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);
604 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
605 if (info->lti_ea_store) {
606 LASSERT(info->lti_ea_store_size);
607 LASSERT(info->lti_ea_store_size < round);
608 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
609 info->lti_ea_store_size, round);
610 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
611 info->lti_ea_store = NULL;
612 info->lti_ea_store_size = 0;
615 OBD_ALLOC_LARGE(info->lti_ea_store, round);
616 if (info->lti_ea_store == NULL)
618 info->lti_ea_store_size = round;
623 static void lod_free_comp_buffer(struct lod_layout_component *entries,
624 __u16 count, __u32 bufsize)
626 struct lod_layout_component *entry;
629 for (i = 0; i < count; i++) {
631 if (entry->llc_pool != NULL)
632 lod_set_pool(&entry->llc_pool, NULL);
633 if (entry->llc_ostlist.op_array)
634 OBD_FREE(entry->llc_ostlist.op_array,
635 entry->llc_ostlist.op_size);
636 LASSERT(entry->llc_stripe == NULL);
637 LASSERT(entry->llc_stripes_allocated == 0);
641 OBD_FREE_LARGE(entries, bufsize);
644 void lod_free_def_comp_entries(struct lod_default_striping *lds)
646 lod_free_comp_buffer(lds->lds_def_comp_entries,
647 lds->lds_def_comp_size_cnt,
649 sizeof(*lds->lds_def_comp_entries) *
650 lds->lds_def_comp_size_cnt));
651 lds->lds_def_comp_entries = NULL;
652 lds->lds_def_comp_cnt = 0;
653 lds->lds_def_striping_is_composite = 0;
654 lds->lds_def_comp_size_cnt = 0;
658 * Resize per-thread storage to hold default striping component entries
660 * A helper function to resize per-thread temporary storage. This storage
661 * is used to hold default LOV/LVM EAs and may be quite large. We do not want
662 * to allocate/release it every time, so instead we put it into the env and
663 * reallocate it on demand. The memory is released when the correspondent
664 * thread is finished.
666 * \param[in,out] lds default striping
667 * \param[in] count new component count to grow the buffer to
669 * \retval 0 on success, -ENOMEM if reallocation failed
671 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
673 struct lod_layout_component *entries;
674 __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
676 __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
677 lds->lds_def_comp_size_cnt);
682 OBD_ALLOC_LARGE(entries, new);
686 if (lds->lds_def_comp_entries != NULL) {
687 CDEBUG(D_INFO, "default striping component size %d is not "
688 "enough, need %d\n", old, new);
689 lod_free_def_comp_entries(lds);
692 lds->lds_def_comp_entries = entries;
693 lds->lds_def_comp_size_cnt = count;
698 void lod_free_comp_entries(struct lod_object *lo)
700 if (lo->ldo_mirrors) {
701 OBD_FREE(lo->ldo_mirrors,
702 sizeof(*lo->ldo_mirrors) * lo->ldo_mirror_count);
703 lo->ldo_mirrors = NULL;
704 lo->ldo_mirror_count = 0;
706 lod_free_comp_buffer(lo->ldo_comp_entries,
708 sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
709 lo->ldo_comp_entries = NULL;
710 lo->ldo_comp_cnt = 0;
711 lo->ldo_is_composite = 0;
714 int lod_alloc_comp_entries(struct lod_object *lo,
715 int mirror_count, int comp_count)
717 LASSERT(comp_count != 0);
718 LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
720 if (mirror_count > 0) {
721 OBD_ALLOC(lo->ldo_mirrors,
722 sizeof(*lo->ldo_mirrors) * mirror_count);
723 if (!lo->ldo_mirrors)
726 lo->ldo_mirror_count = mirror_count;
729 OBD_ALLOC_LARGE(lo->ldo_comp_entries,
730 sizeof(*lo->ldo_comp_entries) * comp_count);
731 if (lo->ldo_comp_entries == NULL) {
732 OBD_FREE(lo->ldo_mirrors,
733 sizeof(*lo->ldo_mirrors) * mirror_count);
734 lo->ldo_mirror_count = 0;
738 lo->ldo_comp_cnt = comp_count;
742 int lod_fill_mirrors(struct lod_object *lo)
744 struct lod_layout_component *lod_comp;
746 __u16 mirror_id = 0xffff;
750 LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
752 if (!lo->ldo_is_composite)
755 lod_comp = &lo->ldo_comp_entries[0];
756 for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
757 int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
758 int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
760 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
761 lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
762 lo->ldo_mirrors[mirror_idx].lme_primary |= preferred;
763 lo->ldo_mirrors[mirror_idx].lme_end = i;
769 if (mirror_idx >= lo->ldo_mirror_count)
772 mirror_id = mirror_id_of(lod_comp->llc_id);
774 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
775 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
776 lo->ldo_mirrors[mirror_idx].lme_primary = preferred;
777 lo->ldo_mirrors[mirror_idx].lme_start = i;
778 lo->ldo_mirrors[mirror_idx].lme_end = i;
780 if (mirror_idx != lo->ldo_mirror_count - 1)
787 * Generate on-disk lov_mds_md structure for each layout component based on
788 * the information in lod_object->ldo_comp_entries[i].
790 * \param[in] env execution environment for this thread
791 * \param[in] lo LOD object
792 * \param[in] comp_idx index of ldo_comp_entries
793 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
794 * \param[in|out] lmm_size buffer size/lmm size
795 * \param[in] is_dir generate lov ea for dir or file? For dir case,
796 * the stripe info is from the default stripe
797 * template, which is collected in lod_ah_init(),
798 * either from parent object or root object; for
799 * file case, it's from the @lo object
801 * \retval 0 if on disk structure is created successfully
802 * \retval negative error number on failure
804 static int lod_gen_component_ea(const struct lu_env *env,
805 struct lod_object *lo, int comp_idx,
806 struct lov_mds_md *lmm, int *lmm_size,
809 struct lod_thread_info *info = lod_env_info(env);
810 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
811 struct lod_device *lod;
812 struct lov_ost_data_v1 *objs;
813 struct lod_layout_component *lod_comp;
822 &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
824 lod_comp = &lo->ldo_comp_entries[comp_idx];
826 magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
827 if (lod_comp->llc_pattern == 0) /* default striping */
828 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
830 lmm->lmm_magic = cpu_to_le32(magic);
831 lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
832 fid_to_lmm_oi(fid, &lmm->lmm_oi);
833 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
834 lmm->lmm_oi.oi.oi_id++;
835 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
837 lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
838 lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
840 * for dir and uninstantiated component, lmm_layout_gen stores
841 * default stripe offset.
843 lmm->lmm_layout_gen =
844 (is_dir || !lod_comp_inited(lod_comp)) ?
845 cpu_to_le16(lod_comp->llc_stripe_offset) :
846 cpu_to_le16(lod_comp->llc_layout_gen);
848 if (magic == LOV_MAGIC_V1) {
849 objs = &lmm->lmm_objects[0];
851 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
852 size_t cplen = strlcpy(v3->lmm_pool_name,
854 sizeof(v3->lmm_pool_name));
855 if (cplen >= sizeof(v3->lmm_pool_name))
857 objs = &v3->lmm_objects[0];
859 stripe_count = lod_comp_entry_stripe_count(lo, lod_comp, is_dir);
860 if (stripe_count == 0 && !is_dir &&
861 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
862 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
865 if (!is_dir && lo->ldo_is_composite)
866 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
868 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
871 /* generate ost_idx of this component stripe */
872 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
873 for (i = 0; i < stripe_count; i++) {
874 struct dt_object *object;
875 __u32 ost_idx = (__u32)-1UL;
876 int type = LU_SEQ_RANGE_OST;
878 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
879 object = lod_comp->llc_stripe[i];
880 /* instantiated component */
881 info->lti_fid = *lu_object_fid(&object->do_lu);
883 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
885 if (cfs_fail_val == 0)
886 cfs_fail_val = info->lti_fid.f_oid;
888 info->lti_fid.f_oid = cfs_fail_val;
891 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
894 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
895 objs[i].l_ost_gen = cpu_to_le32(0);
896 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
899 rc = lod_fld_lookup(env, lod, &info->lti_fid,
902 CERROR("%s: Can not locate "DFID": rc = %d\n",
903 lod2obd(lod)->obd_name,
904 PFID(&info->lti_fid), rc);
907 } else if (lod_comp->llc_ostlist.op_array &&
908 lod_comp->llc_ostlist.op_count) {
909 /* user specified ost list */
910 ost_idx = lod_comp->llc_ostlist.op_array[i];
913 * with un-instantiated or with no specified ost list
914 * component, its l_ost_idx does not matter.
916 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
919 if (lmm_size != NULL)
920 *lmm_size = lov_mds_md_size(stripe_count, magic);
925 * Generate on-disk lov_mds_md structure based on the information in
926 * the lod_object->ldo_comp_entries.
928 * \param[in] env execution environment for this thread
929 * \param[in] lo LOD object
930 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
931 * \param[in|out] lmm_size buffer size/lmm size
932 * \param[in] is_dir generate lov ea for dir or file? For dir case,
933 * the stripe info is from the default stripe
934 * template, which is collected in lod_ah_init(),
935 * either from parent object or root object; for
936 * file case, it's from the @lo object
938 * \retval 0 if on disk structure is created successfully
939 * \retval negative error number on failure
941 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
942 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
944 struct lov_comp_md_entry_v1 *lcme;
945 struct lov_comp_md_v1 *lcm;
946 struct lod_layout_component *comp_entries;
947 __u16 comp_cnt, mirror_cnt;
948 bool is_composite, is_foreign = false;
949 int i, rc = 0, offset;
953 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
954 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
955 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
957 lo->ldo_def_striping->lds_def_striping_is_composite;
959 comp_cnt = lo->ldo_comp_cnt;
960 mirror_cnt = lo->ldo_mirror_count;
961 comp_entries = lo->ldo_comp_entries;
962 is_composite = lo->ldo_is_composite;
963 is_foreign = lo->ldo_is_foreign;
966 LASSERT(lmm_size != NULL);
969 struct lov_foreign_md *lfm;
971 lfm = (struct lov_foreign_md *)lmm;
972 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
973 /* need to store little-endian */
974 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
975 __swab32s(&lfm->lfm_magic);
976 __swab32s(&lfm->lfm_length);
977 __swab32s(&lfm->lfm_type);
978 __swab32s(&lfm->lfm_flags);
980 *lmm_size = lo->ldo_foreign_lov_size;
984 LASSERT(comp_cnt != 0 && comp_entries != NULL);
987 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
991 lcm = (struct lov_comp_md_v1 *)lmm;
992 memset(lcm, 0, sizeof(*lcm));
994 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
995 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
996 lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
997 lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
999 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
1000 LASSERT(offset % sizeof(__u64) == 0);
1002 for (i = 0; i < comp_cnt; i++) {
1003 struct lod_layout_component *lod_comp;
1004 struct lov_mds_md *sub_md;
1007 lod_comp = &comp_entries[i];
1008 lcme = &lcm->lcm_entries[i];
1010 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
1011 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
1013 /* component could be un-inistantiated */
1014 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
1015 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1016 lcme->lcme_timestamp =
1017 cpu_to_le64(lod_comp->llc_timestamp);
1018 lcme->lcme_extent.e_start =
1019 cpu_to_le64(lod_comp->llc_extent.e_start);
1020 lcme->lcme_extent.e_end =
1021 cpu_to_le64(lod_comp->llc_extent.e_end);
1022 lcme->lcme_offset = cpu_to_le32(offset);
1024 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
1025 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
1028 lcme->lcme_size = cpu_to_le32(size);
1030 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
1031 "offset:%d lmm_size:%d\n", offset, *lmm_size);
1033 lcm->lcm_size = cpu_to_le32(offset);
1034 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
1036 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
1047 * Fill lti_ea_store buffer in the environment with a value for the given
1048 * EA. The buffer is reallocated if the value doesn't fit.
1050 * \param[in,out] env execution environment for this thread
1051 * .lti_ea_store buffer is filled with EA's value
1052 * \param[in] lo LOD object
1053 * \param[in] name name of the EA
1055 * \retval > 0 if EA is fetched successfully
1056 * \retval 0 if EA is empty
1057 * \retval negative error number on failure
1059 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
1062 struct lod_thread_info *info = lod_env_info(env);
1063 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1069 if (unlikely(info->lti_ea_store == NULL)) {
1070 /* just to enter in allocation block below */
1074 info->lti_buf.lb_buf = info->lti_ea_store;
1075 info->lti_buf.lb_len = info->lti_ea_store_size;
1076 rc = dt_xattr_get(env, next, &info->lti_buf, name);
1079 /* if object is not striped or inaccessible */
1080 if (rc == -ENODATA || rc == -ENOENT)
1083 if (rc == -ERANGE) {
1084 /* EA doesn't fit, reallocate new buffer */
1085 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
1086 if (rc == -ENODATA || rc == -ENOENT)
1092 rc = lod_ea_store_resize(info, rc);
1102 * Verify the target index is present in the current configuration.
1104 * \param[in] md LOD device where the target table is stored
1105 * \param[in] idx target's index
1107 * \retval 0 if the index is present
1108 * \retval -EINVAL if not
1110 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1112 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1113 !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
1114 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1115 md->lod_ost_descs.ltd_tgts_size);
1119 if (unlikely(OST_TGT(md, idx) == NULL)) {
1120 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1121 lod2obd(md)->obd_name, idx);
1125 if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
1126 CERROR("%s: invalid lod device, for idx: %d\n",
1127 lod2obd(md)->obd_name , idx);
1135 * Instantiate objects for stripes.
1137 * Allocate and initialize LU-objects representing the stripes. The number
1138 * of the stripes (ldo_stripe_count) must be initialized already. The caller
1139 * must ensure nobody else is calling the function on the object at the same
1140 * time. FLDB service must be running to be able to map a FID to the targets
1141 * and find appropriate device representing that target.
1143 * \param[in] env execution environment for this thread
1144 * \param[in,out] lo LOD object
1145 * \param[in] objs an array of IDs to creates the objects from
1146 * \param[in] comp_idx index of ldo_comp_entries
1148 * \retval 0 if the objects are instantiated successfully
1149 * \retval negative error number on failure
1151 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1152 struct lov_ost_data_v1 *objs, int comp_idx)
1154 struct lod_layout_component *lod_comp;
1155 struct lod_thread_info *info = lod_env_info(env);
1156 struct lod_device *md;
1157 struct lu_object *o, *n;
1158 struct lu_device *nd;
1159 struct dt_object **stripe = NULL;
1160 __u32 *ost_indices = NULL;
1166 LASSERT(lo != NULL);
1167 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1169 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1170 lod_comp = &lo->ldo_comp_entries[comp_idx];
1172 LASSERT(lod_comp->llc_stripe == NULL);
1173 LASSERT(lod_comp->llc_stripe_count > 0);
1174 LASSERT(lod_comp->llc_stripe_size > 0);
1176 stripe_len = lod_comp->llc_stripe_count;
1177 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1180 OBD_ALLOC(ost_indices, sizeof(*ost_indices) * stripe_len);
1182 GOTO(out, rc = -ENOMEM);
1184 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1185 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1188 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1189 idx = le32_to_cpu(objs[i].l_ost_idx);
1190 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1193 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1194 PFID(&info->lti_fid));
1195 lod_getref(&md->lod_ost_descs);
1197 rc = validate_lod_and_idx(md, idx);
1198 if (unlikely(rc != 0)) {
1199 lod_putref(md, &md->lod_ost_descs);
1203 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
1204 lod_putref(md, &md->lod_ost_descs);
1206 /* In the function below, .hs_keycmp resolves to
1207 * u_obj_hop_keycmp() */
1208 /* coverity[overrun-buffer-val] */
1209 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1211 GOTO(out, rc = PTR_ERR(o));
1213 n = lu_object_locate(o->lo_header, nd->ld_type);
1216 stripe[i] = container_of(n, struct dt_object, do_lu);
1217 ost_indices[i] = idx;
1222 for (i = 0; i < stripe_len; i++)
1223 if (stripe[i] != NULL)
1224 dt_object_put(env, stripe[i]);
1226 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1227 lod_comp->llc_stripe_count = 0;
1229 OBD_FREE(ost_indices,
1230 sizeof(*ost_indices) * stripe_len);
1232 lod_comp->llc_stripe = stripe;
1233 lod_comp->llc_ost_indices = ost_indices;
1234 lod_comp->llc_stripes_allocated = stripe_len;
1241 * Instantiate objects for striping.
1243 * Parse striping information in \a buf and instantiate the objects
1244 * representing the stripes.
1246 * \param[in] env execution environment for this thread
1247 * \param[in] lo LOD object
1248 * \param[in] buf buffer storing LOV EA to parse
1250 * \retval 0 if parsing and objects creation succeed
1251 * \retval negative error number on failure
1253 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1254 const struct lu_buf *buf)
1256 struct lov_mds_md_v1 *lmm;
1257 struct lov_comp_md_v1 *comp_v1 = NULL;
1258 struct lov_foreign_md *foreign = NULL;
1259 struct lov_ost_data_v1 *objs;
1260 __u32 magic, pattern;
1263 __u16 mirror_cnt = 0;
1267 LASSERT(buf->lb_buf);
1268 LASSERT(buf->lb_len);
1269 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1271 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1272 magic = le32_to_cpu(lmm->lmm_magic);
1274 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1275 magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN)
1276 GOTO(out, rc = -EINVAL);
1278 if (lo->ldo_is_foreign)
1279 lod_free_foreign_lov(lo);
1281 lod_free_comp_entries(lo);
1283 if (magic == LOV_MAGIC_COMP_V1) {
1284 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1285 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1287 GOTO(out, rc = -EINVAL);
1288 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1289 lo->ldo_is_composite = 1;
1290 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1292 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1293 } else if (magic == LOV_MAGIC_FOREIGN) {
1296 foreign = (struct lov_foreign_md *)buf->lb_buf;
1297 length = offsetof(typeof(*foreign), lfm_value);
1298 if (buf->lb_len < length ||
1299 buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1301 "buf len %zu too small for lov_foreign_md\n",
1303 GOTO(out, rc = -EINVAL);
1306 /* just cache foreign LOV EA raw */
1307 rc = lod_alloc_foreign_lov(lo, length);
1310 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1314 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1315 lo->ldo_is_composite = 0;
1318 rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1322 for (i = 0; i < comp_cnt; i++) {
1323 struct lod_layout_component *lod_comp;
1324 struct lu_extent *ext;
1327 lod_comp = &lo->ldo_comp_entries[i];
1328 if (lo->ldo_is_composite) {
1329 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1330 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1331 magic = le32_to_cpu(lmm->lmm_magic);
1333 ext = &comp_v1->lcm_entries[i].lcme_extent;
1334 lod_comp->llc_extent.e_start =
1335 le64_to_cpu(ext->e_start);
1336 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1337 lod_comp->llc_flags =
1338 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1339 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1340 lod_comp->llc_timestamp = le64_to_cpu(
1341 comp_v1->lcm_entries[i].lcme_timestamp);
1343 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1344 if (lod_comp->llc_id == LCME_ID_INVAL)
1345 GOTO(out, rc = -EINVAL);
1347 lod_comp_set_init(lod_comp);
1350 pattern = le32_to_cpu(lmm->lmm_pattern);
1351 if (lov_pattern(pattern) != LOV_PATTERN_RAID0 &&
1352 lov_pattern(pattern) != LOV_PATTERN_MDT)
1353 GOTO(out, rc = -EINVAL);
1355 lod_comp->llc_pattern = pattern;
1356 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1357 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1358 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1360 if (magic == LOV_MAGIC_V3) {
1361 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1362 lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1363 objs = &v3->lmm_objects[0];
1365 lod_set_pool(&lod_comp->llc_pool, NULL);
1366 objs = &lmm->lmm_objects[0];
1370 * If uninstantiated template component has valid l_ost_idx,
1371 * then user has specified ost list for this component.
1373 if (!lod_comp_inited(lod_comp)) {
1376 if (objs[0].l_ost_idx != (__u32)-1UL) {
1377 stripe_count = lod_comp_entry_stripe_count(
1378 lo, lod_comp, false);
1379 if (stripe_count == 0 &&
1380 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1381 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1382 GOTO(out, rc = -E2BIG);
1384 * load the user specified ost list, when this
1385 * component is instantiated later, it will be
1386 * used in lod_alloc_ost_list().
1388 lod_comp->llc_ostlist.op_count = stripe_count;
1389 lod_comp->llc_ostlist.op_size =
1390 stripe_count * sizeof(__u32);
1391 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1392 lod_comp->llc_ostlist.op_size);
1393 if (!lod_comp->llc_ostlist.op_array)
1394 GOTO(out, rc = -ENOMEM);
1396 for (j = 0; j < stripe_count; j++)
1397 lod_comp->llc_ostlist.op_array[j] =
1398 le32_to_cpu(objs[j].l_ost_idx);
1401 * this component OST objects starts from the
1402 * first ost_idx, lod_alloc_ost_list() will
1405 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1408 * for uninstantiated component,
1409 * lmm_layout_gen stores default stripe offset.
1411 lod_comp->llc_stripe_offset =
1412 lmm->lmm_layout_gen;
1416 /* skip un-instantiated component object initialization */
1417 if (!lod_comp_inited(lod_comp))
1420 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1421 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1422 rc = lod_initialize_objects(env, lo, objs, i);
1428 rc = lod_fill_mirrors(lo);
1434 lod_striping_free_nolock(env, lo);
1439 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1440 * is already cached.
1442 * \param[in] lo LOD object
1444 * \retval True if the striping is cached, otherwise
1447 static bool lod_striping_loaded(struct lod_object *lo)
1449 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1450 lo->ldo_comp_cached)
1453 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1454 if (lo->ldo_dir_stripe_loaded)
1457 /* Never load LMV stripe for slaves of striped dir */
1458 if (lo->ldo_dir_slave_stripe)
1466 * A generic function to initialize the stripe objects.
1468 * A protected version of lod_striping_load_locked() - load the striping
1469 * information from storage, parse that and instantiate LU objects to
1470 * represent the stripes. The LOD object \a lo supplies a pointer to the
1471 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1472 * return an array of references to the newly instantiated objects.
1474 * \param[in] env execution environment for this thread
1475 * \param[in,out] lo LOD object, where striping is stored and
1476 * which gets an array of references
1478 * \retval 0 if parsing and object creation succeed
1479 * \retval negative error number on failure
1481 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1483 struct lod_thread_info *info = lod_env_info(env);
1484 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1485 struct lu_buf *buf = &info->lti_buf;
1490 if (!dt_object_exists(next))
1493 if (lod_striping_loaded(lo))
1496 mutex_lock(&lo->ldo_layout_mutex);
1497 if (lod_striping_loaded(lo))
1498 GOTO(unlock, rc = 0);
1500 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1501 rc = lod_get_lov_ea(env, lo);
1506 * there is LOV EA (striping information) in this object
1507 * let's parse it and create in-core objects for the stripes
1509 buf->lb_buf = info->lti_ea_store;
1510 buf->lb_len = info->lti_ea_store_size;
1511 rc = lod_parse_striping(env, lo, buf);
1513 lo->ldo_comp_cached = 1;
1514 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1515 rc = lod_get_lmv_ea(env, lo);
1516 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1)) {
1517 /* Let's set stripe_loaded to avoid further
1518 * stripe loading especially for non-stripe directory,
1519 * which can hurt performance. (See LU-9840)
1522 lo->ldo_dir_stripe_loaded = 1;
1523 GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1525 buf->lb_buf = info->lti_ea_store;
1526 buf->lb_len = info->lti_ea_store_size;
1527 if (rc == sizeof(struct lmv_mds_md_v1)) {
1528 rc = lod_load_lmv_shards(env, lo, buf, true);
1529 if (buf->lb_buf != info->lti_ea_store) {
1530 OBD_FREE_LARGE(info->lti_ea_store,
1531 info->lti_ea_store_size);
1532 info->lti_ea_store = buf->lb_buf;
1533 info->lti_ea_store_size = buf->lb_len;
1541 * there is LMV EA (striping information) in this object
1542 * let's parse it and create in-core objects for the stripes
1544 rc = lod_parse_dir_striping(env, lo, buf);
1546 lo->ldo_dir_stripe_loaded = 1;
1550 mutex_unlock(&lo->ldo_layout_mutex);
1555 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1556 const struct lu_buf *buf)
1562 mutex_lock(&lo->ldo_layout_mutex);
1563 lod_striping_free_nolock(env, lo);
1564 rc = lod_parse_striping(env, lo, buf);
1565 mutex_unlock(&lo->ldo_layout_mutex);
1571 * Verify lov_user_md_v1/v3 striping.
1573 * Check the validity of all fields including the magic, stripe size,
1574 * stripe count, stripe offset and that the pool is present. Also check
1575 * that each target index points to an existing target. The additional
1576 * \a is_from_disk turns additional checks. In some cases zero fields
1577 * are allowed (like pattern=0).
1579 * \param[in] d LOD device
1580 * \param[in] buf buffer with LOV EA to verify
1581 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1582 * 1 - from disk, do not allow
1584 * \retval 0 if the striping is valid
1585 * \retval -EINVAL if striping is invalid
1587 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1590 struct lov_user_md_v1 *lum;
1591 struct lov_user_md_v3 *lum3;
1592 struct pool_desc *pool = NULL;
1596 __u16 stripe_offset;
1603 if (buf->lb_len < sizeof(*lum)) {
1604 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1606 GOTO(out, rc = -EINVAL);
1609 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1610 if (magic != LOV_USER_MAGIC_V1 &&
1611 magic != LOV_USER_MAGIC_V3 &&
1612 magic != LOV_USER_MAGIC_SPECIFIC) {
1613 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1614 le32_to_cpu(lum->lmm_magic));
1615 GOTO(out, rc = -EINVAL);
1618 /* the user uses "0" for default stripe pattern normally. */
1619 if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1620 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1622 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1623 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1624 le32_to_cpu(lum->lmm_pattern));
1625 GOTO(out, rc = -EINVAL);
1628 /* a released lum comes from creating orphan on hsm release,
1629 * doesn't make sense to verify it. */
1630 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1633 /* 64kB is the largest common page size we see (ia64), and matches the
1635 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1636 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1637 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1638 stripe_size, LOV_MIN_STRIPE_SIZE);
1639 GOTO(out, rc = -EINVAL);
1642 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1643 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1644 lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1645 /* if offset is not within valid range [0, osts_size) */
1646 if (stripe_offset >= d->lod_osts_size) {
1647 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1648 stripe_offset, d->lod_osts_size);
1649 GOTO(out, rc = -EINVAL);
1652 /* if lmm_stripe_offset is *not* in bitmap */
1653 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1654 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1656 GOTO(out, rc = -EINVAL);
1660 if (magic == LOV_USER_MAGIC_V1)
1661 lum_size = offsetof(struct lov_user_md_v1,
1663 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1664 lum_size = offsetof(struct lov_user_md_v3,
1667 GOTO(out, rc = -EINVAL);
1669 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1670 if (buf->lb_len < lum_size) {
1671 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1672 "magic %#x and stripe_count %u\n",
1673 buf->lb_len, lum_size, magic, stripe_count);
1674 GOTO(out, rc = -EINVAL);
1677 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1681 /* In the function below, .hs_keycmp resolves to
1682 * pool_hashkey_keycmp() */
1683 /* coverity[overrun-buffer-val] */
1684 pool = lod_find_pool(d, lum3->lmm_pool_name);
1688 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1689 rc = lod_check_index_in_pool(stripe_offset, pool);
1691 GOTO(out, rc = -EINVAL);
1694 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1695 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1696 stripe_count, pool_tgt_count(pool));
1697 GOTO(out, rc = -EINVAL);
1702 lod_pool_putref(pool);
1708 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1710 LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1711 LOV_USER_MAGIC_COMP_V1, "Wrong magic %x\n",
1712 le32_to_cpu(comp->lcm_magic));
1713 LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1714 "bad index %d, max = %d\n",
1715 i, le16_to_cpu(comp->lcm_entry_count));
1717 return &comp->lcm_entries[i];
1720 #define for_each_comp_entry_v1(comp, entry) \
1721 for (entry = comp_entry_v1(comp, 0); \
1722 entry <= comp_entry_v1(comp, \
1723 le16_to_cpu(comp->lcm_entry_count) - 1); \
1726 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1)
1728 struct lov_comp_md_entry_v1 *ent, *dom_ent;
1730 __u32 dom_off, dom_size, comp_size;
1731 void *blob_src, *blob_dst;
1732 unsigned int blob_size, blob_shift;
1734 entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1735 /* if file has only DoM stripe return just error */
1739 comp_size = le32_to_cpu(comp_v1->lcm_size);
1740 dom_ent = &comp_v1->lcm_entries[0];
1741 dom_off = le32_to_cpu(dom_ent->lcme_offset);
1742 dom_size = le32_to_cpu(dom_ent->lcme_size);
1744 /* shift entries array first */
1745 comp_v1->lcm_entry_count = cpu_to_le16(entries);
1746 memmove(dom_ent, dom_ent + 1,
1747 entries * sizeof(struct lov_comp_md_entry_v1));
1749 /* now move blob of layouts */
1750 blob_dst = (void *)comp_v1 + dom_off - sizeof(*dom_ent);
1751 blob_src = (void *)comp_v1 + dom_off + dom_size;
1752 blob_size = (unsigned long)((void *)comp_v1 + comp_size - blob_src);
1753 blob_shift = sizeof(*dom_ent) + dom_size;
1755 memmove(blob_dst, blob_src, blob_size);
1757 for_each_comp_entry_v1(comp_v1, ent) {
1760 off = le32_to_cpu(ent->lcme_offset);
1761 ent->lcme_offset = cpu_to_le32(off - blob_shift);
1764 comp_v1->lcm_size = cpu_to_le32(comp_size - blob_shift);
1766 /* notify a caller to re-check entry */
1770 int lod_fix_dom_stripe(struct lod_device *d, struct lov_comp_md_v1 *comp_v1)
1772 struct lov_comp_md_entry_v1 *ent, *dom_ent;
1773 struct lu_extent *dom_ext, *ext;
1774 struct lov_user_md_v1 *lum;
1779 dom_ent = &comp_v1->lcm_entries[0];
1780 dom_ext = &dom_ent->lcme_extent;
1781 dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1782 stripe_size = d->lod_dom_max_stripesize;
1784 lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1785 CDEBUG(D_LAYOUT, "DoM component size %u was bigger than MDT limit %u, "
1786 "new size is %u\n", le32_to_cpu(lum->lmm_stripe_size),
1787 d->lod_dom_max_stripesize, stripe_size);
1788 lum->lmm_stripe_size = cpu_to_le32(stripe_size);
1790 for_each_comp_entry_v1(comp_v1, ent) {
1794 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1798 ext = &ent->lcme_extent;
1799 if (ext->e_start != dom_ext->e_end)
1802 /* Found next component after the DoM one with the same
1803 * mirror_id and adjust its start with DoM component end.
1805 * NOTE: we are considering here that there can be only one
1806 * DoM component in a file, all replicas are located on OSTs
1807 * always and don't need adjustment since use own layouts.
1809 ext->e_start = cpu_to_le64(stripe_size);
1813 if (stripe_size == 0) {
1814 /* DoM component size is zero due to server setting,
1815 * remove it from the layout */
1816 rc = lod_erase_dom_stripe(comp_v1);
1818 /* Update DoM extent end finally */
1819 dom_ext->e_end = cpu_to_le64(stripe_size);
1826 * Verify LOV striping.
1828 * \param[in] d LOD device
1829 * \param[in] buf buffer with LOV EA to verify
1830 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1831 * 1 - from disk, do not allow
1832 * \param[in] start extent start for composite layout
1834 * \retval 0 if the striping is valid
1835 * \retval -EINVAL if striping is invalid
1837 int lod_verify_striping(struct lod_device *d, struct lod_object *lo,
1838 const struct lu_buf *buf, bool is_from_disk)
1840 struct lov_desc *desc = &d->lod_desc;
1841 struct lov_user_md_v1 *lum;
1842 struct lov_comp_md_v1 *comp_v1;
1843 struct lov_comp_md_entry_v1 *ent;
1844 struct lu_extent *ext;
1847 __u32 stripe_size = 0;
1848 __u16 prev_mid = -1, mirror_id = -1;
1854 if (buf->lb_len < sizeof(lum->lmm_magic)) {
1855 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
1861 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1862 /* treat foreign LOV EA/object case first
1863 * XXX is it expected to try setting again a foreign?
1864 * XXX should we care about different current vs new layouts ?
1866 if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
1867 struct lov_foreign_md *lfm = buf->lb_buf;
1869 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
1871 "buf len %zu < min lov_foreign_md size (%zu)\n",
1872 buf->lb_len, offsetof(typeof(*lfm),
1877 if (foreign_size_le(lfm) > buf->lb_len) {
1879 "buf len %zu < this lov_foreign_md size (%zu)\n",
1880 buf->lb_len, foreign_size_le(lfm));
1883 /* Don't do anything with foreign layouts */
1887 /* normal LOV/layout cases */
1889 if (buf->lb_len < sizeof(*lum)) {
1890 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1895 if (magic != LOV_USER_MAGIC_V1 &&
1896 magic != LOV_USER_MAGIC_V3 &&
1897 magic != LOV_USER_MAGIC_SPECIFIC &&
1898 magic != LOV_USER_MAGIC_COMP_V1) {
1899 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1900 le32_to_cpu(lum->lmm_magic));
1904 if (magic != LOV_USER_MAGIC_COMP_V1)
1905 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1907 /* magic == LOV_USER_MAGIC_COMP_V1 */
1908 comp_v1 = buf->lb_buf;
1909 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1910 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1911 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1917 if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1918 CDEBUG(D_LAYOUT, "entry count is zero\n");
1922 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1923 lo->ldo_comp_cnt > 0) {
1924 /* could be called from lustre.lov.add */
1925 __u32 cnt = lo->ldo_comp_cnt;
1927 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
1928 prev_end = ext->e_end;
1933 for_each_comp_entry_v1(comp_v1, ent) {
1934 ext = &ent->lcme_extent;
1936 if (le64_to_cpu(ext->e_start) >= le64_to_cpu(ext->e_end)) {
1937 CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
1938 le64_to_cpu(ext->e_start),
1939 le64_to_cpu(ext->e_end));
1944 /* lcme_id contains valid value */
1945 if (le32_to_cpu(ent->lcme_id) == 0 ||
1946 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
1947 CDEBUG(D_LAYOUT, "invalid id %u\n",
1948 le32_to_cpu(ent->lcme_id));
1952 if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
1953 mirror_id = mirror_id_of(
1954 le32_to_cpu(ent->lcme_id));
1956 /* first component must start with 0 */
1957 if (mirror_id != prev_mid &&
1958 le64_to_cpu(ext->e_start) != 0) {
1960 "invalid start:%llu, expect:0\n",
1961 le64_to_cpu(ext->e_start));
1965 prev_mid = mirror_id;
1969 if (le64_to_cpu(ext->e_start) == 0) {
1974 /* the next must be adjacent with the previous one */
1975 if (le64_to_cpu(ext->e_start) != prev_end) {
1977 "invalid start actual:%llu, expect:%llu\n",
1978 le64_to_cpu(ext->e_start), prev_end);
1982 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
1983 tmp.lb_len = le32_to_cpu(ent->lcme_size);
1985 /* Check DoM entry is always the first one */
1987 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
1989 /* DoM component can be only the first stripe */
1990 if (le64_to_cpu(ext->e_start) > 0) {
1991 CDEBUG(D_LAYOUT, "invalid DoM component "
1992 "with %llu extent start\n",
1993 le64_to_cpu(ext->e_start));
1996 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1997 /* There is just one stripe on MDT and it must
1998 * cover whole component size. */
1999 if (stripe_size != le64_to_cpu(ext->e_end)) {
2000 CDEBUG(D_LAYOUT, "invalid DoM layout "
2001 "stripe size %u != %llu "
2002 "(component size)\n",
2003 stripe_size, prev_end);
2006 /* Check stripe size againts per-MDT limit */
2007 if (stripe_size > d->lod_dom_max_stripesize) {
2008 CDEBUG(D_LAYOUT, "DoM component size "
2009 "%u is bigger than MDT limit %u, check "
2010 "dom_max_stripesize parameter\n",
2011 stripe_size, d->lod_dom_max_stripesize);
2012 rc = lod_fix_dom_stripe(d, comp_v1);
2013 if (rc == -ERESTART) {
2014 /* DoM entry was removed, re-check
2015 * new layout from start */
2023 prev_end = le64_to_cpu(ext->e_end);
2025 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
2029 if (prev_end == LUSTRE_EOF)
2032 /* extent end must be aligned with the stripe_size */
2033 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2034 if (stripe_size == 0)
2035 stripe_size = desc->ld_default_stripe_size;
2036 if (stripe_size == 0 || (prev_end & (stripe_size - 1))) {
2037 CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
2038 "stripe_sz: %u, [%llu, %llu)\n",
2039 stripe_size, ext->e_start, prev_end);
2044 /* make sure that the mirror_count is telling the truth */
2045 if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
2052 * set the default stripe size, if unset.
2054 * \param[in,out] val number of bytes per OST stripe
2056 * The minimum stripe size is 64KB to ensure that a single stripe is an
2057 * even multiple of a client PAGE_SIZE (IA64, PPC, etc). Otherwise, it
2058 * is difficult to split dirty pages across OSCs during writes.
2060 void lod_fix_desc_stripe_size(__u64 *val)
2062 if (*val < LOV_MIN_STRIPE_SIZE) {
2064 LCONSOLE_INFO("Increasing default stripe size to "
2065 "minimum value %u\n",
2066 LOV_DESC_STRIPE_SIZE_DEFAULT);
2067 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
2068 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
2069 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
2070 LCONSOLE_WARN("Changing default stripe size to %llu (a "
2071 "multiple of %u)\n",
2072 *val, LOV_MIN_STRIPE_SIZE);
2077 * set the filesystem default number of stripes, if unset.
2079 * \param[in,out] val number of stripes
2081 * A value of "0" means "use the system-wide default stripe count", which
2082 * has either been inherited by now, or falls back to 1 stripe per file.
2083 * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
2084 * and is a valid value, so is left unchanged here.
2086 void lod_fix_desc_stripe_count(__u32 *val)
2093 * set the filesystem default layout pattern
2095 * \param[in,out] val LOV_PATTERN_* layout
2097 * A value of "0" means "use the system-wide default layout type", which
2098 * has either been inherited by now, or falls back to plain RAID0 striping.
2100 void lod_fix_desc_pattern(__u32 *val)
2102 /* from lov_setstripe */
2103 if ((*val != 0) && (*val != LOV_PATTERN_RAID0) &&
2104 (*val != LOV_PATTERN_MDT)) {
2105 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
2110 void lod_fix_desc_qos_maxage(__u32 *val)
2112 /* fix qos_maxage */
2114 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2118 * Used to fix insane default striping.
2120 * \param[in] desc striping description
2122 void lod_fix_desc(struct lov_desc *desc)
2124 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2125 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2126 lod_fix_desc_pattern(&desc->ld_pattern);
2127 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2131 * Initialize the structures used to store pools and default striping.
2133 * \param[in] lod LOD device
2134 * \param[in] lcfg configuration structure storing default striping.
2136 * \retval 0 if initialization succeeds
2137 * \retval negative error number on failure
2139 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2141 struct obd_device *obd;
2142 struct lov_desc *desc;
2146 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2147 LASSERT(obd != NULL);
2148 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2150 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2151 CERROR("LOD setup requires a descriptor\n");
2155 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2157 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2158 CERROR("descriptor size wrong: %d > %d\n",
2159 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2163 if (desc->ld_magic != LOV_DESC_MAGIC) {
2164 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2165 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2166 obd->obd_name, desc);
2167 lustre_swab_lov_desc(desc);
2169 CERROR("%s: Bad lov desc magic: %#x\n",
2170 obd->obd_name, desc->ld_magic);
2177 desc->ld_active_tgt_count = 0;
2178 lod->lod_desc = *desc;
2180 lod->lod_sp_me = LUSTRE_SP_CLI;
2182 /* Set up allocation policy (QoS and RR) */
2183 INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
2184 init_rwsem(&lod->lod_qos.lq_rw_sem);
2185 lod->lod_qos.lq_dirty = 1;
2186 lod->lod_qos.lq_rr.lqr_dirty = 1;
2187 lod->lod_qos.lq_reset = 1;
2188 /* Default priority is toward free space balance */
2189 lod->lod_qos.lq_prio_free = 232;
2190 /* Default threshold for rr (roughly 17%) */
2191 lod->lod_qos.lq_threshold_rr = 43;
2193 /* Set up OST pool environment */
2194 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
2195 HASH_POOLS_MAX_BITS,
2196 HASH_POOLS_BKT_BITS, 0,
2199 &pool_hash_operations,
2201 if (lod->lod_pools_hash_body == NULL)
2204 INIT_LIST_HEAD(&lod->lod_pool_list);
2205 lod->lod_pool_count = 0;
2206 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
2209 lod_qos_rr_init(&lod->lod_qos.lq_rr);
2210 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
2212 GOTO(out_pool_info, rc);
2217 lod_ost_pool_free(&lod->lod_pool_info);
2219 cfs_hash_putref(lod->lod_pools_hash_body);
2225 * Release the structures describing the pools.
2227 * \param[in] lod LOD device from which we release the structures
2231 int lod_pools_fini(struct lod_device *lod)
2233 struct obd_device *obd = lod2obd(lod);
2234 struct pool_desc *pool, *tmp;
2237 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2238 /* free pool structs */
2239 CDEBUG(D_INFO, "delete pool %p\n", pool);
2240 /* In the function below, .hs_keycmp resolves to
2241 * pool_hashkey_keycmp() */
2242 /* coverity[overrun-buffer-val] */
2243 lod_pool_del(obd, pool->pool_name);
2246 cfs_hash_putref(lod->lod_pools_hash_body);
2247 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
2248 lod_ost_pool_free(&lod->lod_pool_info);