
1489 * there is LMV EA (striping information) in this object
1490 * let's parse it and create in-core objects for the stripes
1492 rc = lod_parse_dir_striping(env, lo, buf);
1494 lo->ldo_dir_stripe_loaded = 1;
1498 mutex_unlock(&lo->ldo_layout_mutex);
1503 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1504 const struct lu_buf *buf)
1510 mutex_lock(&lo->ldo_layout_mutex);
1511 lod_striping_free_nolock(env, lo);
1512 rc = lod_parse_striping(env, lo, buf);
1513 mutex_unlock(&lo->ldo_layout_mutex);
1519 * Verify lov_user_md_v1/v3 striping.
1521 * Check the validity of all fields including the magic, stripe size,
1522 * stripe count, stripe offset and that the pool is present. Also check
1523 * that each target index points to an existing target. The additional
1524 * \a is_from_disk turns additional checks. In some cases zero fields
1525 * are allowed (like pattern=0).
1527 * \param[in] d LOD device
1528 * \param[in] buf buffer with LOV EA to verify
1529 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1530 * 1 - from disk, do not allow
1532 * \retval 0 if the striping is valid
1533 * \retval -EINVAL if striping is invalid
1535 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1538 struct lov_user_md_v1 *lum;
1539 struct lov_user_md_v3 *lum3;
1540 struct pool_desc *pool = NULL;
1544 __u16 stripe_offset;
1551 if (buf->lb_len < sizeof(*lum)) {
1552 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1554 GOTO(out, rc = -EINVAL);
1557 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1558 if (magic != LOV_USER_MAGIC_V1 &&
1559 magic != LOV_USER_MAGIC_V3 &&
1560 magic != LOV_USER_MAGIC_SPECIFIC) {
1561 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1562 le32_to_cpu(lum->lmm_magic));
1563 GOTO(out, rc = -EINVAL);
1566 /* the user uses "0" for default stripe pattern normally. */
1567 if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1568 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1570 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1571 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1572 le32_to_cpu(lum->lmm_pattern));
1573 GOTO(out, rc = -EINVAL);
1576 /* a released lum comes from creating orphan on hsm release,
1577 * doesn't make sense to verify it. */
1578 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1581 /* 64kB is the largest common page size we see (ia64), and matches the
1583 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1584 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1585 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1586 stripe_size, LOV_MIN_STRIPE_SIZE);
1587 GOTO(out, rc = -EINVAL);
1590 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1591 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1592 lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1593 /* if offset is not within valid range [0, osts_size) */
1594 if (stripe_offset >= d->lod_osts_size) {
1595 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1596 stripe_offset, d->lod_osts_size);
1597 GOTO(out, rc = -EINVAL);
1600 /* if lmm_stripe_offset is *not* in bitmap */
1601 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1602 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1604 GOTO(out, rc = -EINVAL);
1608 if (magic == LOV_USER_MAGIC_V1)
1609 lum_size = offsetof(struct lov_user_md_v1,
1611 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1612 lum_size = offsetof(struct lov_user_md_v3,
1615 GOTO(out, rc = -EINVAL);
1617 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1618 if (buf->lb_len < lum_size) {
1619 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1620 "magic %#x and stripe_count %u\n",
1621 buf->lb_len, lum_size, magic, stripe_count);
1622 GOTO(out, rc = -EINVAL);
1625 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1629 /* In the function below, .hs_keycmp resolves to
1630 * pool_hashkey_keycmp() */
1631 /* coverity[overrun-buffer-val] */
1632 pool = lod_find_pool(d, lum3->lmm_pool_name);
1636 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1637 rc = lod_check_index_in_pool(stripe_offset, pool);
1639 GOTO(out, rc = -EINVAL);
1642 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1643 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1644 stripe_count, pool_tgt_count(pool));
1645 GOTO(out, rc = -EINVAL);
1650 lod_pool_putref(pool);
1656 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1658 LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1659 LOV_USER_MAGIC_COMP_V1, "Wrong magic %x\n",
1660 le32_to_cpu(comp->lcm_magic));
1661 LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1662 "bad index %d, max = %d\n",
1663 i, le16_to_cpu(comp->lcm_entry_count));
1665 return &comp->lcm_entries[i];
1668 #define for_each_comp_entry_v1(comp, entry) \
1669 for (entry = comp_entry_v1(comp, 0); \
1670 entry <= comp_entry_v1(comp, \
1671 le16_to_cpu(comp->lcm_entry_count) - 1); \
1674 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1)
1676 struct lov_comp_md_entry_v1 *ent, *dom_ent;
1678 __u32 dom_off, dom_size, comp_size;
1679 void *blob_src, *blob_dst;
1680 unsigned int blob_size, blob_shift;
1682 entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1683 /* if file has only DoM stripe return just error */
1687 comp_size = le32_to_cpu(comp_v1->lcm_size);
1688 dom_ent = &comp_v1->lcm_entries[0];
1689 dom_off = le32_to_cpu(dom_ent->lcme_offset);
1690 dom_size = le32_to_cpu(dom_ent->lcme_size);
1692 /* shift entries array first */
1693 comp_v1->lcm_entry_count = cpu_to_le16(entries);
1694 memmove(dom_ent, dom_ent + 1,
1695 entries * sizeof(struct lov_comp_md_entry_v1));
1697 /* now move blob of layouts */
1698 blob_dst = (void *)comp_v1 + dom_off - sizeof(*dom_ent);
1699 blob_src = (void *)comp_v1 + dom_off + dom_size;
1700 blob_size = (unsigned long)((void *)comp_v1 + comp_size - blob_src);
1701 blob_shift = sizeof(*dom_ent) + dom_size;
1703 memmove(blob_dst, blob_src, blob_size);
1705 for_each_comp_entry_v1(comp_v1, ent) {
1708 off = le32_to_cpu(ent->lcme_offset);
1709 ent->lcme_offset = cpu_to_le32(off - blob_shift);
1712 comp_v1->lcm_size = cpu_to_le32(comp_size - blob_shift);
1714 /* notify a caller to re-check entry */
1718 int lod_fix_dom_stripe(struct lod_device *d, struct lov_comp_md_v1 *comp_v1)
1720 struct lov_comp_md_entry_v1 *ent, *dom_ent;
1721 struct lu_extent *dom_ext, *ext;
1722 struct lov_user_md_v1 *lum;
1727 dom_ent = &comp_v1->lcm_entries[0];
1728 dom_ext = &dom_ent->lcme_extent;
1729 dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1730 stripe_size = d->lod_dom_max_stripesize;
1732 lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1733 CDEBUG(D_LAYOUT, "DoM component size %u was bigger than MDT limit %u, "
1734 "new size is %u\n", le32_to_cpu(lum->lmm_stripe_size),
1735 d->lod_dom_max_stripesize, stripe_size);
1736 lum->lmm_stripe_size = cpu_to_le32(stripe_size);
1738 for_each_comp_entry_v1(comp_v1, ent) {
1742 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1746 ext = &ent->lcme_extent;
1747 if (ext->e_start != dom_ext->e_end)
1750 /* Found next component after the DoM one with the same
1751 * mirror_id and adjust its start with DoM component end.
1753 * NOTE: we are considering here that there can be only one
1754 * DoM component in a file, all replicas are located on OSTs
1755 * always and don't need adjustment since use own layouts.
1757 ext->e_start = cpu_to_le64(stripe_size);
1761 if (stripe_size == 0) {
1762 /* DoM component size is zero due to server setting,
1763 * remove it from the layout */
1764 rc = lod_erase_dom_stripe(comp_v1);
1766 /* Update DoM extent end finally */
1767 dom_ext->e_end = cpu_to_le64(stripe_size);
1774 * Verify LOV striping.
1776 * \param[in] d LOD device
1777 * \param[in] buf buffer with LOV EA to verify
1778 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1779 * 1 - from disk, do not allow
1780 * \param[in] start extent start for composite layout
1782 * \retval 0 if the striping is valid
1783 * \retval -EINVAL if striping is invalid
1785 int lod_verify_striping(struct lod_device *d, struct lod_object *lo,
1786 const struct lu_buf *buf, bool is_from_disk)
1788 struct lov_desc *desc = &d->lod_desc;
1789 struct lov_user_md_v1 *lum;
1790 struct lov_comp_md_v1 *comp_v1;
1791 struct lov_comp_md_entry_v1 *ent;
1792 struct lu_extent *ext;
1795 __u32 stripe_size = 0;
1796 __u16 prev_mid = -1, mirror_id = -1;
1804 if (buf->lb_len < sizeof(*lum)) {
1805 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1810 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1811 if (magic != LOV_USER_MAGIC_V1 &&
1812 magic != LOV_USER_MAGIC_V3 &&
1813 magic != LOV_USER_MAGIC_SPECIFIC &&
1814 magic != LOV_USER_MAGIC_COMP_V1) {
1815 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1816 le32_to_cpu(lum->lmm_magic));
1820 if (magic != LOV_USER_MAGIC_COMP_V1)
1821 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1823 /* magic == LOV_USER_MAGIC_COMP_V1 */
1824 comp_v1 = buf->lb_buf;
1825 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1826 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1827 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1833 if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1834 CDEBUG(D_LAYOUT, "entry count is zero\n");
1838 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1839 lo->ldo_comp_cnt > 0) {
1840 /* could be called from lustre.lov.add */
1841 __u32 cnt = lo->ldo_comp_cnt;
1843 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
1844 prev_end = ext->e_end;
1849 for_each_comp_entry_v1(comp_v1, ent) {
1850 ext = &ent->lcme_extent;
1852 if (le64_to_cpu(ext->e_start) >= le64_to_cpu(ext->e_end)) {
1853 CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
1854 le64_to_cpu(ext->e_start),
1855 le64_to_cpu(ext->e_end));
1860 /* lcme_id contains valid value */
1861 if (le32_to_cpu(ent->lcme_id) == 0 ||
1862 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
1863 CDEBUG(D_LAYOUT, "invalid id %u\n",
1864 le32_to_cpu(ent->lcme_id));
1868 if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
1869 mirror_id = mirror_id_of(
1870 le32_to_cpu(ent->lcme_id));
1872 /* first component must start with 0 */
1873 if (mirror_id != prev_mid &&
1874 le64_to_cpu(ext->e_start) != 0) {
1876 "invalid start:%llu, expect:0\n",
1877 le64_to_cpu(ext->e_start));
1881 prev_mid = mirror_id;
1885 if (le64_to_cpu(ext->e_start) == 0) {
1890 /* the next must be adjacent with the previous one */
1891 if (le64_to_cpu(ext->e_start) != prev_end) {
1893 "invalid start actual:%llu, expect:%llu\n",
1894 le64_to_cpu(ext->e_start), prev_end);
1898 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
1899 tmp.lb_len = le32_to_cpu(ent->lcme_size);
1901 /* Check DoM entry is always the first one */
1903 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
1905 /* DoM component can be only the first stripe */
1906 if (le64_to_cpu(ext->e_start) > 0) {
1907 CDEBUG(D_LAYOUT, "invalid DoM component "
1908 "with %llu extent start\n",
1909 le64_to_cpu(ext->e_start));
1912 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1913 /* There is just one stripe on MDT and it must
1914 * cover whole component size. */
1915 if (stripe_size != le64_to_cpu(ext->e_end)) {
1916 CDEBUG(D_LAYOUT, "invalid DoM layout "
1917 "stripe size %u != %llu "
1918 "(component size)\n",
1919 stripe_size, prev_end);
1922 /* Check stripe size againts per-MDT limit */
1923 if (stripe_size > d->lod_dom_max_stripesize) {
1924 CDEBUG(D_LAYOUT, "DoM component size "
1925 "%u is bigger than MDT limit %u, check "
1926 "dom_max_stripesize parameter\n",
1927 stripe_size, d->lod_dom_max_stripesize);
1928 rc = lod_fix_dom_stripe(d, comp_v1);
1929 if (rc == -ERESTART) {
1930 /* DoM entry was removed, re-check
1931 * new layout from start */
1937 /* Any stripe count is forbidden on DoM component */
1938 if (lum->lmm_stripe_count) {
1940 "invalid DoM layout stripe count %u, must be 0\n",
1941 le16_to_cpu(lum->lmm_stripe_count));
1945 /* Any pool is forbidden on DoM component */
1946 if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
1947 struct lov_user_md_v3 *v3 = (void *)lum;
1949 if (v3->lmm_pool_name[0] != '\0') {
1951 "DoM component cannot have pool assigned\n");
1957 prev_end = le64_to_cpu(ext->e_end);
1959 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
1963 if (prev_end == LUSTRE_EOF)
1966 /* extent end must be aligned with the stripe_size */
1967 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1968 if (stripe_size == 0)
1969 stripe_size = desc->ld_default_stripe_size;
1970 if (stripe_size == 0 || (prev_end & (stripe_size - 1))) {
1971 CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
1972 "stripe_sz: %u, [%llu, %llu)\n",
1973 stripe_size, ext->e_start, prev_end);
1978 /* make sure that the mirror_count is telling the truth */
1979 if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
1986 * set the default stripe size, if unset.
1988 * \param[in,out] val number of bytes per OST stripe
1990 * The minimum stripe size is 64KB to ensure that a single stripe is an
1991 * even multiple of a client PAGE_SIZE (IA64, PPC, etc). Otherwise, it
1992 * is difficult to split dirty pages across OSCs during writes.
1994 void lod_fix_desc_stripe_size(__u64 *val)
1996 if (*val < LOV_MIN_STRIPE_SIZE) {
1998 LCONSOLE_INFO("Increasing default stripe size to "
1999 "minimum value %u\n",
2000 LOV_DESC_STRIPE_SIZE_DEFAULT);
2001 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
2002 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
2003 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
2004 LCONSOLE_WARN("Changing default stripe size to %llu (a "
2005 "multiple of %u)\n",
2006 *val, LOV_MIN_STRIPE_SIZE);
2011 * set the filesystem default number of stripes, if unset.
2013 * \param[in,out] val number of stripes
2015 * A value of "0" means "use the system-wide default stripe count", which
2016 * has either been inherited by now, or falls back to 1 stripe per file.
2017 * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
2018 * and is a valid value, so is left unchanged here.
2020 void lod_fix_desc_stripe_count(__u32 *val)
2027 * set the filesystem default layout pattern
2029 * \param[in,out] val LOV_PATTERN_* layout
2031 * A value of "0" means "use the system-wide default layout type", which
2032 * has either been inherited by now, or falls back to plain RAID0 striping.
2034 void lod_fix_desc_pattern(__u32 *val)
2036 /* from lov_setstripe */
2037 if ((*val != 0) && (*val != LOV_PATTERN_RAID0) &&
2038 (*val != LOV_PATTERN_MDT)) {
2039 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
2044 void lod_fix_desc_qos_maxage(__u32 *val)
2046 /* fix qos_maxage */
2048 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2052 * Used to fix insane default striping.
2054 * \param[in] desc striping description
2056 void lod_fix_desc(struct lov_desc *desc)
2058 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2059 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2060 lod_fix_desc_pattern(&desc->ld_pattern);
2061 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2065 * Initialize the structures used to store pools and default striping.
2067 * \param[in] lod LOD device
2068 * \param[in] lcfg configuration structure storing default striping.
2070 * \retval 0 if initialization succeeds
2071 * \retval negative error number on failure
2073 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2075 struct obd_device *obd;
2076 struct lov_desc *desc;
2080 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2081 LASSERT(obd != NULL);
2082 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2084 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2085 CERROR("LOD setup requires a descriptor\n");
2089 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2091 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2092 CERROR("descriptor size wrong: %d > %d\n",
2093 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2097 if (desc->ld_magic != LOV_DESC_MAGIC) {
2098 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2099 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2100 obd->obd_name, desc);
2101 lustre_swab_lov_desc(desc);
2103 CERROR("%s: Bad lov desc magic: %#x\n",
2104 obd->obd_name, desc->ld_magic);
2111 desc->ld_active_tgt_count = 0;
2112 lod->lod_desc = *desc;
2114 lod->lod_sp_me = LUSTRE_SP_CLI;
2116 /* Set up allocation policy (QoS and RR) */
2117 INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
2118 init_rwsem(&lod->lod_qos.lq_rw_sem);
2119 lod->lod_qos.lq_dirty = 1;
2120 lod->lod_qos.lq_rr.lqr_dirty = 1;
2121 lod->lod_qos.lq_reset = 1;
2122 /* Default priority is toward free space balance */
2123 lod->lod_qos.lq_prio_free = 232;
2124 /* Default threshold for rr (roughly 17%) */
2125 lod->lod_qos.lq_threshold_rr = 43;
2127 /* Set up OST pool environment */
2128 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
2129 HASH_POOLS_MAX_BITS,
2130 HASH_POOLS_BKT_BITS, 0,
2133 &pool_hash_operations,
2135 if (lod->lod_pools_hash_body == NULL)
2138 INIT_LIST_HEAD(&lod->lod_pool_list);
2139 lod->lod_pool_count = 0;
2140 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
2143 lod_qos_rr_init(&lod->lod_qos.lq_rr);
2144 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
2146 GOTO(out_pool_info, rc);
2151 lod_ost_pool_free(&lod->lod_pool_info);
2153 cfs_hash_putref(lod->lod_pools_hash_body);
2159 * Release the structures describing the pools.
2161 * \param[in] lod LOD device from which we release the structures
2165 int lod_pools_fini(struct lod_device *lod)
2167 struct obd_device *obd = lod2obd(lod);
2168 struct pool_desc *pool, *tmp;
2171 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2172 /* free pool structs */
2173 CDEBUG(D_INFO, "delete pool %p\n", pool);
2174 /* In the function below, .hs_keycmp resolves to
2175 * pool_hashkey_keycmp() */
2176 /* coverity[overrun-buffer-val] */
2177 lod_pool_del(obd, pool->pool_name);
2180 cfs_hash_putref(lod->lod_pools_hash_body);
2181 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
2182 lod_ost_pool_free(&lod->lod_pool_info);