4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * lustre/lod/lod_lov.c
31 * A set of helpers to maintain Logical Object Volume (LOV)
32 * Extended Attribute (EA) and known OST targets
34 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
37 #define DEBUG_SUBSYSTEM S_MDS
39 #include <obd_class.h>
40 #include <lustre_lfsck.h>
41 #include <lustre_lmv.h>
42 #include <lustre_swab.h>
44 #include "lod_internal.h"
47 * Increase reference count on the target table.
49 * Increase reference count on the target table usage to prevent racing with
50 * addition/deletion. Any function that expects the table to remain
51 * stationary must take a ref.
53 * \param[in] ltd target table (lod_ost_descs or lod_mdt_descs)
55 void lod_getref(struct lod_tgt_descs *ltd)
57 down_read(<d->ltd_rw_sem);
58 mutex_lock(<d->ltd_mutex);
60 mutex_unlock(<d->ltd_mutex);
64 * Decrease reference count on the target table.
66 * Companion of lod_getref() to release a reference on the target table.
67 * If this is the last reference and the OST entry was scheduled for deletion,
68 * the descriptor is removed from the table.
70 * \param[in] lod LOD device from which we release a reference
71 * \param[in] ltd target table (lod_ost_descs or lod_mdt_descs)
73 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
75 mutex_lock(<d->ltd_mutex);
77 if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
78 struct lod_tgt_desc *tgt_desc, *tmp;
81 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
84 ltd_foreach_tgt_safe(ltd, tgt_desc, tmp) {
86 if (!tgt_desc->ltd_reap)
89 list_add(&tgt_desc->ltd_kill, &kill);
90 lu_tgt_pool_remove(<d->ltd_tgt_pool,
92 ltd_del_tgt(ltd, tgt_desc);
95 mutex_unlock(<d->ltd_mutex);
96 up_read(<d->ltd_rw_sem);
98 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
101 list_del(&tgt_desc->ltd_kill);
102 rc = obd_disconnect(tgt_desc->ltd_exp);
104 CERROR("%s: failed to disconnect %s: rc = %d\n",
105 lod2obd(lod)->obd_name,
106 obd_uuid2str(&tgt_desc->ltd_uuid), rc);
107 OBD_FREE_PTR(tgt_desc);
110 mutex_unlock(<d->ltd_mutex);
111 up_read(<d->ltd_rw_sem);
116 * Connect LOD to a new OSP and add it to the target table.
118 * Connect to the OSP device passed, initialize all the internal
119 * structures related to the device and add it to the target table.
121 * \param[in] env execution environment for this thread
122 * \param[in] lod LOD device to be connected to the new OSP
123 * \param[in] osp name of OSP device name to be added
124 * \param[in] index index of the new target
125 * \param[in] gen target's generation number
126 * \param[in] tgt_index OSP's group
127 * \param[in] type type of device (mdc or osc)
128 * \param[in] active state of OSP: 0 - inactive, 1 - active
130 * \retval 0 if added successfully
131 * \retval negative error number on failure
133 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
134 char *osp, unsigned index, unsigned gen, int tgt_index,
135 char *type, int active)
137 struct obd_connect_data *data = NULL;
138 struct obd_export *exp = NULL;
139 struct obd_device *obd;
140 struct lu_device *lu_dev;
141 struct dt_device *dt_dev;
143 struct lod_tgt_desc *tgt_desc;
144 struct lod_tgt_descs *ltd;
145 struct lustre_cfg *lcfg;
146 struct obd_uuid obd_uuid;
148 bool connected = false;
151 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
154 CERROR("request to add OBD %s with invalid generation: %d\n",
159 obd_str2uuid(&obd_uuid, osp);
161 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
162 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
164 CERROR("can't find %s device\n", osp);
168 LASSERT(obd->obd_lu_dev != NULL);
169 LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
171 lu_dev = obd->obd_lu_dev;
172 dt_dev = lu2dt_dev(lu_dev);
176 GOTO(out_cleanup, rc = -ENOMEM);
178 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
179 data->ocd_version = LUSTRE_VERSION_CODE;
180 data->ocd_index = index;
182 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
184 data->ocd_connect_flags |= OBD_CONNECT_AT |
187 #ifdef HAVE_LRU_RESIZE_SUPPORT
188 OBD_CONNECT_LRU_RESIZE |
191 OBD_CONNECT_REQPORTAL |
192 OBD_CONNECT_SKIP_ORPHAN |
194 OBD_CONNECT_LVB_TYPE |
195 OBD_CONNECT_VERSION |
196 OBD_CONNECT_PINGLESS |
198 OBD_CONNECT_BULK_MBITS;
200 data->ocd_group = tgt_index;
201 ltd = &lod->lod_ost_descs;
203 struct obd_import *imp = obd->u.cli.cl_import;
206 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
207 data->ocd_connect_flags |= OBD_CONNECT_ACL |
209 OBD_CONNECT_MDS_MDS |
210 OBD_CONNECT_MULTIMODRPCS |
215 OBD_CONNECT_BULK_MBITS;
216 spin_lock(&imp->imp_lock);
217 imp->imp_server_timeout = 1;
218 spin_unlock(&imp->imp_lock);
219 imp->imp_client->cli_request_portal = OUT_PORTAL;
220 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
222 ltd = &lod->lod_mdt_descs;
225 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
228 CERROR("%s: cannot connect to next dev %s (%d)\n",
229 obd->obd_name, osp, rc);
230 GOTO(out_cleanup, rc);
234 /* Allocate ost descriptor and fill it */
235 OBD_ALLOC_PTR(tgt_desc);
237 GOTO(out_cleanup, rc = -ENOMEM);
239 tgt_desc->ltd_tgt = dt_dev;
240 tgt_desc->ltd_exp = exp;
241 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
242 tgt_desc->ltd_gen = gen;
243 tgt_desc->ltd_index = index;
244 tgt_desc->ltd_active = active;
246 down_write(<d->ltd_rw_sem);
247 mutex_lock(<d->ltd_mutex);
248 rc = ltd_add_tgt(ltd, tgt_desc);
252 rc = lu_qos_add_tgt(<d->ltd_qos, tgt_desc);
254 GOTO(out_del_tgt, rc);
256 rc = lu_tgt_pool_add(<d->ltd_tgt_pool, index,
257 ltd->ltd_lov_desc.ld_tgt_count);
259 CERROR("%s: can't set up pool, failed with %d\n",
261 GOTO(out_del_tgt, rc);
264 mutex_unlock(<d->ltd_mutex);
265 up_write(<d->ltd_rw_sem);
267 if (lod->lod_recovery_completed)
268 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
270 if (!for_ost && lod->lod_initialized) {
271 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
273 CERROR("%s: cannot start llog on %s:rc = %d\n",
274 lod2obd(lod)->obd_name, osp, rc);
279 rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
281 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
282 "index = %u, rc = %d\n", osp, type, index, rc);
283 GOTO(out_fini_llog, rc);
287 lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
288 &tgt_desc->ltd_recovery_task);
290 down_write(<d->ltd_rw_sem);
291 mutex_lock(<d->ltd_mutex);
292 lu_tgt_pool_remove(<d->ltd_tgt_pool, index);
294 ltd_del_tgt(ltd, tgt_desc);
296 mutex_unlock(<d->ltd_mutex);
297 up_write(<d->ltd_rw_sem);
298 OBD_FREE_PTR(tgt_desc);
300 /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
301 * objects from the MDT stack. See LU-7184. */
302 lcfg = &lod_env_info(env)->lti_lustre_cfg;
303 memset(lcfg, 0, sizeof(*lcfg));
304 lcfg->lcfg_version = LUSTRE_CFG_VERSION;
305 lcfg->lcfg_command = LCFG_CLEANUP;
306 lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
315 * Schedule target removal from the target table.
317 * Mark the device as dead. The device is not removed here because it may
318 * still be in use. The device will be removed in lod_putref() when the
319 * last reference is released.
321 * \param[in] env execution environment for this thread
322 * \param[in] lod LOD device the target table belongs to
323 * \param[in] ltd target table
324 * \param[in] tgt target
326 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
327 struct lod_tgt_descs *ltd, struct lu_tgt_desc *tgt)
329 lfsck_del_target(env, lod->lod_child, tgt->ltd_tgt, tgt->ltd_index,
332 if (!tgt->ltd_reap) {
334 ltd->ltd_death_row++;
339 * Schedule removal of all the targets from the given target table.
341 * See more details in the description for __lod_del_device()
343 * \param[in] env execution environment for this thread
344 * \param[in] lod LOD device the target table belongs to
345 * \param[in] ltd target table
349 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
350 struct lod_tgt_descs *ltd)
352 struct lu_tgt_desc *tgt;
354 if (ltd->ltd_tgts_size <= 0)
358 mutex_lock(<d->ltd_mutex);
359 ltd_foreach_tgt(ltd, tgt)
360 __lod_del_device(env, lod, ltd, tgt);
361 mutex_unlock(<d->ltd_mutex);
362 lod_putref(lod, ltd);
364 lu_tgt_descs_fini(ltd);
370 * Remove device by name.
372 * Remove a device identified by \a osp from the target table. Given
373 * the device can be in use, the real deletion happens in lod_putref().
375 * \param[in] env execution environment for this thread
376 * \param[in] lod LOD device to be connected to the new OSP
377 * \param[in] ltd target table
378 * \param[in] osp name of OSP device to be removed
379 * \param[in] idx index of the target
380 * \param[in] gen generation number, not used currently
382 * \retval 0 if the device was scheduled for removal
383 * \retval -EINVAL if no device was found
385 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
386 struct lod_tgt_descs *ltd, char *osp, unsigned int idx,
389 struct obd_device *obd;
390 struct lu_tgt_desc *tgt;
391 struct obd_uuid uuid;
396 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
398 obd_str2uuid(&uuid, osp);
400 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
401 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
403 CERROR("can't find %s device\n", osp);
408 CERROR("%s: request to remove OBD %s with invalid generation %d"
409 "\n", obd->obd_name, osp, gen);
413 obd_str2uuid(&uuid, osp);
416 mutex_lock(<d->ltd_mutex);
417 tgt = LTD_TGT(ltd, idx);
418 /* check that the index is allocated in the bitmap */
419 if (!test_bit(idx, ltd->ltd_tgt_bitmap) || !tgt) {
420 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
421 GOTO(out, rc = -EINVAL);
424 /* check that the UUID matches */
425 if (!obd_uuid_equals(&uuid, &tgt->ltd_uuid)) {
426 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
427 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), idx, osp);
428 GOTO(out, rc = -EINVAL);
431 __lod_del_device(env, lod, ltd, tgt);
434 mutex_unlock(<d->ltd_mutex);
435 lod_putref(lod, ltd);
440 * Resize per-thread storage to hold specified size.
442 * A helper function to resize per-thread temporary storage. This storage
443 * is used to process LOV/LVM EAs and may be quite large. We do not want to
444 * allocate/release it every time, so instead we put it into the env and
445 * reallocate on demand. The memory is released when the correspondent thread
448 * \param[in] info LOD-specific storage in the environment
449 * \param[in] size new size to grow the buffer to
451 * \retval 0 on success, -ENOMEM if reallocation failed
453 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
455 __u32 round = size_roundup_power2(size);
457 if (info->lti_ea_store) {
458 LASSERT(info->lti_ea_store_size);
459 LASSERT(info->lti_ea_store_size < round);
460 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
461 info->lti_ea_store_size, round);
462 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
463 info->lti_ea_store = NULL;
464 info->lti_ea_store_size = 0;
467 OBD_ALLOC_LARGE(info->lti_ea_store, round);
468 if (info->lti_ea_store == NULL)
470 info->lti_ea_store_size = round;
475 static void lod_free_comp_buffer(struct lod_layout_component *entries,
476 __u16 count, __u32 bufsize)
478 struct lod_layout_component *entry;
481 for (i = 0; i < count; i++) {
483 if (entry->llc_pool != NULL)
484 lod_set_pool(&entry->llc_pool, NULL);
485 if (entry->llc_ostlist.op_array)
486 OBD_FREE(entry->llc_ostlist.op_array,
487 entry->llc_ostlist.op_size);
488 LASSERT(entry->llc_stripe == NULL);
489 LASSERT(entry->llc_stripes_allocated == 0);
493 OBD_FREE_LARGE(entries, bufsize);
496 void lod_free_def_comp_entries(struct lod_default_striping *lds)
498 lod_free_comp_buffer(lds->lds_def_comp_entries,
499 lds->lds_def_comp_size_cnt,
501 sizeof(*lds->lds_def_comp_entries) *
502 lds->lds_def_comp_size_cnt));
503 lds->lds_def_comp_entries = NULL;
504 lds->lds_def_comp_cnt = 0;
505 lds->lds_def_striping_is_composite = 0;
506 lds->lds_def_comp_size_cnt = 0;
510 * Resize per-thread storage to hold default striping component entries
512 * A helper function to resize per-thread temporary storage. This storage
513 * is used to hold default LOV/LVM EAs and may be quite large. We do not want
514 * to allocate/release it every time, so instead we put it into the env and
515 * reallocate it on demand. The memory is released when the correspondent
516 * thread is finished.
518 * \param[in,out] lds default striping
519 * \param[in] count new component count to grow the buffer to
521 * \retval 0 on success, -ENOMEM if reallocation failed
523 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
525 struct lod_layout_component *entries;
526 __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
528 __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
529 lds->lds_def_comp_size_cnt);
534 OBD_ALLOC_LARGE(entries, new);
538 if (lds->lds_def_comp_entries != NULL) {
539 CDEBUG(D_INFO, "default striping component size %d is not "
540 "enough, need %d\n", old, new);
541 lod_free_def_comp_entries(lds);
544 lds->lds_def_comp_entries = entries;
545 lds->lds_def_comp_size_cnt = count;
550 void lod_free_comp_entries(struct lod_object *lo)
552 if (lo->ldo_mirrors) {
553 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, lo->ldo_mirror_count);
554 lo->ldo_mirrors = NULL;
555 lo->ldo_mirror_count = 0;
557 lod_free_comp_buffer(lo->ldo_comp_entries,
559 sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
560 lo->ldo_comp_entries = NULL;
561 lo->ldo_comp_cnt = 0;
562 lo->ldo_is_composite = 0;
565 int lod_alloc_comp_entries(struct lod_object *lo,
566 int mirror_count, int comp_count)
568 LASSERT(comp_count != 0);
569 LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
571 if (mirror_count > 0) {
572 OBD_ALLOC_PTR_ARRAY(lo->ldo_mirrors, mirror_count);
573 if (!lo->ldo_mirrors)
576 lo->ldo_mirror_count = mirror_count;
579 OBD_ALLOC_LARGE(lo->ldo_comp_entries,
580 sizeof(*lo->ldo_comp_entries) * comp_count);
581 if (lo->ldo_comp_entries == NULL) {
582 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, mirror_count);
583 lo->ldo_mirrors = NULL;
584 lo->ldo_mirror_count = 0;
588 lo->ldo_comp_cnt = comp_count;
589 lo->ldo_is_foreign = 0;
593 int lod_fill_mirrors(struct lod_object *lo)
595 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
596 struct lod_layout_component *lod_comp;
597 bool found_preferred = false;
599 __u16 mirror_id = 0xffff;
603 LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
605 if (!lo->ldo_is_composite)
608 lod_comp = &lo->ldo_comp_entries[0];
610 for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
611 bool stale = lod_comp->llc_flags & LCME_FL_STALE;
612 bool preferred = lod_comp->llc_flags & LCME_FL_PREF_WR;
613 bool init = (lod_comp->llc_stripe != NULL) &&
614 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
615 !(lod_comp->llc_pattern & LOV_PATTERN_MDT);
619 /* calculate component preference over all used OSTs */
620 for (j = 0; init && j < lod_comp->llc_stripes_allocated; j++) {
621 __u32 idx = lod_comp->llc_ost_indices[j];
622 struct lod_tgt_desc *ltd;
624 if (lod_comp->llc_stripe[j] == NULL)
627 if (unlikely(idx >= lod->lod_ost_descs.ltd_tgts_size)) {
628 CERROR("%s: "DFID" OST idx %u > max %u\n",
629 lod2obd(lod)->obd_name,
630 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
631 idx, lod->lod_ost_descs.ltd_tgts_size);
634 ltd = OST_TGT(lod, idx);
635 if (unlikely(!ltd)) {
636 CERROR("%s: "DFID" OST idx %u is NULL\n",
637 lod2obd(lod)->obd_name,
638 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
643 if (ltd->ltd_statfs.os_state & OS_STATFS_NONROT)
647 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
648 lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
649 lo->ldo_mirrors[mirror_idx].lme_prefer |= preferred;
650 lo->ldo_mirrors[mirror_idx].lme_preference += pref;
651 lo->ldo_mirrors[mirror_idx].lme_end = i;
655 if (mirror_idx >= 0 && preferred &&
656 !lo->ldo_mirrors[mirror_idx].lme_stale)
657 found_preferred = true;
661 if (mirror_idx >= lo->ldo_mirror_count)
664 mirror_id = mirror_id_of(lod_comp->llc_id);
666 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
667 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
668 lo->ldo_mirrors[mirror_idx].lme_prefer = preferred;
669 lo->ldo_mirrors[mirror_idx].lme_preference = pref;
670 lo->ldo_mirrors[mirror_idx].lme_start = i;
671 lo->ldo_mirrors[mirror_idx].lme_end = i;
673 if (mirror_idx != lo->ldo_mirror_count - 1)
676 if (!found_preferred && mirror_idx > 0) {
680 * if no explicited preferred found, then find a mirror
681 * with higher number of non-rotational OSTs
684 for (i = 0; i <= mirror_idx; i++) {
685 if (lo->ldo_mirrors[i].lme_stale)
687 if (lo->ldo_mirrors[i].lme_preference > pref) {
688 pref = lo->ldo_mirrors[i].lme_preference;
694 lo->ldo_mirrors[best].lme_prefer = 1;
701 * Generate on-disk lov_mds_md structure for each layout component based on
702 * the information in lod_object->ldo_comp_entries[i].
704 * \param[in] env execution environment for this thread
705 * \param[in] lo LOD object
706 * \param[in] comp_idx index of ldo_comp_entries
707 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
708 * \param[in|out] lmm_size buffer size/lmm size
709 * \param[in] is_dir generate lov ea for dir or file? For dir case,
710 * the stripe info is from the default stripe
711 * template, which is collected in lod_ah_init(),
712 * either from parent object or root object; for
713 * file case, it's from the @lo object
715 * \retval 0 if on disk structure is created successfully
716 * \retval negative error number on failure
718 static int lod_gen_component_ea(const struct lu_env *env,
719 struct lod_object *lo, int comp_idx,
720 struct lov_mds_md *lmm, int *lmm_size,
723 struct lod_thread_info *info = lod_env_info(env);
724 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
725 struct lod_device *lod;
726 struct lov_ost_data_v1 *objs;
727 struct lod_layout_component *lod_comp;
736 &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
738 lod_comp = &lo->ldo_comp_entries[comp_idx];
740 magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
741 if (lod_comp->llc_pattern == 0) /* default striping */
742 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
744 lmm->lmm_magic = cpu_to_le32(magic);
745 lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
746 fid_to_lmm_oi(fid, &lmm->lmm_oi);
747 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
748 lmm->lmm_oi.oi.oi_id++;
749 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
751 lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
752 lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
754 * for dir and uninstantiated component, lmm_layout_gen stores
755 * default stripe offset.
757 lmm->lmm_layout_gen =
758 (is_dir || !lod_comp_inited(lod_comp)) ?
759 cpu_to_le16(lod_comp->llc_stripe_offset) :
760 cpu_to_le16(lod_comp->llc_layout_gen);
762 if (magic == LOV_MAGIC_V1) {
763 objs = &lmm->lmm_objects[0];
765 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
766 size_t cplen = strlcpy(v3->lmm_pool_name,
768 sizeof(v3->lmm_pool_name));
769 if (cplen >= sizeof(v3->lmm_pool_name))
771 objs = &v3->lmm_objects[0];
773 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
774 stripe_count = lod_comp_entry_stripe_count(lo, comp_idx, is_dir);
775 if (stripe_count == 0 && !is_dir &&
776 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
777 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
778 /* Try again if all active targets are disconnected.
779 * It is possible when MDS does failover. */
780 if (!lod->lod_ost_active_count &&
786 if (!is_dir && lo->ldo_is_composite)
787 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
789 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
792 /* generate ost_idx of this component stripe */
793 for (i = 0; i < stripe_count; i++) {
794 struct dt_object *object;
795 __u32 ost_idx = (__u32)-1UL;
796 int type = LU_SEQ_RANGE_OST;
798 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
799 object = lod_comp->llc_stripe[i];
800 /* instantiated component */
801 info->lti_fid = *lu_object_fid(&object->do_lu);
803 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
805 if (cfs_fail_val == 0)
806 cfs_fail_val = info->lti_fid.f_oid;
808 info->lti_fid.f_oid = cfs_fail_val;
811 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
814 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
815 objs[i].l_ost_gen = cpu_to_le32(0);
816 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
819 rc = lod_fld_lookup(env, lod, &info->lti_fid,
822 CERROR("%s: Can not locate "DFID": rc = %d\n",
823 lod2obd(lod)->obd_name,
824 PFID(&info->lti_fid), rc);
827 } else if (lod_comp->llc_ostlist.op_array &&
828 lod_comp->llc_ostlist.op_count) {
829 /* user specified ost list */
830 ost_idx = lod_comp->llc_ostlist.op_array[i];
833 * with un-instantiated or with no specified ost list
834 * component, its l_ost_idx does not matter.
836 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
839 if (lmm_size != NULL)
840 *lmm_size = lov_mds_md_size(stripe_count, magic);
845 * Generate on-disk lov_mds_md structure based on the information in
846 * the lod_object->ldo_comp_entries.
848 * \param[in] env execution environment for this thread
849 * \param[in] lo LOD object
850 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
851 * \param[in|out] lmm_size buffer size/lmm size
852 * \param[in] is_dir generate lov ea for dir or file? For dir case,
853 * the stripe info is from the default stripe
854 * template, which is collected in lod_ah_init(),
855 * either from parent object or root object; for
856 * file case, it's from the @lo object
858 * \retval 0 if on disk structure is created successfully
859 * \retval negative error number on failure
861 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
862 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
864 struct lov_comp_md_entry_v1 *lcme;
865 struct lov_comp_md_v1 *lcm;
866 struct lod_layout_component *comp_entries;
867 __u16 comp_cnt, mirror_cnt;
868 bool is_composite, is_foreign = false;
869 int i, rc = 0, offset;
873 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
874 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
875 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
877 lo->ldo_def_striping->lds_def_striping_is_composite;
879 comp_cnt = lo->ldo_comp_cnt;
880 mirror_cnt = lo->ldo_mirror_count;
881 comp_entries = lo->ldo_comp_entries;
882 is_composite = lo->ldo_is_composite;
883 is_foreign = lo->ldo_is_foreign;
886 LASSERT(lmm_size != NULL);
889 struct lov_foreign_md *lfm;
891 lfm = (struct lov_foreign_md *)lmm;
892 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
893 /* need to store little-endian */
894 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
895 __swab32s(&lfm->lfm_magic);
896 __swab32s(&lfm->lfm_length);
897 __swab32s(&lfm->lfm_type);
898 __swab32s(&lfm->lfm_flags);
900 *lmm_size = lo->ldo_foreign_lov_size;
904 LASSERT(comp_cnt != 0 && comp_entries != NULL);
907 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
911 lcm = (struct lov_comp_md_v1 *)lmm;
912 memset(lcm, 0, sizeof(*lcm));
914 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
915 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
916 lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
918 lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
920 lcm->lcm_flags = LCM_FL_NONE;
922 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
923 LASSERT(offset % sizeof(__u64) == 0);
925 for (i = 0; i < comp_cnt; i++) {
926 struct lod_layout_component *lod_comp;
927 struct lov_mds_md *sub_md;
930 lod_comp = &comp_entries[i];
931 lcme = &lcm->lcm_entries[i];
933 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
934 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
936 /* component could be un-inistantiated */
937 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
938 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
939 lcme->lcme_timestamp =
940 cpu_to_le64(lod_comp->llc_timestamp);
941 if (lod_comp->llc_flags & LCME_FL_EXTENSION && !is_dir)
942 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
944 lcme->lcme_extent.e_start =
945 cpu_to_le64(lod_comp->llc_extent.e_start);
946 lcme->lcme_extent.e_end =
947 cpu_to_le64(lod_comp->llc_extent.e_end);
948 lcme->lcme_offset = cpu_to_le32(offset);
950 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
951 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
954 lcme->lcme_size = cpu_to_le32(size);
956 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
957 "offset:%d lmm_size:%d\n", offset, *lmm_size);
959 lcm->lcm_size = cpu_to_le32(offset);
960 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
962 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
973 * Fill lti_ea_store buffer in the environment with a value for the given
974 * EA. The buffer is reallocated if the value doesn't fit.
976 * \param[in,out] env execution environment for this thread
977 * .lti_ea_store buffer is filled with EA's value
978 * \param[in] lo LOD object
979 * \param[in] name name of the EA
981 * \retval > 0 if EA is fetched successfully
982 * \retval 0 if EA is empty
983 * \retval negative error number on failure
985 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
988 struct lod_thread_info *info = lod_env_info(env);
989 struct dt_object *next = dt_object_child(&lo->ldo_obj);
995 if (unlikely(info->lti_ea_store == NULL)) {
996 /* just to enter in allocation block below */
1000 info->lti_buf.lb_buf = info->lti_ea_store;
1001 info->lti_buf.lb_len = info->lti_ea_store_size;
1002 rc = dt_xattr_get(env, next, &info->lti_buf, name);
1005 /* if object is not striped or inaccessible */
1006 if (rc == -ENODATA || rc == -ENOENT)
1009 if (rc == -ERANGE) {
1010 /* EA doesn't fit, reallocate new buffer */
1011 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
1012 if (rc == -ENODATA || rc == -ENOENT)
1018 rc = lod_ea_store_resize(info, rc);
1028 * Verify the target index is present in the current configuration.
1030 * \param[in] md LOD device where the target table is stored
1031 * \param[in] idx target's index
1033 * \retval 0 if the index is present
1034 * \retval -EINVAL if not
1036 int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1038 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1039 !test_bit(idx, md->lod_ost_bitmap))) {
1040 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1041 md->lod_ost_descs.ltd_tgts_size);
1045 if (unlikely(OST_TGT(md, idx) == NULL)) {
1046 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1047 lod2obd(md)->obd_name, idx);
1051 if (unlikely(OST_TGT(md, idx)->ltd_tgt == NULL)) {
1052 CERROR("%s: invalid lod device, for idx: %d\n",
1053 lod2obd(md)->obd_name , idx);
1061 * Instantiate objects for stripes.
1063 * Allocate and initialize LU-objects representing the stripes. The number
1064 * of the stripes (llc_stripe_count) must be initialized already. The caller
1065 * must ensure nobody else is calling the function on the object at the same
1066 * time. FLDB service must be running to be able to map a FID to the targets
1067 * and find appropriate device representing that target.
1069 * \param[in] env execution environment for this thread
1070 * \param[in,out] lo LOD object
1071 * \param[in] objs an array of IDs to creates the objects from
1072 * \param[in] comp_idx index of ldo_comp_entries
1074 * \retval 0 if the objects are instantiated successfully
1075 * \retval negative error number on failure
1077 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1078 struct lov_ost_data_v1 *objs, int comp_idx)
1080 struct lod_layout_component *lod_comp;
1081 struct lod_thread_info *info = lod_env_info(env);
1082 struct lod_device *md;
1083 struct lu_object *o, *n;
1084 struct lu_device *nd;
1085 struct dt_object **stripe = NULL;
1086 __u32 *ost_indices = NULL;
1092 LASSERT(lo != NULL);
1093 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1095 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1096 lod_comp = &lo->ldo_comp_entries[comp_idx];
1098 LASSERT(lod_comp->llc_stripe == NULL);
1099 LASSERT(lod_comp->llc_stripe_count > 0);
1100 LASSERT(lod_comp->llc_stripe_size > 0);
1102 stripe_len = lod_comp->llc_stripe_count;
1103 OBD_ALLOC_PTR_ARRAY(stripe, stripe_len);
1106 OBD_ALLOC_PTR_ARRAY(ost_indices, stripe_len);
1108 GOTO(out, rc = -ENOMEM);
1110 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1111 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1114 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1115 idx = le32_to_cpu(objs[i].l_ost_idx);
1116 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1119 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1120 PFID(&info->lti_fid));
1121 lod_getref(&md->lod_ost_descs);
1123 rc = validate_lod_and_idx(md, idx);
1124 if (unlikely(rc != 0)) {
1125 lod_putref(md, &md->lod_ost_descs);
1129 nd = &OST_TGT(md, idx)->ltd_tgt->dd_lu_dev;
1130 lod_putref(md, &md->lod_ost_descs);
1132 /* In the function below, .hs_keycmp resolves to
1133 * u_obj_hop_keycmp() */
1134 /* coverity[overrun-buffer-val] */
1135 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1137 GOTO(out, rc = PTR_ERR(o));
1139 n = lu_object_locate(o->lo_header, nd->ld_type);
1142 stripe[i] = container_of(n, struct dt_object, do_lu);
1143 ost_indices[i] = idx;
1148 for (i = 0; i < stripe_len; i++)
1149 if (stripe[i] != NULL)
1150 dt_object_put(env, stripe[i]);
1152 OBD_FREE_PTR_ARRAY(stripe, stripe_len);
1153 lod_comp->llc_stripe_count = 0;
1155 OBD_FREE_PTR_ARRAY(ost_indices, stripe_len);
1157 lod_comp->llc_stripe = stripe;
1158 lod_comp->llc_ost_indices = ost_indices;
1159 lod_comp->llc_stripes_allocated = stripe_len;
1166 * Instantiate objects for striping.
1168 * Parse striping information in \a buf and instantiate the objects
1169 * representing the stripes.
1171 * \param[in] env execution environment for this thread
1172 * \param[in] lo LOD object
1173 * \param[in] buf buffer storing LOV EA to parse
1174 * \param[in] lvf verify flags when parsing the layout
1176 * \retval 0 if parsing and objects creation succeed
1177 * \retval negative error number on failure
1179 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1180 const struct lu_buf *buf, enum layout_verify_flags lvf)
1182 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1183 struct lov_mds_md_v1 *lmm;
1184 struct lov_comp_md_v1 *comp_v1 = NULL;
1185 struct lov_foreign_md *foreign = NULL;
1186 struct lov_ost_data_v1 *objs;
1187 __u32 magic, pattern;
1188 __u16 mirror_cnt = 0;
1191 __u16 mirror_id = MIRROR_ID_NEG;
1193 int stale_mirrors = 0;
1197 LASSERT(buf->lb_buf);
1198 LASSERT(buf->lb_len);
1199 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1201 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1202 magic = le32_to_cpu(lmm->lmm_magic);
1204 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1205 magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN &&
1206 magic != LOV_MAGIC_SEL)
1207 GOTO(out, rc = -EINVAL);
1209 lod_striping_free_nolock(env, lo);
1211 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1212 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1213 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1215 GOTO(out, rc = -EINVAL);
1216 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1217 lo->ldo_is_composite = 1;
1218 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1220 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1223 lo->ldo_flr_state = LCM_FL_NONE;
1224 } else if (magic == LOV_MAGIC_FOREIGN) {
1227 foreign = (struct lov_foreign_md *)buf->lb_buf;
1228 length = offsetof(typeof(*foreign), lfm_value);
1229 if (buf->lb_len < length ||
1230 buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1232 "buf len %zu too small for lov_foreign_md\n",
1234 GOTO(out, rc = -EINVAL);
1237 /* just cache foreign LOV EA raw */
1238 rc = lod_alloc_foreign_lov(lo, length);
1241 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1245 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1246 lo->ldo_is_composite = 0;
1249 rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1253 for (i = 0; i < comp_cnt; i++) {
1254 struct lod_layout_component *lod_comp;
1255 struct lu_extent *ext;
1258 lod_comp = &lo->ldo_comp_entries[i];
1259 if (lo->ldo_is_composite) {
1260 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1261 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1263 ext = &comp_v1->lcm_entries[i].lcme_extent;
1264 lod_comp->llc_extent.e_start =
1265 le64_to_cpu(ext->e_start);
1266 if (lod_comp->llc_extent.e_start &
1267 (LOV_MIN_STRIPE_SIZE - 1)) {
1269 "extent start %llu is not a multiple of min size %u\n",
1270 lod_comp->llc_extent.e_start,
1271 LOV_MIN_STRIPE_SIZE);
1272 GOTO(out, rc = -EINVAL);
1275 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1276 if (lod_comp->llc_extent.e_end != LUSTRE_EOF &&
1277 lod_comp->llc_extent.e_end &
1278 (LOV_MIN_STRIPE_SIZE - 1)) {
1280 "extent end %llu is not a multiple of min size %u\n",
1281 lod_comp->llc_extent.e_end,
1282 LOV_MIN_STRIPE_SIZE);
1283 GOTO(out, rc = -EINVAL);
1286 lod_comp->llc_flags =
1287 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1289 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1290 lod_comp->llc_timestamp = le64_to_cpu(
1291 comp_v1->lcm_entries[i].lcme_timestamp);
1293 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1294 if (lod_comp->llc_id == LCME_ID_INVAL)
1295 GOTO(out, rc = -EINVAL);
1297 if (lvf & LVF_ALL_STALE) {
1298 if (mirror_id_of(lod_comp->llc_id) ==
1300 /* remaining comps in the mirror */
1301 stale |= lod_comp->llc_flags &
1305 * new mirror, check last mirror's
1312 mirror_id_of(lod_comp->llc_id);
1314 /* the first comp of the new mirror */
1315 stale = lod_comp->llc_flags &
1320 if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
1321 comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
1322 CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
1323 lod2obd(d)->obd_name,
1324 lod_comp->llc_flags, lod_comp->llc_id, i,
1325 PFID(lod_object_fid(lo)),
1326 le32_to_cpu(comp_v1->lcm_magic));
1329 lod_comp_set_init(lod_comp);
1332 pattern = le32_to_cpu(lmm->lmm_pattern);
1333 if (!lov_pattern_supported(lov_pattern(pattern)))
1334 GOTO(out, rc = -EINVAL);
1336 lod_comp->llc_pattern = pattern;
1337 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1338 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1339 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1341 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1342 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1344 lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1345 objs = &v3->lmm_objects[0];
1347 lod_set_pool(&lod_comp->llc_pool, NULL);
1348 objs = &lmm->lmm_objects[0];
1352 * If uninstantiated template component has valid l_ost_idx,
1353 * then user has specified ost list for this component.
1355 if (!lod_comp_inited(lod_comp)) {
1358 if (objs[0].l_ost_idx != (__u32)-1UL) {
1361 stripe_count = lod_comp_entry_stripe_count(
1363 if (stripe_count == 0 &&
1364 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1365 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1366 GOTO(out, rc = -E2BIG);
1368 * load the user specified ost list, when this
1369 * component is instantiated later, it will be
1370 * used in lod_alloc_ost_list().
1372 lod_comp->llc_ostlist.op_count = stripe_count;
1373 lod_comp->llc_ostlist.op_size =
1374 stripe_count * sizeof(__u32);
1375 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1376 lod_comp->llc_ostlist.op_size);
1377 if (!lod_comp->llc_ostlist.op_array)
1378 GOTO(out, rc = -ENOMEM);
1380 for (j = 0; j < stripe_count; j++)
1381 lod_comp->llc_ostlist.op_array[j] =
1382 le32_to_cpu(objs[j].l_ost_idx);
1385 * this component OST objects starts from the
1386 * first ost_idx, lod_alloc_ost_list() will
1389 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1392 * for uninstantiated component,
1393 * lmm_layout_gen stores default stripe offset.
1395 lod_comp->llc_stripe_offset =
1396 lmm->lmm_layout_gen;
1400 /* skip un-instantiated component object initialization */
1401 if (!lod_comp_inited(lod_comp))
1404 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1405 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1406 rc = lod_initialize_objects(env, lo, objs, i);
1412 if (lo->ldo_is_composite && (lvf & LVF_ALL_STALE)) {
1413 /* check the last mirror stale-ness */
1417 if (mirror_cnt == stale_mirrors) {
1419 CERROR("%s: can not set all stale mirrors for "
1421 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
1427 rc = lod_fill_mirrors(lo);
1433 lod_striping_free_nolock(env, lo);
1438 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1439 * is already cached.
1441 * \param[in] lo LOD object
1443 * \retval True if the striping is cached, otherwise
1446 static bool lod_striping_loaded(struct lod_object *lo)
1448 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1449 lo->ldo_comp_cached)
1452 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1453 if (lo->ldo_dir_stripe_loaded)
1456 /* Never load LMV stripe for slaves of striped dir */
1457 if (lo->ldo_dir_slave_stripe)
1465 * A generic function to initialize the stripe objects.
1467 * A protected version of lod_striping_load_locked() - load the striping
1468 * information from storage, parse that and instantiate LU objects to
1469 * represent the stripes. The LOD object \a lo supplies a pointer to the
1470 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1471 * return an array of references to the newly instantiated objects.
1473 * \param[in] env execution environment for this thread
1474 * \param[in,out] lo LOD object, where striping is stored and
1475 * which gets an array of references
1477 * \retval 0 if parsing and object creation succeed
1478 * \retval negative error number on failure
1480 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1482 struct lod_thread_info *info = lod_env_info(env);
1483 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1484 struct lu_buf *buf = &info->lti_buf;
1489 if (!dt_object_exists(next))
1492 if (lod_striping_loaded(lo))
1495 mutex_lock(&lo->ldo_layout_mutex);
1496 if (lod_striping_loaded(lo))
1497 GOTO(unlock, rc = 0);
1499 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1500 rc = lod_get_lov_ea(env, lo);
1505 * there is LOV EA (striping information) in this object
1506 * let's parse it and create in-core objects for the stripes
1508 buf->lb_buf = info->lti_ea_store;
1509 buf->lb_len = info->lti_ea_store_size;
1510 rc = lod_parse_striping(env, lo, buf, 0);
1512 lo->ldo_comp_cached = 1;
1513 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1514 rc = lod_get_lmv_ea(env, lo);
1515 if (rc > sizeof(struct lmv_foreign_md)) {
1516 struct lmv_foreign_md *lfm = info->lti_ea_store;
1518 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
1519 lo->ldo_foreign_lmv = info->lti_ea_store;
1520 lo->ldo_foreign_lmv_size =
1521 info->lti_ea_store_size;
1522 info->lti_ea_store = NULL;
1523 info->lti_ea_store_size = 0;
1525 lo->ldo_dir_stripe_loaded = 1;
1526 lo->ldo_is_foreign = 1;
1527 GOTO(unlock, rc = 0);
1531 if (rc < (int)sizeof(struct lmv_mds_md_v1)) {
1532 /* Let's set stripe_loaded to avoid further
1533 * stripe loading especially for non-stripe directory,
1534 * which can hurt performance. (See LU-9840)
1537 lo->ldo_dir_stripe_loaded = 1;
1538 GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1540 buf->lb_buf = info->lti_ea_store;
1541 buf->lb_len = info->lti_ea_store_size;
1542 if (rc == sizeof(struct lmv_mds_md_v1)) {
1543 rc = lod_load_lmv_shards(env, lo, buf, true);
1544 if (buf->lb_buf != info->lti_ea_store) {
1545 OBD_FREE_LARGE(info->lti_ea_store,
1546 info->lti_ea_store_size);
1547 info->lti_ea_store = buf->lb_buf;
1548 info->lti_ea_store_size = buf->lb_len;
1556 * there is LMV EA (striping information) in this object
1557 * let's parse it and create in-core objects for the stripes
1559 rc = lod_parse_dir_striping(env, lo, buf);
1561 lo->ldo_dir_stripe_loaded = 1;
1565 mutex_unlock(&lo->ldo_layout_mutex);
1570 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1571 const struct lu_buf *buf, enum layout_verify_flags lvf)
1577 mutex_lock(&lo->ldo_layout_mutex);
1578 rc = lod_parse_striping(env, lo, buf, lvf);
1579 mutex_unlock(&lo->ldo_layout_mutex);
1585 * Verify lov_user_md_v1/v3 striping.
1587 * Check the validity of all fields including the magic, stripe size,
1588 * stripe count, stripe offset and that the pool is present. Also check
1589 * that each target index points to an existing target. The additional
1590 * \a is_from_disk turns additional checks. In some cases zero fields
1591 * are allowed (like pattern=0).
1593 * \param[in] d LOD device
1594 * \param[in] buf buffer with LOV EA to verify
1595 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1596 * 1 - from disk, do not allow
1598 * \retval 0 if the striping is valid
1599 * \retval -EINVAL if striping is invalid
1601 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1604 struct lov_user_md_v1 *lum;
1605 struct lov_user_md_v3 *lum3;
1606 struct pool_desc *pool = NULL;
1610 __u16 stripe_offset;
1617 if (buf->lb_len < sizeof(*lum)) {
1618 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1620 GOTO(out, rc = -EINVAL);
1623 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1624 if (magic != LOV_USER_MAGIC_V1 &&
1625 magic != LOV_USER_MAGIC_V3 &&
1626 magic != LOV_USER_MAGIC_SPECIFIC) {
1627 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1628 le32_to_cpu(lum->lmm_magic));
1629 GOTO(out, rc = -EINVAL);
1632 /* the user uses "0" for default stripe pattern normally. */
1633 if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1634 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1636 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1637 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1638 le32_to_cpu(lum->lmm_pattern));
1639 GOTO(out, rc = -EINVAL);
1642 /* a released lum comes from creating orphan on hsm release,
1643 * doesn't make sense to verify it. */
1644 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1647 /* 64kB is the largest common page size we see (ia64), and matches the
1649 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1650 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1651 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1652 stripe_size, LOV_MIN_STRIPE_SIZE);
1653 GOTO(out, rc = -EINVAL);
1656 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1657 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1658 lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1659 /* if offset is not within valid range [0, osts_size) */
1660 if (stripe_offset >= d->lod_ost_descs.ltd_tgts_size) {
1661 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1662 stripe_offset, d->lod_ost_descs.ltd_tgts_size);
1663 GOTO(out, rc = -EINVAL);
1666 /* if lmm_stripe_offset is *not* in bitmap */
1667 if (!test_bit(stripe_offset, d->lod_ost_bitmap)) {
1668 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1670 GOTO(out, rc = -EINVAL);
1674 if (magic == LOV_USER_MAGIC_V1)
1675 lum_size = offsetof(struct lov_user_md_v1,
1677 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1678 lum_size = offsetof(struct lov_user_md_v3,
1681 GOTO(out, rc = -EINVAL);
1683 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1684 if (buf->lb_len < lum_size) {
1685 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1686 "magic %#x and stripe_count %u\n",
1687 buf->lb_len, lum_size, magic, stripe_count);
1688 GOTO(out, rc = -EINVAL);
1691 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1695 /* In the function below, .hs_keycmp resolves to
1696 * pool_hashkey_keycmp() */
1697 /* coverity[overrun-buffer-val] */
1698 pool = lod_find_pool(d, lum3->lmm_pool_name);
1702 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1703 rc = lod_check_index_in_pool(stripe_offset, pool);
1705 GOTO(out, rc = -EINVAL);
1708 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1709 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1710 stripe_count, pool_tgt_count(pool));
1711 GOTO(out, rc = -EINVAL);
1716 lod_pool_putref(pool);
1722 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1724 LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1725 LOV_USER_MAGIC_COMP_V1 ||
1726 (le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1727 LOV_USER_MAGIC_SEL, "Wrong magic %x\n",
1728 le32_to_cpu(comp->lcm_magic));
1729 LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1730 "bad index %d, max = %d\n",
1731 i, le16_to_cpu(comp->lcm_entry_count));
1733 return &comp->lcm_entries[i];
1736 #define for_each_comp_entry_v1(comp, entry) \
1737 for (entry = comp_entry_v1(comp, 0); \
1738 entry <= comp_entry_v1(comp, \
1739 le16_to_cpu(comp->lcm_entry_count) - 1); \
1742 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
1743 struct lov_comp_md_entry_v1 *dom_ent)
1745 struct lov_comp_md_entry_v1 *ent;
1747 __u32 dom_off, dom_size, comp_size, off;
1749 unsigned int size, shift;
1751 entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1752 LASSERT(entries > 0);
1753 comp_v1->lcm_entry_count = cpu_to_le16(entries);
1755 comp_size = le32_to_cpu(comp_v1->lcm_size);
1756 dom_off = le32_to_cpu(dom_ent->lcme_offset);
1757 dom_size = le32_to_cpu(dom_ent->lcme_size);
1759 /* all entries offsets are shifted by entry size at least */
1760 shift = sizeof(*dom_ent);
1761 for_each_comp_entry_v1(comp_v1, ent) {
1762 off = le32_to_cpu(ent->lcme_offset);
1763 if (off == dom_off) {
1764 /* Entry deletion creates two holes in layout data:
1765 * - hole in entries array
1766 * - hole in layout data at dom_off with dom_size
1768 * First memmove is one entry shift from next entry
1769 * start with size up to dom_off in blob
1772 src = (void *)(ent + 1);
1773 size = (unsigned long)((void *)comp_v1 + dom_off - src);
1774 memmove(dst, src, size);
1775 /* take 'off' from just moved entry */
1776 off = le32_to_cpu(ent->lcme_offset);
1777 /* second memmove is blob tail after 'off' up to
1780 dst = (void *)comp_v1 + dom_off - sizeof(*ent);
1781 src = (void *)comp_v1 + off;
1782 size = (unsigned long)(comp_size - off);
1783 memmove(dst, src, size);
1784 /* all entries offsets after DoM entry are shifted by
1785 * dom_size additionally
1789 ent->lcme_offset = cpu_to_le32(off - shift);
1791 comp_v1->lcm_size = cpu_to_le32(comp_size - shift);
1793 /* notify a caller to re-check entry */
1797 void lod_dom_stripesize_recalc(struct lod_device *d)
1799 __u64 threshold_mb = d->lod_dom_threshold_free_mb;
1800 __u32 max_size = d->lod_dom_stripesize_max_kb;
1801 __u32 def_size = d->lod_dom_stripesize_cur_kb;
1803 /* use maximum allowed value if free space is above threshold */
1804 if (d->lod_lsfs_free_mb >= threshold_mb) {
1805 def_size = max_size;
1806 } else if (!d->lod_lsfs_free_mb || max_size <= LOD_DOM_MIN_SIZE_KB) {
1809 /* recalc threshold like it would be with def_size as max */
1810 threshold_mb = mult_frac(threshold_mb, def_size, max_size);
1811 if (d->lod_lsfs_free_mb < threshold_mb)
1812 def_size = rounddown(def_size / 2, LOD_DOM_MIN_SIZE_KB);
1813 else if (d->lod_lsfs_free_mb > threshold_mb * 2)
1814 def_size = max_t(unsigned int, def_size * 2,
1815 LOD_DOM_MIN_SIZE_KB);
1818 if (d->lod_dom_stripesize_cur_kb != def_size) {
1819 CDEBUG(D_LAYOUT, "Change default DOM stripe size %d->%d\n",
1820 d->lod_dom_stripesize_cur_kb, def_size);
1821 d->lod_dom_stripesize_cur_kb = def_size;
1825 static __u32 lod_dom_stripesize_limit(const struct lu_env *env,
1826 struct lod_device *d)
1830 /* set bfree as fraction of total space */
1831 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_SPOOF)) {
1832 spin_lock(&d->lod_lsfs_lock);
1833 d->lod_lsfs_free_mb = mult_frac(d->lod_lsfs_total_mb,
1834 min_t(int, cfs_fail_val, 100), 100);
1835 GOTO(recalc, rc = 0);
1838 if (d->lod_lsfs_age < ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE) {
1839 struct obd_statfs sfs;
1841 spin_lock(&d->lod_lsfs_lock);
1842 if (d->lod_lsfs_age > ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE)
1843 GOTO(unlock, rc = 0);
1845 d->lod_lsfs_age = ktime_get_seconds();
1846 spin_unlock(&d->lod_lsfs_lock);
1847 rc = dt_statfs(env, d->lod_child, &sfs);
1850 "%s: failed to get OSD statfs: rc = %d\n",
1851 lod2obd(d)->obd_name, rc);
1854 /* udpate local OSD cached statfs data */
1855 spin_lock(&d->lod_lsfs_lock);
1856 d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
1857 d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
1859 lod_dom_stripesize_recalc(d);
1861 spin_unlock(&d->lod_lsfs_lock);
1864 return d->lod_dom_stripesize_cur_kb << 10;
1867 int lod_dom_stripesize_choose(const struct lu_env *env, struct lod_device *d,
1868 struct lov_comp_md_v1 *comp_v1,
1869 struct lov_comp_md_entry_v1 *dom_ent,
1872 struct lov_comp_md_entry_v1 *ent;
1873 struct lu_extent *dom_ext, *ext;
1874 struct lov_user_md_v1 *lum;
1875 __u32 max_stripe_size;
1878 bool dom_next_entry = false;
1880 dom_ext = &dom_ent->lcme_extent;
1881 dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1882 max_stripe_size = lod_dom_stripesize_limit(env, d);
1884 /* Check stripe size againts current per-MDT limit */
1885 if (stripe_size <= max_stripe_size)
1888 lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1889 CDEBUG(D_LAYOUT, "overwrite DoM component size %u with MDT limit %u\n",
1890 stripe_size, max_stripe_size);
1891 lum->lmm_stripe_size = cpu_to_le32(max_stripe_size);
1893 /* In common case the DoM stripe is first entry in a mirror and
1894 * can be deleted only if it is not single entry in layout or
1895 * mirror, otherwise error should be returned.
1897 for_each_comp_entry_v1(comp_v1, ent) {
1901 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1905 ext = &ent->lcme_extent;
1906 if (ext->e_start != dom_ext->e_end)
1909 /* Found next component after the DoM one with the same
1910 * mirror_id and adjust its start with DoM component end.
1912 * NOTE: we are considering here that there can be only one
1913 * DoM component in a file, all replicas are located on OSTs
1914 * always and don't need adjustment since use own layouts.
1916 ext->e_start = cpu_to_le64(max_stripe_size);
1917 dom_next_entry = true;
1921 if (max_stripe_size == 0) {
1922 /* DoM component size is zero due to server setting, remove
1923 * it from the layout but only if next component exists in
1924 * the same mirror. That must be checked prior calling the
1925 * lod_erase_dom_stripe().
1927 if (!dom_next_entry)
1930 rc = lod_erase_dom_stripe(comp_v1, dom_ent);
1932 /* Update DoM extent end finally */
1933 dom_ext->e_end = cpu_to_le64(max_stripe_size);
1940 * Verify LOV striping.
1942 * \param[in] d LOD device
1943 * \param[in] buf buffer with LOV EA to verify
1944 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1945 * 1 - from disk, do not allow
1946 * \param[in] start extent start for composite layout
1948 * \retval 0 if the striping is valid
1949 * \retval -EINVAL if striping is invalid
1951 int lod_verify_striping(const struct lu_env *env, struct lod_device *d,
1952 struct lod_object *lo, const struct lu_buf *buf,
1955 struct lov_user_md_v1 *lum;
1956 struct lov_comp_md_v1 *comp_v1;
1957 struct lov_comp_md_entry_v1 *ent;
1958 struct lu_extent *ext;
1961 __u32 stripe_size = 0;
1962 __u16 prev_mid = -1, mirror_id = -1;
1968 if (buf->lb_len < sizeof(lum->lmm_magic)) {
1969 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
1975 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1976 /* treat foreign LOV EA/object case first
1977 * XXX is it expected to try setting again a foreign?
1978 * XXX should we care about different current vs new layouts ?
1980 if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
1981 struct lov_foreign_md *lfm = buf->lb_buf;
1983 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
1985 "buf len %zu < min lov_foreign_md size (%zu)\n",
1986 buf->lb_len, offsetof(typeof(*lfm),
1991 if (foreign_size_le(lfm) > buf->lb_len) {
1993 "buf len %zu < this lov_foreign_md size (%zu)\n",
1994 buf->lb_len, foreign_size_le(lfm));
1997 /* Don't do anything with foreign layouts */
2001 /* normal LOV/layout cases */
2003 if (buf->lb_len < sizeof(*lum)) {
2004 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
2010 case LOV_USER_MAGIC_FOREIGN:
2012 case LOV_USER_MAGIC_V1:
2013 case LOV_USER_MAGIC_V3:
2014 case LOV_USER_MAGIC_SPECIFIC:
2015 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
2016 case LOV_USER_MAGIC_COMP_V1:
2017 case LOV_USER_MAGIC_SEL:
2020 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
2021 le32_to_cpu(lum->lmm_magic));
2025 /* magic == LOV_USER_MAGIC_COMP_V1 */
2026 comp_v1 = buf->lb_buf;
2027 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
2028 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
2029 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
2035 if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
2036 CDEBUG(D_LAYOUT, "entry count is zero\n");
2040 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
2041 lo->ldo_comp_cnt > 0) {
2042 /* could be called from lustre.lov.add */
2043 __u32 cnt = lo->ldo_comp_cnt;
2045 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
2046 prev_end = ext->e_end;
2051 for_each_comp_entry_v1(comp_v1, ent) {
2052 ext = &ent->lcme_extent;
2054 if (le64_to_cpu(ext->e_start) > le64_to_cpu(ext->e_end) ||
2055 le64_to_cpu(ext->e_start) & (LOV_MIN_STRIPE_SIZE - 1) ||
2056 (le64_to_cpu(ext->e_end) != LUSTRE_EOF &&
2057 le64_to_cpu(ext->e_end) & (LOV_MIN_STRIPE_SIZE - 1))) {
2058 CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
2059 le64_to_cpu(ext->e_start),
2060 le64_to_cpu(ext->e_end));
2065 /* lcme_id contains valid value */
2066 if (le32_to_cpu(ent->lcme_id) == 0 ||
2067 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
2068 CDEBUG(D_LAYOUT, "invalid id %u\n",
2069 le32_to_cpu(ent->lcme_id));
2073 if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
2074 mirror_id = mirror_id_of(
2075 le32_to_cpu(ent->lcme_id));
2077 /* first component must start with 0 */
2078 if (mirror_id != prev_mid &&
2079 le64_to_cpu(ext->e_start) != 0) {
2081 "invalid start:%llu, expect:0\n",
2082 le64_to_cpu(ext->e_start));
2086 prev_mid = mirror_id;
2090 if (le64_to_cpu(ext->e_start) == 0) {
2095 /* the next must be adjacent with the previous one */
2096 if (le64_to_cpu(ext->e_start) != prev_end) {
2098 "invalid start actual:%llu, expect:%llu\n",
2099 le64_to_cpu(ext->e_start), prev_end);
2103 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
2104 tmp.lb_len = le32_to_cpu(ent->lcme_size);
2106 /* Check DoM entry is always the first one */
2108 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
2110 /* DoM component must be the first in a mirror */
2111 if (le64_to_cpu(ext->e_start) > 0) {
2112 CDEBUG(D_LAYOUT, "invalid DoM component "
2113 "with %llu extent start\n",
2114 le64_to_cpu(ext->e_start));
2117 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2118 /* There is just one stripe on MDT and it must
2119 * cover whole component size. */
2120 if (stripe_size != le64_to_cpu(ext->e_end)) {
2121 CDEBUG(D_LAYOUT, "invalid DoM layout "
2122 "stripe size %u != %llu "
2123 "(component size)\n",
2124 stripe_size, prev_end);
2127 /* Check and adjust stripe size by per-MDT limit */
2128 rc = lod_dom_stripesize_choose(env, d, comp_v1, ent,
2130 /* DoM entry was removed, re-check layout from start */
2131 if (rc == -ERESTART)
2136 if (le16_to_cpu(lum->lmm_stripe_count) == 1)
2137 lum->lmm_stripe_count = 0;
2138 /* Any stripe count is forbidden on DoM component */
2139 if (lum->lmm_stripe_count > 0) {
2141 "invalid DoM layout stripe count %u, must be 0\n",
2142 le16_to_cpu(lum->lmm_stripe_count));
2146 /* Any pool is forbidden on DoM component */
2147 if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
2148 struct lov_user_md_v3 *v3 = (void *)lum;
2150 if (v3->lmm_pool_name[0] != '\0') {
2152 "DoM component cannot have pool assigned\n");
2158 prev_end = le64_to_cpu(ext->e_end);
2160 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
2164 if (prev_end == LUSTRE_EOF || ext->e_start == prev_end)
2167 /* extent end must be aligned with the stripe_size */
2168 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2169 if (stripe_size && prev_end % stripe_size) {
2170 CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
2171 "stripe_sz: %u, [%llu, %llu)\n",
2172 stripe_size, ext->e_start, prev_end);
2177 /* make sure that the mirror_count is telling the truth */
2178 if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
2185 * set the default stripe size, if unset.
2187 * \param[in,out] val number of bytes per OST stripe
2189 * The minimum stripe size is 64KB to ensure that a single stripe is an
2190 * even multiple of a client PAGE_SIZE (IA64, PPC, etc). Otherwise, it
2191 * is difficult to split dirty pages across OSCs during writes.
2193 void lod_fix_desc_stripe_size(__u64 *val)
2195 if (*val < LOV_MIN_STRIPE_SIZE) {
2197 LCONSOLE_INFO("Increasing default stripe size to "
2198 "minimum value %u\n",
2199 LOV_DESC_STRIPE_SIZE_DEFAULT);
2200 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
2201 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
2202 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
2203 LCONSOLE_WARN("Changing default stripe size to %llu (a "
2204 "multiple of %u)\n",
2205 *val, LOV_MIN_STRIPE_SIZE);
2210 * set the filesystem default number of stripes, if unset.
2212 * \param[in,out] val number of stripes
2214 * A value of "0" means "use the system-wide default stripe count", which
2215 * has either been inherited by now, or falls back to 1 stripe per file.
2216 * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
2217 * and is a valid value, so is left unchanged here.
2219 void lod_fix_desc_stripe_count(__u32 *val)
2226 * set the filesystem default layout pattern
2228 * \param[in,out] val LOV_PATTERN_* layout
2230 * A value of "0" means "use the system-wide default layout type", which
2231 * has either been inherited by now, or falls back to plain RAID0 striping.
2233 void lod_fix_desc_pattern(__u32 *val)
2235 /* from lov_setstripe */
2236 if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
2237 LCONSOLE_WARN("lod: Unknown stripe pattern: %#x\n", *val);
2242 void lod_fix_lmv_desc_pattern(__u32 *val)
2244 if ((*val) && !lmv_is_known_hash_type(*val)) {
2245 LCONSOLE_WARN("lod: Unknown md stripe pattern: %#x\n", *val);
2250 void lod_fix_desc_qos_maxage(__u32 *val)
2252 /* fix qos_maxage */
2254 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2258 * Used to fix insane default striping.
2260 * \param[in] desc striping description
2262 void lod_fix_desc(struct lov_desc *desc)
2264 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2265 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2266 lod_fix_desc_pattern(&desc->ld_pattern);
2267 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2270 static void lod_fix_lmv_desc(struct lmv_desc *desc)
2272 desc->ld_active_tgt_count = 0;
2273 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2274 lod_fix_lmv_desc_pattern(&desc->ld_pattern);
2275 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2279 * Initialize the structures used to store pools and default striping.
2281 * \param[in] lod LOD device
2282 * \param[in] lcfg configuration structure storing default striping.
2284 * \retval 0 if initialization succeeds
2285 * \retval negative error number on failure
2287 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2289 struct obd_device *obd;
2290 struct lov_desc *desc;
2294 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2295 LASSERT(obd != NULL);
2296 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2298 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2299 CERROR("LOD setup requires a descriptor\n");
2303 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2305 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2306 CERROR("descriptor size wrong: %d > %d\n",
2307 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2311 if (desc->ld_magic != LOV_DESC_MAGIC) {
2312 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2313 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2314 obd->obd_name, desc);
2315 lustre_swab_lov_desc(desc);
2317 CERROR("%s: Bad lov desc magic: %#x\n",
2318 obd->obd_name, desc->ld_magic);
2325 desc->ld_active_tgt_count = 0;
2326 lod->lod_ost_descs.ltd_lov_desc = *desc;
2328 /* NB: config doesn't contain lmv_desc, alter it via sysfs. */
2329 lod_fix_lmv_desc(&lod->lod_mdt_descs.ltd_lmv_desc);
2331 lod->lod_sp_me = LUSTRE_SP_CLI;
2333 /* Set up OST pool environment */
2334 lod->lod_pool_count = 0;
2335 rc = lod_pool_hash_init(&lod->lod_pools_hash_body);
2339 INIT_LIST_HEAD(&lod->lod_pool_list);
2340 lod->lod_pool_count = 0;
2341 rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_tgt_pool, 0);
2345 rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool, 0);
2347 GOTO(out_mdt_pool, rc);
2349 rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_tgt_pool, 0);
2351 GOTO(out_mdt_rr_pool, rc);
2353 rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
2355 GOTO(out_ost_pool, rc);
2360 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2362 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2364 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
2366 lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2372 * Release the structures describing the pools.
2374 * \param[in] lod LOD device from which we release the structures
2378 int lod_pools_fini(struct lod_device *lod)
2380 struct obd_device *obd = lod2obd(lod);
2381 struct pool_desc *pool, *tmp;
2384 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2385 /* free pool structs */
2386 CDEBUG(D_INFO, "delete pool %p\n", pool);
2387 /* In the function below, .hs_keycmp resolves to
2388 * pool_hashkey_keycmp() */
2389 /* coverity[overrun-buffer-val] */
2390 lod_pool_del(obd, pool->pool_name);
2393 lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2394 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool);
2395 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2396 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2397 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);