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;
225 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
228 CERROR("request to add OBD %s with invalid generation: %d\n",
233 obd_str2uuid(&obd_uuid, osp);
235 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
236 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
238 CERROR("can't find %s device\n", osp);
242 LASSERT(obd->obd_lu_dev != NULL);
243 LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
245 lu_dev = obd->obd_lu_dev;
246 dt_dev = lu2dt_dev(lu_dev);
250 GOTO(out_cleanup, rc = -ENOMEM);
252 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
253 data->ocd_version = LUSTRE_VERSION_CODE;
254 data->ocd_index = index;
256 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
258 data->ocd_connect_flags |= OBD_CONNECT_AT |
261 #ifdef HAVE_LRU_RESIZE_SUPPORT
262 OBD_CONNECT_LRU_RESIZE |
265 OBD_CONNECT_REQPORTAL |
266 OBD_CONNECT_SKIP_ORPHAN |
268 OBD_CONNECT_LVB_TYPE |
269 OBD_CONNECT_VERSION |
270 OBD_CONNECT_PINGLESS |
272 OBD_CONNECT_BULK_MBITS;
274 data->ocd_group = tgt_index;
275 ltd = &lod->lod_ost_descs;
277 struct obd_import *imp = obd->u.cli.cl_import;
280 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
281 data->ocd_connect_flags |= OBD_CONNECT_ACL |
283 OBD_CONNECT_MDS_MDS |
288 OBD_CONNECT_BULK_MBITS;
289 spin_lock(&imp->imp_lock);
290 imp->imp_server_timeout = 1;
291 spin_unlock(&imp->imp_lock);
292 imp->imp_client->cli_request_portal = OUT_PORTAL;
293 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
295 ltd = &lod->lod_mdt_descs;
298 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
301 CERROR("%s: cannot connect to next dev %s (%d)\n",
302 obd->obd_name, osp, rc);
303 GOTO(out_cleanup, rc);
306 /* Allocate ost descriptor and fill it */
307 OBD_ALLOC_PTR(tgt_desc);
309 GOTO(out_conn, rc = -ENOMEM);
311 tgt_desc->ltd_tgt = dt_dev;
312 tgt_desc->ltd_exp = exp;
313 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
314 tgt_desc->ltd_gen = gen;
315 tgt_desc->ltd_index = index;
316 tgt_desc->ltd_active = active;
319 if (index >= ltd->ltd_tgts_size) {
320 /* we have to increase the size of the lod_osts array */
323 newsize = max(ltd->ltd_tgts_size, (__u32)2);
324 while (newsize < index + 1)
325 newsize = newsize << 1;
327 /* lod_bitmap_resize() needs lod_rw_sem
328 * which we hold with th reference */
329 lod_putref(lod, ltd);
331 rc = ltd_bitmap_resize(ltd, newsize);
338 mutex_lock(<d->ltd_mutex);
340 if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
341 CERROR("%s: device %d is registered already\n", obd->obd_name,
343 GOTO(out_mutex, rc = -EEXIST);
346 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
347 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
348 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
349 CERROR("can't allocate index to add %s\n",
351 GOTO(out_mutex, rc = -ENOMEM);
356 /* pool and qos are not supported for MDS stack yet */
357 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
360 CERROR("%s: can't set up pool, failed with %d\n",
365 rc = qos_add_tgt(lod, tgt_desc);
367 CERROR("%s: qos_add_tgt failed with %d\n",
372 /* The new OST is now a full citizen */
373 if (index >= lod->lod_desc.ld_tgt_count)
374 lod->lod_desc.ld_tgt_count = index + 1;
376 lod->lod_desc.ld_active_tgt_count++;
379 LTD_TGT(ltd, index) = tgt_desc;
380 cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
382 mutex_unlock(<d->ltd_mutex);
383 lod_putref(lod, ltd);
385 if (lod->lod_recovery_completed)
386 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
388 if (!for_ost && lod->lod_initialized) {
389 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
391 CERROR("%s: cannot start llog on %s:rc = %d\n",
392 lod2obd(lod)->obd_name, osp, rc);
397 rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
399 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
400 "index = %u, rc = %d\n", osp, type, index, rc);
401 GOTO(out_fini_llog, rc);
405 lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
406 tgt_desc->ltd_recovery_thread);
409 mutex_lock(<d->ltd_mutex);
411 if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
412 struct ptlrpc_thread *thread;
414 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
415 OBD_FREE_PTR(thread);
418 cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index);
419 LTD_TGT(ltd, index) = NULL;
421 lod_ost_pool_remove(&lod->lod_pool_info, index);
424 mutex_unlock(<d->ltd_mutex);
425 lod_putref(lod, ltd);
428 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);
444 * Schedule target removal from the target table.
446 * Mark the device as dead. The device is not removed here because it may
447 * still be in use. The device will be removed in lod_putref() when the
448 * last reference is released.
450 * \param[in] env execution environment for this thread
451 * \param[in] lod LOD device the target table belongs to
452 * \param[in] ltd target table
453 * \param[in] idx index of the target
454 * \param[in] for_ost type of the target: 0 - MDT, 1 - OST
456 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
457 struct lod_tgt_descs *ltd, unsigned idx,
460 LASSERT(LTD_TGT(ltd, idx));
462 lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
465 if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
466 struct ptlrpc_thread *thread;
468 thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
469 OBD_FREE_PTR(thread);
472 if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
473 LTD_TGT(ltd, idx)->ltd_reap = 1;
474 ltd->ltd_death_row++;
479 * Schedule removal of all the targets from the given target table.
481 * See more details in the description for __lod_del_device()
483 * \param[in] env execution environment for this thread
484 * \param[in] lod LOD device the target table belongs to
485 * \param[in] ltd target table
486 * \param[in] for_ost type of the target: MDT or OST
490 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
491 struct lod_tgt_descs *ltd, bool for_ost)
495 if (ltd->ltd_tgts_size <= 0)
498 mutex_lock(<d->ltd_mutex);
499 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
500 __lod_del_device(env, lod, ltd, idx, for_ost);
501 mutex_unlock(<d->ltd_mutex);
502 lod_putref(lod, ltd);
503 CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
504 for (idx = 0; idx < TGT_PTRS; idx++) {
505 if (ltd->ltd_tgt_idx[idx])
506 OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
508 ltd->ltd_tgts_size = 0;
513 * Remove device by name.
515 * Remove a device identified by \a osp from the target table. Given
516 * the device can be in use, the real deletion happens in lod_putref().
518 * \param[in] env execution environment for this thread
519 * \param[in] lod LOD device to be connected to the new OSP
520 * \param[in] ltd target table
521 * \param[in] osp name of OSP device to be removed
522 * \param[in] idx index of the target
523 * \param[in] gen generation number, not used currently
524 * \param[in] for_ost type of the target: 0 - MDT, 1 - OST
526 * \retval 0 if the device was scheduled for removal
527 * \retval -EINVAL if no device was found
529 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
530 struct lod_tgt_descs *ltd, char *osp, unsigned idx,
531 unsigned gen, bool for_ost)
533 struct obd_device *obd;
535 struct obd_uuid uuid;
538 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
540 obd_str2uuid(&uuid, osp);
542 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
543 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
545 CERROR("can't find %s device\n", osp);
550 CERROR("%s: request to remove OBD %s with invalid generation %d"
551 "\n", obd->obd_name, osp, gen);
555 obd_str2uuid(&uuid, osp);
558 mutex_lock(<d->ltd_mutex);
559 /* check that the index is allocated in the bitmap */
560 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
561 !LTD_TGT(ltd, idx)) {
562 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
563 GOTO(out, rc = -EINVAL);
566 /* check that the UUID matches */
567 if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
568 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
569 obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
571 GOTO(out, rc = -EINVAL);
574 __lod_del_device(env, lod, ltd, idx, for_ost);
577 mutex_unlock(<d->ltd_mutex);
578 lod_putref(lod, ltd);
583 * Resize per-thread storage to hold specified size.
585 * A helper function to resize per-thread temporary storage. This storage
586 * is used to process LOV/LVM EAs and may be quite large. We do not want to
587 * allocate/release it every time, so instead we put it into the env and
588 * reallocate on demand. The memory is released when the correspondent thread
591 * \param[in] info LOD-specific storage in the environment
592 * \param[in] size new size to grow the buffer to
594 * \retval 0 on success, -ENOMEM if reallocation failed
596 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
598 __u32 round = size_roundup_power2(size);
601 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
602 if (info->lti_ea_store) {
603 LASSERT(info->lti_ea_store_size);
604 LASSERT(info->lti_ea_store_size < round);
605 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
606 info->lti_ea_store_size, round);
607 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
608 info->lti_ea_store = NULL;
609 info->lti_ea_store_size = 0;
612 OBD_ALLOC_LARGE(info->lti_ea_store, round);
613 if (info->lti_ea_store == NULL)
615 info->lti_ea_store_size = round;
620 static void lod_free_comp_buffer(struct lod_layout_component *entries,
621 __u16 count, __u32 bufsize)
623 struct lod_layout_component *entry;
626 for (i = 0; i < count; i++) {
628 if (entry->llc_pool != NULL)
629 lod_set_pool(&entry->llc_pool, NULL);
630 if (entry->llc_ostlist.op_array)
631 OBD_FREE(entry->llc_ostlist.op_array,
632 entry->llc_ostlist.op_size);
633 LASSERT(entry->llc_stripe == NULL);
634 LASSERT(entry->llc_stripes_allocated == 0);
638 OBD_FREE_LARGE(entries, bufsize);
641 void lod_free_def_comp_entries(struct lod_default_striping *lds)
643 lod_free_comp_buffer(lds->lds_def_comp_entries,
644 lds->lds_def_comp_size_cnt,
646 sizeof(*lds->lds_def_comp_entries) *
647 lds->lds_def_comp_size_cnt));
648 lds->lds_def_comp_entries = NULL;
649 lds->lds_def_comp_cnt = 0;
650 lds->lds_def_striping_is_composite = 0;
651 lds->lds_def_comp_size_cnt = 0;
655 * Resize per-thread storage to hold default striping component entries
657 * A helper function to resize per-thread temporary storage. This storage
658 * is used to hold default LOV/LVM EAs and may be quite large. We do not want
659 * to allocate/release it every time, so instead we put it into the env and
660 * reallocate it on demand. The memory is released when the correspondent
661 * thread is finished.
663 * \param[in,out] lds default striping
664 * \param[in] count new component count to grow the buffer to
666 * \retval 0 on success, -ENOMEM if reallocation failed
668 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
670 struct lod_layout_component *entries;
671 __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
673 __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
674 lds->lds_def_comp_size_cnt);
679 OBD_ALLOC_LARGE(entries, new);
683 if (lds->lds_def_comp_entries != NULL) {
684 CDEBUG(D_INFO, "default striping component size %d is not "
685 "enough, need %d\n", old, new);
686 lod_free_def_comp_entries(lds);
689 lds->lds_def_comp_entries = entries;
690 lds->lds_def_comp_size_cnt = count;
695 void lod_free_comp_entries(struct lod_object *lo)
697 if (lo->ldo_mirrors) {
698 OBD_FREE(lo->ldo_mirrors,
699 sizeof(*lo->ldo_mirrors) * lo->ldo_mirror_count);
700 lo->ldo_mirrors = NULL;
701 lo->ldo_mirror_count = 0;
703 lod_free_comp_buffer(lo->ldo_comp_entries,
705 sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
706 lo->ldo_comp_entries = NULL;
707 lo->ldo_comp_cnt = 0;
708 lo->ldo_is_composite = 0;
711 int lod_alloc_comp_entries(struct lod_object *lo,
712 int mirror_count, int comp_count)
714 LASSERT(comp_count != 0);
715 LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
717 if (mirror_count > 0) {
718 OBD_ALLOC(lo->ldo_mirrors,
719 sizeof(*lo->ldo_mirrors) * mirror_count);
720 if (!lo->ldo_mirrors)
723 lo->ldo_mirror_count = mirror_count;
726 OBD_ALLOC_LARGE(lo->ldo_comp_entries,
727 sizeof(*lo->ldo_comp_entries) * comp_count);
728 if (lo->ldo_comp_entries == NULL) {
729 OBD_FREE(lo->ldo_mirrors,
730 sizeof(*lo->ldo_mirrors) * mirror_count);
731 lo->ldo_mirror_count = 0;
735 lo->ldo_comp_cnt = comp_count;
739 int lod_fill_mirrors(struct lod_object *lo)
741 struct lod_layout_component *lod_comp;
743 __u16 mirror_id = 0xffff;
747 LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
749 if (!lo->ldo_is_composite)
752 lod_comp = &lo->ldo_comp_entries[0];
753 for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
754 int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
755 int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
757 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
758 lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
759 lo->ldo_mirrors[mirror_idx].lme_primary |= preferred;
760 lo->ldo_mirrors[mirror_idx].lme_end = i;
766 if (mirror_idx >= lo->ldo_mirror_count)
769 mirror_id = mirror_id_of(lod_comp->llc_id);
771 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
772 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
773 lo->ldo_mirrors[mirror_idx].lme_primary = preferred;
774 lo->ldo_mirrors[mirror_idx].lme_start = i;
775 lo->ldo_mirrors[mirror_idx].lme_end = i;
777 if (mirror_idx != lo->ldo_mirror_count - 1)
784 * Generate on-disk lov_mds_md structure for each layout component based on
785 * the information in lod_object->ldo_comp_entries[i].
787 * \param[in] env execution environment for this thread
788 * \param[in] lo LOD object
789 * \param[in] comp_idx index of ldo_comp_entries
790 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
791 * \param[in|out] lmm_size buffer size/lmm size
792 * \param[in] is_dir generate lov ea for dir or file? For dir case,
793 * the stripe info is from the default stripe
794 * template, which is collected in lod_ah_init(),
795 * either from parent object or root object; for
796 * file case, it's from the @lo object
798 * \retval 0 if on disk structure is created successfully
799 * \retval negative error number on failure
801 static int lod_gen_component_ea(const struct lu_env *env,
802 struct lod_object *lo, int comp_idx,
803 struct lov_mds_md *lmm, int *lmm_size,
806 struct lod_thread_info *info = lod_env_info(env);
807 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
808 struct lod_device *lod;
809 struct lov_ost_data_v1 *objs;
810 struct lod_layout_component *lod_comp;
819 &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
821 lod_comp = &lo->ldo_comp_entries[comp_idx];
823 magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
824 if (lod_comp->llc_pattern == 0) /* default striping */
825 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
827 lmm->lmm_magic = cpu_to_le32(magic);
828 lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
829 fid_to_lmm_oi(fid, &lmm->lmm_oi);
830 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
831 lmm->lmm_oi.oi.oi_id++;
832 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
834 lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
835 lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
837 * for dir and uninstantiated component, lmm_layout_gen stores
838 * default stripe offset.
840 lmm->lmm_layout_gen =
841 (is_dir || !lod_comp_inited(lod_comp)) ?
842 cpu_to_le16(lod_comp->llc_stripe_offset) :
843 cpu_to_le16(lod_comp->llc_layout_gen);
845 if (magic == LOV_MAGIC_V1) {
846 objs = &lmm->lmm_objects[0];
848 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
849 size_t cplen = strlcpy(v3->lmm_pool_name,
851 sizeof(v3->lmm_pool_name));
852 if (cplen >= sizeof(v3->lmm_pool_name))
854 objs = &v3->lmm_objects[0];
856 stripe_count = lod_comp_entry_stripe_count(lo, lod_comp, is_dir);
857 if (!is_dir && lo->ldo_is_composite)
858 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
860 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
863 /* generate ost_idx of this component stripe */
864 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
865 for (i = 0; i < stripe_count; i++) {
866 struct dt_object *object;
867 __u32 ost_idx = (__u32)-1UL;
868 int type = LU_SEQ_RANGE_OST;
870 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
871 object = lod_comp->llc_stripe[i];
872 /* instantiated component */
873 info->lti_fid = *lu_object_fid(&object->do_lu);
875 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
877 if (cfs_fail_val == 0)
878 cfs_fail_val = info->lti_fid.f_oid;
880 info->lti_fid.f_oid = cfs_fail_val;
883 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
886 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
887 objs[i].l_ost_gen = cpu_to_le32(0);
888 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
891 rc = lod_fld_lookup(env, lod, &info->lti_fid,
894 CERROR("%s: Can not locate "DFID": rc = %d\n",
895 lod2obd(lod)->obd_name,
896 PFID(&info->lti_fid), rc);
899 } else if (lod_comp->llc_ostlist.op_array) {
900 /* user specified ost list */
901 ost_idx = lod_comp->llc_ostlist.op_array[i];
904 * with un-instantiated or with no specified ost list
905 * component, its l_ost_idx does not matter.
907 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
910 if (lmm_size != NULL)
911 *lmm_size = lov_mds_md_size(stripe_count, magic);
916 * Generate on-disk lov_mds_md structure based on the information in
917 * the lod_object->ldo_comp_entries.
919 * \param[in] env execution environment for this thread
920 * \param[in] lo LOD object
921 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
922 * \param[in|out] lmm_size buffer size/lmm size
923 * \param[in] is_dir generate lov ea for dir or file? For dir case,
924 * the stripe info is from the default stripe
925 * template, which is collected in lod_ah_init(),
926 * either from parent object or root object; for
927 * file case, it's from the @lo object
929 * \retval 0 if on disk structure is created successfully
930 * \retval negative error number on failure
932 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
933 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
935 struct lov_comp_md_entry_v1 *lcme;
936 struct lov_comp_md_v1 *lcm;
937 struct lod_layout_component *comp_entries;
938 __u16 comp_cnt, mirror_cnt;
940 int i, rc = 0, offset;
944 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
945 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
946 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
948 lo->ldo_def_striping->lds_def_striping_is_composite;
950 comp_cnt = lo->ldo_comp_cnt;
951 mirror_cnt = lo->ldo_mirror_count;
952 comp_entries = lo->ldo_comp_entries;
953 is_composite = lo->ldo_is_composite;
956 LASSERT(lmm_size != NULL);
957 LASSERT(comp_cnt != 0 && comp_entries != NULL);
960 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
964 lcm = (struct lov_comp_md_v1 *)lmm;
965 memset(lcm, 0, sizeof(*lcm));
967 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
968 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
969 lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
970 lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
972 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
973 LASSERT(offset % sizeof(__u64) == 0);
975 for (i = 0; i < comp_cnt; i++) {
976 struct lod_layout_component *lod_comp;
977 struct lov_mds_md *sub_md;
980 lod_comp = &comp_entries[i];
981 lcme = &lcm->lcm_entries[i];
983 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
984 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
986 /* component could be un-inistantiated */
987 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
988 lcme->lcme_extent.e_start =
989 cpu_to_le64(lod_comp->llc_extent.e_start);
990 lcme->lcme_extent.e_end =
991 cpu_to_le64(lod_comp->llc_extent.e_end);
992 lcme->lcme_offset = cpu_to_le32(offset);
994 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
995 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
998 lcme->lcme_size = cpu_to_le32(size);
1000 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
1001 "offset:%d lmm_size:%d\n", offset, *lmm_size);
1003 lcm->lcm_size = cpu_to_le32(offset);
1004 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
1006 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
1017 * Fill lti_ea_store buffer in the environment with a value for the given
1018 * EA. The buffer is reallocated if the value doesn't fit.
1020 * \param[in,out] env execution environment for this thread
1021 * .lti_ea_store buffer is filled with EA's value
1022 * \param[in] lo LOD object
1023 * \param[in] name name of the EA
1025 * \retval > 0 if EA is fetched successfully
1026 * \retval 0 if EA is empty
1027 * \retval negative error number on failure
1029 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
1032 struct lod_thread_info *info = lod_env_info(env);
1033 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1039 if (unlikely(info->lti_ea_store == NULL)) {
1040 /* just to enter in allocation block below */
1044 info->lti_buf.lb_buf = info->lti_ea_store;
1045 info->lti_buf.lb_len = info->lti_ea_store_size;
1046 rc = dt_xattr_get(env, next, &info->lti_buf, name);
1049 /* if object is not striped or inaccessible */
1050 if (rc == -ENODATA || rc == -ENOENT)
1053 if (rc == -ERANGE) {
1054 /* EA doesn't fit, reallocate new buffer */
1055 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
1056 if (rc == -ENODATA || rc == -ENOENT)
1062 rc = lod_ea_store_resize(info, rc);
1072 * Verify the target index is present in the current configuration.
1074 * \param[in] md LOD device where the target table is stored
1075 * \param[in] idx target's index
1077 * \retval 0 if the index is present
1078 * \retval -EINVAL if not
1080 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1082 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1083 !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
1084 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1085 md->lod_ost_descs.ltd_tgts_size);
1089 if (unlikely(OST_TGT(md, idx) == NULL)) {
1090 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1091 lod2obd(md)->obd_name, idx);
1095 if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
1096 CERROR("%s: invalid lod device, for idx: %d\n",
1097 lod2obd(md)->obd_name , idx);
1105 * Instantiate objects for stripes.
1107 * Allocate and initialize LU-objects representing the stripes. The number
1108 * of the stripes (ldo_stripe_count) must be initialized already. The caller
1109 * must ensure nobody else is calling the function on the object at the same
1110 * time. FLDB service must be running to be able to map a FID to the targets
1111 * and find appropriate device representing that target.
1113 * \param[in] env execution environment for this thread
1114 * \param[in,out] lo LOD object
1115 * \param[in] objs an array of IDs to creates the objects from
1116 * \param[in] comp_idx index of ldo_comp_entries
1118 * \retval 0 if the objects are instantiated successfully
1119 * \retval negative error number on failure
1121 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1122 struct lov_ost_data_v1 *objs, int comp_idx)
1124 struct lod_layout_component *lod_comp;
1125 struct lod_thread_info *info = lod_env_info(env);
1126 struct lod_device *md;
1127 struct lu_object *o, *n;
1128 struct lu_device *nd;
1129 struct dt_object **stripe;
1135 LASSERT(lo != NULL);
1136 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1138 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1139 lod_comp = &lo->ldo_comp_entries[comp_idx];
1141 LASSERT(lod_comp->llc_stripe == NULL);
1142 LASSERT(lod_comp->llc_stripe_count > 0);
1143 LASSERT(lod_comp->llc_stripe_size > 0);
1145 stripe_len = lod_comp->llc_stripe_count;
1146 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1150 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1151 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1154 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1155 idx = le32_to_cpu(objs[i].l_ost_idx);
1156 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1159 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1160 PFID(&info->lti_fid));
1161 lod_getref(&md->lod_ost_descs);
1163 rc = validate_lod_and_idx(md, idx);
1164 if (unlikely(rc != 0)) {
1165 lod_putref(md, &md->lod_ost_descs);
1169 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
1170 lod_putref(md, &md->lod_ost_descs);
1172 /* In the function below, .hs_keycmp resolves to
1173 * u_obj_hop_keycmp() */
1174 /* coverity[overrun-buffer-val] */
1175 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1177 GOTO(out, rc = PTR_ERR(o));
1179 n = lu_object_locate(o->lo_header, nd->ld_type);
1182 stripe[i] = container_of(n, struct dt_object, do_lu);
1187 for (i = 0; i < stripe_len; i++)
1188 if (stripe[i] != NULL)
1189 dt_object_put(env, stripe[i]);
1191 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1192 lod_comp->llc_stripe_count = 0;
1194 lod_comp->llc_stripe = stripe;
1195 lod_comp->llc_stripes_allocated = stripe_len;
1202 * Instantiate objects for striping.
1204 * Parse striping information in \a buf and instantiate the objects
1205 * representing the stripes.
1207 * \param[in] env execution environment for this thread
1208 * \param[in] lo LOD object
1209 * \param[in] buf buffer storing LOV EA to parse
1211 * \retval 0 if parsing and objects creation succeed
1212 * \retval negative error number on failure
1214 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1215 const struct lu_buf *buf)
1217 struct lov_mds_md_v1 *lmm;
1218 struct lov_comp_md_v1 *comp_v1 = NULL;
1219 struct lov_ost_data_v1 *objs;
1220 __u32 magic, pattern;
1223 __u16 mirror_cnt = 0;
1227 LASSERT(buf->lb_buf);
1228 LASSERT(buf->lb_len);
1230 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1231 magic = le32_to_cpu(lmm->lmm_magic);
1233 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1234 magic != LOV_MAGIC_COMP_V1)
1235 GOTO(out, rc = -EINVAL);
1237 lod_free_comp_entries(lo);
1239 if (magic == LOV_MAGIC_COMP_V1) {
1240 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1241 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1243 GOTO(out, rc = -EINVAL);
1244 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1245 lo->ldo_is_composite = 1;
1246 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1248 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1251 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1252 lo->ldo_is_composite = 0;
1255 rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1259 for (i = 0; i < comp_cnt; i++) {
1260 struct lod_layout_component *lod_comp;
1261 struct lu_extent *ext;
1264 lod_comp = &lo->ldo_comp_entries[i];
1265 if (lo->ldo_is_composite) {
1266 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1267 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1268 magic = le32_to_cpu(lmm->lmm_magic);
1270 ext = &comp_v1->lcm_entries[i].lcme_extent;
1271 lod_comp->llc_extent.e_start =
1272 le64_to_cpu(ext->e_start);
1273 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1274 lod_comp->llc_flags =
1275 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1277 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1278 if (lod_comp->llc_id == LCME_ID_INVAL)
1279 GOTO(out, rc = -EINVAL);
1281 lod_comp_set_init(lod_comp);
1284 pattern = le32_to_cpu(lmm->lmm_pattern);
1285 if (lov_pattern(pattern) != LOV_PATTERN_RAID0 &&
1286 lov_pattern(pattern) != LOV_PATTERN_MDT)
1287 GOTO(out, rc = -EINVAL);
1289 lod_comp->llc_pattern = pattern;
1290 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1291 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1292 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1294 if (magic == LOV_MAGIC_V3) {
1295 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1296 lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1297 objs = &v3->lmm_objects[0];
1299 lod_set_pool(&lod_comp->llc_pool, NULL);
1300 objs = &lmm->lmm_objects[0];
1304 * If uninstantiated template component has valid l_ost_idx,
1305 * then user has specified ost list for this component.
1307 if (!lod_comp_inited(lod_comp)) {
1310 if (objs[0].l_ost_idx != (__u32)-1UL) {
1311 stripe_count = lod_comp_entry_stripe_count(
1312 lo, lod_comp, false);
1314 * load the user specified ost list, when this
1315 * component is instantiated later, it will be
1316 * used in lod_alloc_ost_list().
1318 lod_comp->llc_ostlist.op_count = stripe_count;
1319 lod_comp->llc_ostlist.op_size =
1320 stripe_count * sizeof(__u32);
1321 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1322 lod_comp->llc_ostlist.op_size);
1323 if (!lod_comp->llc_ostlist.op_array)
1324 GOTO(out, rc = -ENOMEM);
1326 for (j = 0; j < stripe_count; j++)
1327 lod_comp->llc_ostlist.op_array[j] =
1328 le32_to_cpu(objs[j].l_ost_idx);
1331 * this component OST objects starts from the
1332 * first ost_idx, lod_alloc_ost_list() will
1335 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1338 * for uninstantiated component,
1339 * lmm_layout_gen stores default stripe offset.
1341 lod_comp->llc_stripe_offset =
1342 lmm->lmm_layout_gen;
1346 /* skip un-instantiated component object initialization */
1347 if (!lod_comp_inited(lod_comp))
1350 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1351 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1352 rc = lod_initialize_objects(env, lo, objs, i);
1358 rc = lod_fill_mirrors(lo);
1364 lod_object_free_striping(env, lo);
1369 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1370 * is already cached.
1372 * \param[in] lo LOD object
1374 * \retval True if the striping is cached, otherwise
1377 static bool lod_striping_loaded(struct lod_object *lo)
1379 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1380 lo->ldo_comp_cached)
1383 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1384 if (lo->ldo_stripe != NULL || lo->ldo_dir_stripe_loaded)
1387 /* Never load LMV stripe for slaves of striped dir */
1388 if (lo->ldo_dir_slave_stripe)
1396 * Initialize the object representing the stripes.
1398 * Unless the stripes are initialized already, fetch LOV (for regular
1399 * objects) or LMV (for directory objects) EA and call lod_parse_striping()
1400 * to instantiate the objects representing the stripes. Caller should
1401 * hold the dt_write_lock(next).
1403 * \param[in] env execution environment for this thread
1404 * \param[in,out] lo LOD object
1406 * \retval 0 if parsing and object creation succeed
1407 * \retval negative error number on failure
1409 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
1411 struct lod_thread_info *info = lod_env_info(env);
1412 struct lu_buf *buf = &info->lti_buf;
1413 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1417 if (!dt_object_exists(next))
1420 if (lod_striping_loaded(lo))
1423 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1424 rc = lod_get_lov_ea(env, lo);
1428 * there is LOV EA (striping information) in this object
1429 * let's parse it and create in-core objects for the stripes
1431 buf->lb_buf = info->lti_ea_store;
1432 buf->lb_len = info->lti_ea_store_size;
1433 rc = lod_parse_striping(env, lo, buf);
1435 lo->ldo_comp_cached = 1;
1436 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1437 rc = lod_get_lmv_ea(env, lo);
1438 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1)) {
1439 /* Let's set stripe_loaded to avoid further
1440 * stripe loading especially for non-stripe directory,
1441 * which can hurt performance. (See LU-9840)
1444 lo->ldo_dir_stripe_loaded = 1;
1445 GOTO(out, rc = rc > 0 ? -EINVAL : rc);
1447 buf->lb_buf = info->lti_ea_store;
1448 buf->lb_len = info->lti_ea_store_size;
1449 if (rc == sizeof(struct lmv_mds_md_v1)) {
1450 rc = lod_load_lmv_shards(env, lo, buf, true);
1451 if (buf->lb_buf != info->lti_ea_store) {
1452 OBD_FREE_LARGE(info->lti_ea_store,
1453 info->lti_ea_store_size);
1454 info->lti_ea_store = buf->lb_buf;
1455 info->lti_ea_store_size = buf->lb_len;
1463 * there is LMV EA (striping information) in this object
1464 * let's parse it and create in-core objects for the stripes
1466 rc = lod_parse_dir_striping(env, lo, buf);
1468 lo->ldo_dir_stripe_loaded = 1;
1475 * A generic function to initialize the stripe objects.
1477 * A protected version of lod_load_striping_locked() - load the striping
1478 * information from storage, parse that and instantiate LU objects to
1479 * represent the stripes. The LOD object \a lo supplies a pointer to the
1480 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1481 * return an array of references to the newly instantiated objects.
1483 * \param[in] env execution environment for this thread
1484 * \param[in,out] lo LOD object, where striping is stored and
1485 * which gets an array of references
1487 * \retval 0 if parsing and object creation succeed
1488 * \retval negative error number on failure
1490 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
1492 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1495 if (!dt_object_exists(next))
1498 /* Check without locking first */
1499 if (lod_striping_loaded(lo))
1502 /* currently this code is supposed to be called from declaration
1503 * phase only, thus the object is not expected to be locked by caller */
1504 dt_write_lock(env, next, 0);
1505 rc = lod_load_striping_locked(env, lo);
1506 dt_write_unlock(env, next);
1511 * Verify lov_user_md_v1/v3 striping.
1513 * Check the validity of all fields including the magic, stripe size,
1514 * stripe count, stripe offset and that the pool is present. Also check
1515 * that each target index points to an existing target. The additional
1516 * \a is_from_disk turns additional checks. In some cases zero fields
1517 * are allowed (like pattern=0).
1519 * \param[in] d LOD device
1520 * \param[in] buf buffer with LOV EA to verify
1521 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1522 * 1 - from disk, do not allow
1524 * \retval 0 if the striping is valid
1525 * \retval -EINVAL if striping is invalid
1527 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1530 struct lov_user_md_v1 *lum;
1531 struct lov_user_md_v3 *lum3;
1532 struct pool_desc *pool = NULL;
1536 __u16 stripe_offset;
1543 if (buf->lb_len < sizeof(*lum)) {
1544 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1546 GOTO(out, rc = -EINVAL);
1549 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1550 if (magic != LOV_USER_MAGIC_V1 &&
1551 magic != LOV_USER_MAGIC_V3 &&
1552 magic != LOV_USER_MAGIC_SPECIFIC) {
1553 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1554 le32_to_cpu(lum->lmm_magic));
1555 GOTO(out, rc = -EINVAL);
1558 /* the user uses "0" for default stripe pattern normally. */
1559 if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1560 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1562 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1563 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1564 le32_to_cpu(lum->lmm_pattern));
1565 GOTO(out, rc = -EINVAL);
1568 /* a released lum comes from creating orphan on hsm release,
1569 * doesn't make sense to verify it. */
1570 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1573 /* 64kB is the largest common page size we see (ia64), and matches the
1575 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1576 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1577 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1578 stripe_size, LOV_MIN_STRIPE_SIZE);
1579 GOTO(out, rc = -EINVAL);
1582 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1583 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1584 lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1585 /* if offset is not within valid range [0, osts_size) */
1586 if (stripe_offset >= d->lod_osts_size) {
1587 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1588 stripe_offset, d->lod_osts_size);
1589 GOTO(out, rc = -EINVAL);
1592 /* if lmm_stripe_offset is *not* in bitmap */
1593 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1594 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1596 GOTO(out, rc = -EINVAL);
1600 if (magic == LOV_USER_MAGIC_V1)
1601 lum_size = offsetof(struct lov_user_md_v1,
1603 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1604 lum_size = offsetof(struct lov_user_md_v3,
1607 GOTO(out, rc = -EINVAL);
1609 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1610 if (buf->lb_len < lum_size) {
1611 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1612 "magic %#x and stripe_count %u\n",
1613 buf->lb_len, lum_size, magic, stripe_count);
1614 GOTO(out, rc = -EINVAL);
1617 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1621 /* In the function below, .hs_keycmp resolves to
1622 * pool_hashkey_keycmp() */
1623 /* coverity[overrun-buffer-val] */
1624 pool = lod_find_pool(d, lum3->lmm_pool_name);
1628 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1629 rc = lod_check_index_in_pool(stripe_offset, pool);
1631 GOTO(out, rc = -EINVAL);
1634 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1635 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1636 stripe_count, pool_tgt_count(pool));
1637 GOTO(out, rc = -EINVAL);
1642 lod_pool_putref(pool);
1648 * Verify LOV striping.
1650 * \param[in] d LOD device
1651 * \param[in] buf buffer with LOV EA to verify
1652 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1653 * 1 - from disk, do not allow
1654 * \param[in] start extent start for composite layout
1656 * \retval 0 if the striping is valid
1657 * \retval -EINVAL if striping is invalid
1659 int lod_verify_striping(struct lod_device *d, struct lod_object *lo,
1660 const struct lu_buf *buf, bool is_from_disk)
1662 struct lov_desc *desc = &d->lod_desc;
1663 struct lov_user_md_v1 *lum;
1664 struct lov_comp_md_v1 *comp_v1;
1665 struct lov_comp_md_entry_v1 *ent;
1666 struct lu_extent *ext;
1669 __u32 stripe_size = 0;
1670 __u16 prev_mid = -1, mirror_id = -1;
1671 __u32 mirror_count = 0;
1678 if (buf->lb_len < sizeof(*lum)) {
1679 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1684 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1685 if (magic != LOV_USER_MAGIC_V1 &&
1686 magic != LOV_USER_MAGIC_V3 &&
1687 magic != LOV_USER_MAGIC_SPECIFIC &&
1688 magic != LOV_USER_MAGIC_COMP_V1) {
1689 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1690 le32_to_cpu(lum->lmm_magic));
1694 if (magic != LOV_USER_MAGIC_COMP_V1)
1695 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1697 /* magic == LOV_USER_MAGIC_COMP_V1 */
1698 comp_v1 = buf->lb_buf;
1699 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1700 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1701 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1705 if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1706 CDEBUG(D_LAYOUT, "entry count is zero\n");
1710 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1711 lo->ldo_comp_cnt > 0) {
1712 /* could be called from lustre.lov.add */
1713 __u32 cnt = lo->ldo_comp_cnt;
1715 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
1716 prev_end = ext->e_end;
1721 for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
1722 ent = &comp_v1->lcm_entries[i];
1723 ext = &ent->lcme_extent;
1725 if (le64_to_cpu(ext->e_start) >= le64_to_cpu(ext->e_end)) {
1726 CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
1727 le64_to_cpu(ext->e_start),
1728 le64_to_cpu(ext->e_end));
1733 /* lcme_id contains valid value */
1734 if (le32_to_cpu(ent->lcme_id) == 0 ||
1735 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
1736 CDEBUG(D_LAYOUT, "invalid id %u\n",
1737 le32_to_cpu(ent->lcme_id));
1741 if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
1742 mirror_id = mirror_id_of(
1743 le32_to_cpu(ent->lcme_id));
1745 /* first component must start with 0 */
1746 if (mirror_id != prev_mid &&
1747 le64_to_cpu(ext->e_start) != 0) {
1749 "invalid start:%llu, expect:0\n",
1750 le64_to_cpu(ext->e_start));
1754 prev_mid = mirror_id;
1758 if (le64_to_cpu(ext->e_start) == 0) {
1763 /* the next must be adjacent with the previous one */
1764 if (le64_to_cpu(ext->e_start) != prev_end) {
1766 "invalid start actual:%llu, expect:%llu\n",
1767 le64_to_cpu(ext->e_start), prev_end);
1771 prev_end = le64_to_cpu(ext->e_end);
1773 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
1774 tmp.lb_len = le32_to_cpu(ent->lcme_size);
1776 /* Check DoM entry is always the first one */
1778 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
1780 /* DoM component can be only the first entry */
1782 CDEBUG(D_LAYOUT, "invalid DoM layout "
1783 "entry found at %i index\n", i);
1786 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1787 /* There is just one stripe on MDT and it must
1788 * cover whole component size. */
1789 if (stripe_size != prev_end) {
1790 CDEBUG(D_LAYOUT, "invalid DoM layout "
1791 "stripe size %u != %llu "
1792 "(component size)\n",
1793 stripe_size, prev_end);
1796 /* Check stripe size againts per-MDT limit */
1797 if (stripe_size > d->lod_dom_max_stripesize) {
1798 CDEBUG(D_LAYOUT, "DoM component size "
1799 "%u is bigger than MDT limit %u, check "
1800 "dom_max_stripesize parameter\n",
1801 stripe_size, d->lod_dom_max_stripesize);
1806 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
1810 if (prev_end == LUSTRE_EOF)
1813 /* extent end must be aligned with the stripe_size */
1814 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1815 if (stripe_size == 0)
1816 stripe_size = desc->ld_default_stripe_size;
1817 if (stripe_size == 0 || (prev_end & (stripe_size - 1))) {
1818 CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
1819 "stripe_sz: %u, [%llu, %llu)\n",
1820 stripe_size, ext->e_start, prev_end);
1825 /* make sure that the mirror_count is telling the truth */
1826 if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
1833 * set the default stripe size, if unset.
1835 * \param[in,out] val number of bytes per OST stripe
1837 * The minimum stripe size is 64KB to ensure that a single stripe is an
1838 * even multiple of a client PAGE_SIZE (IA64, PPC, etc). Otherwise, it
1839 * is difficult to split dirty pages across OSCs during writes.
1841 void lod_fix_desc_stripe_size(__u64 *val)
1843 if (*val < LOV_MIN_STRIPE_SIZE) {
1845 LCONSOLE_INFO("Increasing default stripe size to "
1846 "minimum value %u\n",
1847 LOV_DESC_STRIPE_SIZE_DEFAULT);
1848 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1849 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1850 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1851 LCONSOLE_WARN("Changing default stripe size to %llu (a "
1852 "multiple of %u)\n",
1853 *val, LOV_MIN_STRIPE_SIZE);
1858 * set the filesystem default number of stripes, if unset.
1860 * \param[in,out] val number of stripes
1862 * A value of "0" means "use the system-wide default stripe count", which
1863 * has either been inherited by now, or falls back to 1 stripe per file.
1864 * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
1865 * and is a valid value, so is left unchanged here.
1867 void lod_fix_desc_stripe_count(__u32 *val)
1874 * set the filesystem default layout pattern
1876 * \param[in,out] val LOV_PATTERN_* layout
1878 * A value of "0" means "use the system-wide default layout type", which
1879 * has either been inherited by now, or falls back to plain RAID0 striping.
1881 void lod_fix_desc_pattern(__u32 *val)
1883 /* from lov_setstripe */
1884 if ((*val != 0) && (*val != LOV_PATTERN_RAID0) &&
1885 (*val != LOV_PATTERN_MDT)) {
1886 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1891 void lod_fix_desc_qos_maxage(__u32 *val)
1893 /* fix qos_maxage */
1895 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1899 * Used to fix insane default striping.
1901 * \param[in] desc striping description
1903 void lod_fix_desc(struct lov_desc *desc)
1905 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1906 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1907 lod_fix_desc_pattern(&desc->ld_pattern);
1908 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1912 * Initialize the structures used to store pools and default striping.
1914 * \param[in] lod LOD device
1915 * \param[in] lcfg configuration structure storing default striping.
1917 * \retval 0 if initialization succeeds
1918 * \retval negative error number on failure
1920 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1922 struct obd_device *obd;
1923 struct lov_desc *desc;
1927 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1928 LASSERT(obd != NULL);
1929 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1931 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1932 CERROR("LOD setup requires a descriptor\n");
1936 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1938 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1939 CERROR("descriptor size wrong: %d > %d\n",
1940 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1944 if (desc->ld_magic != LOV_DESC_MAGIC) {
1945 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1946 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1947 obd->obd_name, desc);
1948 lustre_swab_lov_desc(desc);
1950 CERROR("%s: Bad lov desc magic: %#x\n",
1951 obd->obd_name, desc->ld_magic);
1958 desc->ld_active_tgt_count = 0;
1959 lod->lod_desc = *desc;
1961 lod->lod_sp_me = LUSTRE_SP_CLI;
1963 /* Set up allocation policy (QoS and RR) */
1964 INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1965 init_rwsem(&lod->lod_qos.lq_rw_sem);
1966 lod->lod_qos.lq_dirty = 1;
1967 lod->lod_qos.lq_rr.lqr_dirty = 1;
1968 lod->lod_qos.lq_reset = 1;
1969 /* Default priority is toward free space balance */
1970 lod->lod_qos.lq_prio_free = 232;
1971 /* Default threshold for rr (roughly 17%) */
1972 lod->lod_qos.lq_threshold_rr = 43;
1974 /* Set up OST pool environment */
1975 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1976 HASH_POOLS_MAX_BITS,
1977 HASH_POOLS_BKT_BITS, 0,
1980 &pool_hash_operations,
1982 if (lod->lod_pools_hash_body == NULL)
1985 INIT_LIST_HEAD(&lod->lod_pool_list);
1986 lod->lod_pool_count = 0;
1987 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1990 lod_qos_rr_init(&lod->lod_qos.lq_rr);
1991 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1993 GOTO(out_pool_info, rc);
1998 lod_ost_pool_free(&lod->lod_pool_info);
2000 cfs_hash_putref(lod->lod_pools_hash_body);
2006 * Release the structures describing the pools.
2008 * \param[in] lod LOD device from which we release the structures
2012 int lod_pools_fini(struct lod_device *lod)
2014 struct obd_device *obd = lod2obd(lod);
2015 struct pool_desc *pool, *tmp;
2018 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2019 /* free pool structs */
2020 CDEBUG(D_INFO, "delete pool %p\n", pool);
2021 /* In the function below, .hs_keycmp resolves to
2022 * pool_hashkey_keycmp() */
2023 /* coverity[overrun-buffer-val] */
2024 lod_pool_del(obd, pool->pool_name);
2027 cfs_hash_putref(lod->lod_pools_hash_body);
2028 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
2029 lod_ost_pool_free(&lod->lod_pool_info);