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_inited(lod_comp);
617 /* calculate component preference over all used OSTs */
618 for (j = 0; init && j < lod_comp->llc_stripes_allocated; j++) {
619 __u32 idx = lod_comp->llc_ost_indices[j];
620 struct lod_tgt_desc *ltd;
622 if (unlikely(idx > lod->lod_ost_descs.ltd_tgts_size)) {
623 CERROR("%s: "DFID" OST idx %u > max %u\n",
624 lod2obd(lod)->obd_name,
625 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
626 idx, lod->lod_ost_descs.ltd_tgts_size);
629 ltd = OST_TGT(lod, idx);
630 if (unlikely(!ltd)) {
631 CERROR("%s: "DFID" OST idx %u is NULL\n",
632 lod2obd(lod)->obd_name,
633 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
638 if (ltd->ltd_statfs.os_state & OS_STATFS_NONROT)
642 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
643 lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
644 lo->ldo_mirrors[mirror_idx].lme_prefer |= preferred;
645 lo->ldo_mirrors[mirror_idx].lme_preference += pref;
646 lo->ldo_mirrors[mirror_idx].lme_end = i;
650 if (mirror_idx >= 0 && preferred &&
651 !lo->ldo_mirrors[mirror_idx].lme_stale)
652 found_preferred = true;
656 if (mirror_idx >= lo->ldo_mirror_count)
659 mirror_id = mirror_id_of(lod_comp->llc_id);
661 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
662 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
663 lo->ldo_mirrors[mirror_idx].lme_prefer = preferred;
664 lo->ldo_mirrors[mirror_idx].lme_preference = pref;
665 lo->ldo_mirrors[mirror_idx].lme_start = i;
666 lo->ldo_mirrors[mirror_idx].lme_end = i;
668 if (mirror_idx != lo->ldo_mirror_count - 1)
671 if (!found_preferred && mirror_idx > 0) {
675 * if no explicited preferred found, then find a mirror
676 * with higher number of non-rotational OSTs
679 for (i = 0; i <= mirror_idx; i++) {
680 if (lo->ldo_mirrors[i].lme_stale)
682 if (lo->ldo_mirrors[i].lme_preference > pref) {
683 pref = lo->ldo_mirrors[i].lme_preference;
689 lo->ldo_mirrors[best].lme_prefer = 1;
696 * Generate on-disk lov_mds_md structure for each layout component based on
697 * the information in lod_object->ldo_comp_entries[i].
699 * \param[in] env execution environment for this thread
700 * \param[in] lo LOD object
701 * \param[in] comp_idx index of ldo_comp_entries
702 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
703 * \param[in|out] lmm_size buffer size/lmm size
704 * \param[in] is_dir generate lov ea for dir or file? For dir case,
705 * the stripe info is from the default stripe
706 * template, which is collected in lod_ah_init(),
707 * either from parent object or root object; for
708 * file case, it's from the @lo object
710 * \retval 0 if on disk structure is created successfully
711 * \retval negative error number on failure
713 static int lod_gen_component_ea(const struct lu_env *env,
714 struct lod_object *lo, int comp_idx,
715 struct lov_mds_md *lmm, int *lmm_size,
718 struct lod_thread_info *info = lod_env_info(env);
719 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
720 struct lod_device *lod;
721 struct lov_ost_data_v1 *objs;
722 struct lod_layout_component *lod_comp;
731 &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
733 lod_comp = &lo->ldo_comp_entries[comp_idx];
735 magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
736 if (lod_comp->llc_pattern == 0) /* default striping */
737 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
739 lmm->lmm_magic = cpu_to_le32(magic);
740 lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
741 fid_to_lmm_oi(fid, &lmm->lmm_oi);
742 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
743 lmm->lmm_oi.oi.oi_id++;
744 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
746 lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
747 lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
749 * for dir and uninstantiated component, lmm_layout_gen stores
750 * default stripe offset.
752 lmm->lmm_layout_gen =
753 (is_dir || !lod_comp_inited(lod_comp)) ?
754 cpu_to_le16(lod_comp->llc_stripe_offset) :
755 cpu_to_le16(lod_comp->llc_layout_gen);
757 if (magic == LOV_MAGIC_V1) {
758 objs = &lmm->lmm_objects[0];
760 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
761 size_t cplen = strlcpy(v3->lmm_pool_name,
763 sizeof(v3->lmm_pool_name));
764 if (cplen >= sizeof(v3->lmm_pool_name))
766 objs = &v3->lmm_objects[0];
768 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
769 stripe_count = lod_comp_entry_stripe_count(lo, comp_idx, is_dir);
770 if (stripe_count == 0 && !is_dir &&
771 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
772 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
773 /* Try again if all active targets are disconnected.
774 * It is possible when MDS does failover. */
775 if (!lod->lod_ost_active_count &&
781 if (!is_dir && lo->ldo_is_composite)
782 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
784 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
787 /* generate ost_idx of this component stripe */
788 for (i = 0; i < stripe_count; i++) {
789 struct dt_object *object;
790 __u32 ost_idx = (__u32)-1UL;
791 int type = LU_SEQ_RANGE_OST;
793 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
794 object = lod_comp->llc_stripe[i];
795 /* instantiated component */
796 info->lti_fid = *lu_object_fid(&object->do_lu);
798 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
800 if (cfs_fail_val == 0)
801 cfs_fail_val = info->lti_fid.f_oid;
803 info->lti_fid.f_oid = cfs_fail_val;
806 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
809 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
810 objs[i].l_ost_gen = cpu_to_le32(0);
811 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
814 rc = lod_fld_lookup(env, lod, &info->lti_fid,
817 CERROR("%s: Can not locate "DFID": rc = %d\n",
818 lod2obd(lod)->obd_name,
819 PFID(&info->lti_fid), rc);
822 } else if (lod_comp->llc_ostlist.op_array &&
823 lod_comp->llc_ostlist.op_count) {
824 /* user specified ost list */
825 ost_idx = lod_comp->llc_ostlist.op_array[i];
828 * with un-instantiated or with no specified ost list
829 * component, its l_ost_idx does not matter.
831 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
834 if (lmm_size != NULL)
835 *lmm_size = lov_mds_md_size(stripe_count, magic);
840 * Generate on-disk lov_mds_md structure based on the information in
841 * the lod_object->ldo_comp_entries.
843 * \param[in] env execution environment for this thread
844 * \param[in] lo LOD object
845 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
846 * \param[in|out] lmm_size buffer size/lmm size
847 * \param[in] is_dir generate lov ea for dir or file? For dir case,
848 * the stripe info is from the default stripe
849 * template, which is collected in lod_ah_init(),
850 * either from parent object or root object; for
851 * file case, it's from the @lo object
853 * \retval 0 if on disk structure is created successfully
854 * \retval negative error number on failure
856 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
857 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
859 struct lov_comp_md_entry_v1 *lcme;
860 struct lov_comp_md_v1 *lcm;
861 struct lod_layout_component *comp_entries;
862 __u16 comp_cnt, mirror_cnt;
863 bool is_composite, is_foreign = false;
864 int i, rc = 0, offset;
868 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
869 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
870 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
872 lo->ldo_def_striping->lds_def_striping_is_composite;
874 comp_cnt = lo->ldo_comp_cnt;
875 mirror_cnt = lo->ldo_mirror_count;
876 comp_entries = lo->ldo_comp_entries;
877 is_composite = lo->ldo_is_composite;
878 is_foreign = lo->ldo_is_foreign;
881 LASSERT(lmm_size != NULL);
884 struct lov_foreign_md *lfm;
886 lfm = (struct lov_foreign_md *)lmm;
887 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
888 /* need to store little-endian */
889 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
890 __swab32s(&lfm->lfm_magic);
891 __swab32s(&lfm->lfm_length);
892 __swab32s(&lfm->lfm_type);
893 __swab32s(&lfm->lfm_flags);
895 *lmm_size = lo->ldo_foreign_lov_size;
899 LASSERT(comp_cnt != 0 && comp_entries != NULL);
902 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
906 lcm = (struct lov_comp_md_v1 *)lmm;
907 memset(lcm, 0, sizeof(*lcm));
909 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
910 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
911 lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
912 lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
914 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
915 LASSERT(offset % sizeof(__u64) == 0);
917 for (i = 0; i < comp_cnt; i++) {
918 struct lod_layout_component *lod_comp;
919 struct lov_mds_md *sub_md;
922 lod_comp = &comp_entries[i];
923 lcme = &lcm->lcm_entries[i];
925 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
926 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
928 /* component could be un-inistantiated */
929 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
930 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
931 lcme->lcme_timestamp =
932 cpu_to_le64(lod_comp->llc_timestamp);
933 if (lod_comp->llc_flags & LCME_FL_EXTENSION && !is_dir)
934 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
936 lcme->lcme_extent.e_start =
937 cpu_to_le64(lod_comp->llc_extent.e_start);
938 lcme->lcme_extent.e_end =
939 cpu_to_le64(lod_comp->llc_extent.e_end);
940 lcme->lcme_offset = cpu_to_le32(offset);
942 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
943 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
946 lcme->lcme_size = cpu_to_le32(size);
948 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
949 "offset:%d lmm_size:%d\n", offset, *lmm_size);
951 lcm->lcm_size = cpu_to_le32(offset);
952 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
954 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
965 * Fill lti_ea_store buffer in the environment with a value for the given
966 * EA. The buffer is reallocated if the value doesn't fit.
968 * \param[in,out] env execution environment for this thread
969 * .lti_ea_store buffer is filled with EA's value
970 * \param[in] lo LOD object
971 * \param[in] name name of the EA
973 * \retval > 0 if EA is fetched successfully
974 * \retval 0 if EA is empty
975 * \retval negative error number on failure
977 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
980 struct lod_thread_info *info = lod_env_info(env);
981 struct dt_object *next = dt_object_child(&lo->ldo_obj);
987 if (unlikely(info->lti_ea_store == NULL)) {
988 /* just to enter in allocation block below */
992 info->lti_buf.lb_buf = info->lti_ea_store;
993 info->lti_buf.lb_len = info->lti_ea_store_size;
994 rc = dt_xattr_get(env, next, &info->lti_buf, name);
997 /* if object is not striped or inaccessible */
998 if (rc == -ENODATA || rc == -ENOENT)
1001 if (rc == -ERANGE) {
1002 /* EA doesn't fit, reallocate new buffer */
1003 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
1004 if (rc == -ENODATA || rc == -ENOENT)
1010 rc = lod_ea_store_resize(info, rc);
1020 * Verify the target index is present in the current configuration.
1022 * \param[in] md LOD device where the target table is stored
1023 * \param[in] idx target's index
1025 * \retval 0 if the index is present
1026 * \retval -EINVAL if not
1028 int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1030 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1031 !test_bit(idx, md->lod_ost_bitmap))) {
1032 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1033 md->lod_ost_descs.ltd_tgts_size);
1037 if (unlikely(OST_TGT(md, idx) == NULL)) {
1038 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1039 lod2obd(md)->obd_name, idx);
1043 if (unlikely(OST_TGT(md, idx)->ltd_tgt == NULL)) {
1044 CERROR("%s: invalid lod device, for idx: %d\n",
1045 lod2obd(md)->obd_name , idx);
1053 * Instantiate objects for stripes.
1055 * Allocate and initialize LU-objects representing the stripes. The number
1056 * of the stripes (llc_stripe_count) must be initialized already. The caller
1057 * must ensure nobody else is calling the function on the object at the same
1058 * time. FLDB service must be running to be able to map a FID to the targets
1059 * and find appropriate device representing that target.
1061 * \param[in] env execution environment for this thread
1062 * \param[in,out] lo LOD object
1063 * \param[in] objs an array of IDs to creates the objects from
1064 * \param[in] comp_idx index of ldo_comp_entries
1066 * \retval 0 if the objects are instantiated successfully
1067 * \retval negative error number on failure
1069 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1070 struct lov_ost_data_v1 *objs, int comp_idx)
1072 struct lod_layout_component *lod_comp;
1073 struct lod_thread_info *info = lod_env_info(env);
1074 struct lod_device *md;
1075 struct lu_object *o, *n;
1076 struct lu_device *nd;
1077 struct dt_object **stripe = NULL;
1078 __u32 *ost_indices = NULL;
1084 LASSERT(lo != NULL);
1085 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1087 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1088 lod_comp = &lo->ldo_comp_entries[comp_idx];
1090 LASSERT(lod_comp->llc_stripe == NULL);
1091 LASSERT(lod_comp->llc_stripe_count > 0);
1092 LASSERT(lod_comp->llc_stripe_size > 0);
1094 stripe_len = lod_comp->llc_stripe_count;
1095 OBD_ALLOC_PTR_ARRAY(stripe, stripe_len);
1098 OBD_ALLOC_PTR_ARRAY(ost_indices, stripe_len);
1100 GOTO(out, rc = -ENOMEM);
1102 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1103 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1106 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1107 idx = le32_to_cpu(objs[i].l_ost_idx);
1108 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1111 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1112 PFID(&info->lti_fid));
1113 lod_getref(&md->lod_ost_descs);
1115 rc = validate_lod_and_idx(md, idx);
1116 if (unlikely(rc != 0)) {
1117 lod_putref(md, &md->lod_ost_descs);
1121 nd = &OST_TGT(md, idx)->ltd_tgt->dd_lu_dev;
1122 lod_putref(md, &md->lod_ost_descs);
1124 /* In the function below, .hs_keycmp resolves to
1125 * u_obj_hop_keycmp() */
1126 /* coverity[overrun-buffer-val] */
1127 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1129 GOTO(out, rc = PTR_ERR(o));
1131 n = lu_object_locate(o->lo_header, nd->ld_type);
1134 stripe[i] = container_of(n, struct dt_object, do_lu);
1135 ost_indices[i] = idx;
1140 for (i = 0; i < stripe_len; i++)
1141 if (stripe[i] != NULL)
1142 dt_object_put(env, stripe[i]);
1144 OBD_FREE_PTR_ARRAY(stripe, stripe_len);
1145 lod_comp->llc_stripe_count = 0;
1147 OBD_FREE_PTR_ARRAY(ost_indices, stripe_len);
1149 lod_comp->llc_stripe = stripe;
1150 lod_comp->llc_ost_indices = ost_indices;
1151 lod_comp->llc_stripes_allocated = stripe_len;
1158 * Instantiate objects for striping.
1160 * Parse striping information in \a buf and instantiate the objects
1161 * representing the stripes.
1163 * \param[in] env execution environment for this thread
1164 * \param[in] lo LOD object
1165 * \param[in] buf buffer storing LOV EA to parse
1166 * \param[in] lvf verify flags when parsing the layout
1168 * \retval 0 if parsing and objects creation succeed
1169 * \retval negative error number on failure
1171 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1172 const struct lu_buf *buf, enum layout_verify_flags lvf)
1174 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1175 struct lov_mds_md_v1 *lmm;
1176 struct lov_comp_md_v1 *comp_v1 = NULL;
1177 struct lov_foreign_md *foreign = NULL;
1178 struct lov_ost_data_v1 *objs;
1179 __u32 magic, pattern;
1180 __u16 mirror_cnt = 0;
1183 __u16 mirror_id = MIRROR_ID_NEG;
1185 int stale_mirrors = 0;
1189 LASSERT(buf->lb_buf);
1190 LASSERT(buf->lb_len);
1191 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1193 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1194 magic = le32_to_cpu(lmm->lmm_magic);
1196 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1197 magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN &&
1198 magic != LOV_MAGIC_SEL)
1199 GOTO(out, rc = -EINVAL);
1201 lod_striping_free_nolock(env, lo);
1203 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1204 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1205 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1207 GOTO(out, rc = -EINVAL);
1208 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1209 lo->ldo_is_composite = 1;
1210 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1212 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1215 lo->ldo_flr_state = LCM_FL_NONE;
1216 } else if (magic == LOV_MAGIC_FOREIGN) {
1219 foreign = (struct lov_foreign_md *)buf->lb_buf;
1220 length = offsetof(typeof(*foreign), lfm_value);
1221 if (buf->lb_len < length ||
1222 buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1224 "buf len %zu too small for lov_foreign_md\n",
1226 GOTO(out, rc = -EINVAL);
1229 /* just cache foreign LOV EA raw */
1230 rc = lod_alloc_foreign_lov(lo, length);
1233 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1237 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1238 lo->ldo_is_composite = 0;
1241 rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1245 for (i = 0; i < comp_cnt; i++) {
1246 struct lod_layout_component *lod_comp;
1247 struct lu_extent *ext;
1250 lod_comp = &lo->ldo_comp_entries[i];
1251 if (lo->ldo_is_composite) {
1252 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1253 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1255 ext = &comp_v1->lcm_entries[i].lcme_extent;
1256 lod_comp->llc_extent.e_start =
1257 le64_to_cpu(ext->e_start);
1258 if (lod_comp->llc_extent.e_start &
1259 (LOV_MIN_STRIPE_SIZE - 1)) {
1261 "extent start %llu is not a multiple of min size %u\n",
1262 lod_comp->llc_extent.e_start,
1263 LOV_MIN_STRIPE_SIZE);
1264 GOTO(out, rc = -EINVAL);
1267 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1268 if (lod_comp->llc_extent.e_end != LUSTRE_EOF &&
1269 lod_comp->llc_extent.e_end &
1270 (LOV_MIN_STRIPE_SIZE - 1)) {
1272 "extent end %llu is not a multiple of min size %u\n",
1273 lod_comp->llc_extent.e_end,
1274 LOV_MIN_STRIPE_SIZE);
1275 GOTO(out, rc = -EINVAL);
1278 lod_comp->llc_flags =
1279 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1281 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1282 lod_comp->llc_timestamp = le64_to_cpu(
1283 comp_v1->lcm_entries[i].lcme_timestamp);
1285 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1286 if (lod_comp->llc_id == LCME_ID_INVAL)
1287 GOTO(out, rc = -EINVAL);
1289 if (lvf & LVF_ALL_STALE) {
1290 if (mirror_id_of(lod_comp->llc_id) ==
1292 /* remaining comps in the mirror */
1293 stale |= lod_comp->llc_flags &
1297 * new mirror, check last mirror's
1304 mirror_id_of(lod_comp->llc_id);
1306 /* the first comp of the new mirror */
1307 stale = lod_comp->llc_flags &
1312 if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
1313 comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
1314 CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
1315 lod2obd(d)->obd_name,
1316 lod_comp->llc_flags, lod_comp->llc_id, i,
1317 PFID(lod_object_fid(lo)),
1318 le32_to_cpu(comp_v1->lcm_magic));
1321 lod_comp_set_init(lod_comp);
1324 pattern = le32_to_cpu(lmm->lmm_pattern);
1325 if (!lov_pattern_supported(lov_pattern(pattern)))
1326 GOTO(out, rc = -EINVAL);
1328 lod_comp->llc_pattern = pattern;
1329 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1330 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1331 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1333 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1334 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1336 lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1337 objs = &v3->lmm_objects[0];
1339 lod_set_pool(&lod_comp->llc_pool, NULL);
1340 objs = &lmm->lmm_objects[0];
1344 * If uninstantiated template component has valid l_ost_idx,
1345 * then user has specified ost list for this component.
1347 if (!lod_comp_inited(lod_comp)) {
1350 if (objs[0].l_ost_idx != (__u32)-1UL) {
1353 stripe_count = lod_comp_entry_stripe_count(
1355 if (stripe_count == 0 &&
1356 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1357 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1358 GOTO(out, rc = -E2BIG);
1360 * load the user specified ost list, when this
1361 * component is instantiated later, it will be
1362 * used in lod_alloc_ost_list().
1364 lod_comp->llc_ostlist.op_count = stripe_count;
1365 lod_comp->llc_ostlist.op_size =
1366 stripe_count * sizeof(__u32);
1367 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1368 lod_comp->llc_ostlist.op_size);
1369 if (!lod_comp->llc_ostlist.op_array)
1370 GOTO(out, rc = -ENOMEM);
1372 for (j = 0; j < stripe_count; j++)
1373 lod_comp->llc_ostlist.op_array[j] =
1374 le32_to_cpu(objs[j].l_ost_idx);
1377 * this component OST objects starts from the
1378 * first ost_idx, lod_alloc_ost_list() will
1381 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1384 * for uninstantiated component,
1385 * lmm_layout_gen stores default stripe offset.
1387 lod_comp->llc_stripe_offset =
1388 lmm->lmm_layout_gen;
1392 /* skip un-instantiated component object initialization */
1393 if (!lod_comp_inited(lod_comp))
1396 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1397 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1398 rc = lod_initialize_objects(env, lo, objs, i);
1404 if (lo->ldo_is_composite && (lvf & LVF_ALL_STALE)) {
1405 /* check the last mirror stale-ness */
1409 if (mirror_cnt == stale_mirrors) {
1411 CERROR("%s: can not set all stale mirrors for "
1413 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
1419 rc = lod_fill_mirrors(lo);
1425 lod_striping_free_nolock(env, lo);
1430 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1431 * is already cached.
1433 * \param[in] lo LOD object
1435 * \retval True if the striping is cached, otherwise
1438 static bool lod_striping_loaded(struct lod_object *lo)
1440 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1441 lo->ldo_comp_cached)
1444 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1445 if (lo->ldo_dir_stripe_loaded)
1448 /* Never load LMV stripe for slaves of striped dir */
1449 if (lo->ldo_dir_slave_stripe)
1457 * A generic function to initialize the stripe objects.
1459 * A protected version of lod_striping_load_locked() - load the striping
1460 * information from storage, parse that and instantiate LU objects to
1461 * represent the stripes. The LOD object \a lo supplies a pointer to the
1462 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1463 * return an array of references to the newly instantiated objects.
1465 * \param[in] env execution environment for this thread
1466 * \param[in,out] lo LOD object, where striping is stored and
1467 * which gets an array of references
1469 * \retval 0 if parsing and object creation succeed
1470 * \retval negative error number on failure
1472 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1474 struct lod_thread_info *info = lod_env_info(env);
1475 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1476 struct lu_buf *buf = &info->lti_buf;
1481 if (!dt_object_exists(next))
1484 if (lod_striping_loaded(lo))
1487 mutex_lock(&lo->ldo_layout_mutex);
1488 if (lod_striping_loaded(lo))
1489 GOTO(unlock, rc = 0);
1491 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1492 rc = lod_get_lov_ea(env, lo);
1497 * there is LOV EA (striping information) in this object
1498 * let's parse it and create in-core objects for the stripes
1500 buf->lb_buf = info->lti_ea_store;
1501 buf->lb_len = info->lti_ea_store_size;
1502 rc = lod_parse_striping(env, lo, buf, 0);
1504 lo->ldo_comp_cached = 1;
1505 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1506 rc = lod_get_lmv_ea(env, lo);
1507 if (rc > sizeof(struct lmv_foreign_md)) {
1508 struct lmv_foreign_md *lfm = info->lti_ea_store;
1510 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
1511 lo->ldo_foreign_lmv = info->lti_ea_store;
1512 lo->ldo_foreign_lmv_size =
1513 info->lti_ea_store_size;
1514 info->lti_ea_store = NULL;
1515 info->lti_ea_store_size = 0;
1517 lo->ldo_dir_stripe_loaded = 1;
1518 lo->ldo_is_foreign = 1;
1519 GOTO(unlock, rc = 0);
1523 if (rc < (int)sizeof(struct lmv_mds_md_v1)) {
1524 /* Let's set stripe_loaded to avoid further
1525 * stripe loading especially for non-stripe directory,
1526 * which can hurt performance. (See LU-9840)
1529 lo->ldo_dir_stripe_loaded = 1;
1530 GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1532 buf->lb_buf = info->lti_ea_store;
1533 buf->lb_len = info->lti_ea_store_size;
1534 if (rc == sizeof(struct lmv_mds_md_v1)) {
1535 rc = lod_load_lmv_shards(env, lo, buf, true);
1536 if (buf->lb_buf != info->lti_ea_store) {
1537 OBD_FREE_LARGE(info->lti_ea_store,
1538 info->lti_ea_store_size);
1539 info->lti_ea_store = buf->lb_buf;
1540 info->lti_ea_store_size = buf->lb_len;
1548 * there is LMV EA (striping information) in this object
1549 * let's parse it and create in-core objects for the stripes
1551 rc = lod_parse_dir_striping(env, lo, buf);
1553 lo->ldo_dir_stripe_loaded = 1;
1557 mutex_unlock(&lo->ldo_layout_mutex);
1562 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1563 const struct lu_buf *buf, enum layout_verify_flags lvf)
1569 mutex_lock(&lo->ldo_layout_mutex);
1570 rc = lod_parse_striping(env, lo, buf, lvf);
1571 mutex_unlock(&lo->ldo_layout_mutex);
1577 * Verify lov_user_md_v1/v3 striping.
1579 * Check the validity of all fields including the magic, stripe size,
1580 * stripe count, stripe offset and that the pool is present. Also check
1581 * that each target index points to an existing target. The additional
1582 * \a is_from_disk turns additional checks. In some cases zero fields
1583 * are allowed (like pattern=0).
1585 * \param[in] d LOD device
1586 * \param[in] buf buffer with LOV EA to verify
1587 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1588 * 1 - from disk, do not allow
1590 * \retval 0 if the striping is valid
1591 * \retval -EINVAL if striping is invalid
1593 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1596 struct lov_user_md_v1 *lum;
1597 struct lov_user_md_v3 *lum3;
1598 struct pool_desc *pool = NULL;
1602 __u16 stripe_offset;
1609 if (buf->lb_len < sizeof(*lum)) {
1610 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1612 GOTO(out, rc = -EINVAL);
1615 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1616 if (magic != LOV_USER_MAGIC_V1 &&
1617 magic != LOV_USER_MAGIC_V3 &&
1618 magic != LOV_USER_MAGIC_SPECIFIC) {
1619 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1620 le32_to_cpu(lum->lmm_magic));
1621 GOTO(out, rc = -EINVAL);
1624 /* the user uses "0" for default stripe pattern normally. */
1625 if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1626 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1628 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1629 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1630 le32_to_cpu(lum->lmm_pattern));
1631 GOTO(out, rc = -EINVAL);
1634 /* a released lum comes from creating orphan on hsm release,
1635 * doesn't make sense to verify it. */
1636 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1639 /* 64kB is the largest common page size we see (ia64), and matches the
1641 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1642 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1643 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1644 stripe_size, LOV_MIN_STRIPE_SIZE);
1645 GOTO(out, rc = -EINVAL);
1648 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1649 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1650 lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1651 /* if offset is not within valid range [0, osts_size) */
1652 if (stripe_offset >= d->lod_ost_descs.ltd_tgts_size) {
1653 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1654 stripe_offset, d->lod_ost_descs.ltd_tgts_size);
1655 GOTO(out, rc = -EINVAL);
1658 /* if lmm_stripe_offset is *not* in bitmap */
1659 if (!test_bit(stripe_offset, d->lod_ost_bitmap)) {
1660 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1662 GOTO(out, rc = -EINVAL);
1666 if (magic == LOV_USER_MAGIC_V1)
1667 lum_size = offsetof(struct lov_user_md_v1,
1669 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1670 lum_size = offsetof(struct lov_user_md_v3,
1673 GOTO(out, rc = -EINVAL);
1675 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1676 if (buf->lb_len < lum_size) {
1677 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1678 "magic %#x and stripe_count %u\n",
1679 buf->lb_len, lum_size, magic, stripe_count);
1680 GOTO(out, rc = -EINVAL);
1683 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1687 /* In the function below, .hs_keycmp resolves to
1688 * pool_hashkey_keycmp() */
1689 /* coverity[overrun-buffer-val] */
1690 pool = lod_find_pool(d, lum3->lmm_pool_name);
1694 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1695 rc = lod_check_index_in_pool(stripe_offset, pool);
1697 GOTO(out, rc = -EINVAL);
1700 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1701 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1702 stripe_count, pool_tgt_count(pool));
1703 GOTO(out, rc = -EINVAL);
1708 lod_pool_putref(pool);
1714 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1716 LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1717 LOV_USER_MAGIC_COMP_V1 ||
1718 (le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1719 LOV_USER_MAGIC_SEL, "Wrong magic %x\n",
1720 le32_to_cpu(comp->lcm_magic));
1721 LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1722 "bad index %d, max = %d\n",
1723 i, le16_to_cpu(comp->lcm_entry_count));
1725 return &comp->lcm_entries[i];
1728 #define for_each_comp_entry_v1(comp, entry) \
1729 for (entry = comp_entry_v1(comp, 0); \
1730 entry <= comp_entry_v1(comp, \
1731 le16_to_cpu(comp->lcm_entry_count) - 1); \
1734 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
1735 struct lov_comp_md_entry_v1 *dom_ent)
1737 struct lov_comp_md_entry_v1 *ent;
1739 __u32 dom_off, dom_size, comp_size, off;
1741 unsigned int size, shift;
1743 entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1744 LASSERT(entries > 0);
1745 comp_v1->lcm_entry_count = cpu_to_le16(entries);
1747 comp_size = le32_to_cpu(comp_v1->lcm_size);
1748 dom_off = le32_to_cpu(dom_ent->lcme_offset);
1749 dom_size = le32_to_cpu(dom_ent->lcme_size);
1751 /* all entries offsets are shifted by entry size at least */
1752 shift = sizeof(*dom_ent);
1753 for_each_comp_entry_v1(comp_v1, ent) {
1754 off = le32_to_cpu(ent->lcme_offset);
1755 if (off == dom_off) {
1756 /* Entry deletion creates two holes in layout data:
1757 * - hole in entries array
1758 * - hole in layout data at dom_off with dom_size
1760 * First memmove is one entry shift from next entry
1761 * start with size up to dom_off in blob
1764 src = (void *)(ent + 1);
1765 size = (unsigned long)((void *)comp_v1 + dom_off - src);
1766 memmove(dst, src, size);
1767 /* take 'off' from just moved entry */
1768 off = le32_to_cpu(ent->lcme_offset);
1769 /* second memmove is blob tail after 'off' up to
1772 dst = (void *)comp_v1 + dom_off - sizeof(*ent);
1773 src = (void *)comp_v1 + off;
1774 size = (unsigned long)(comp_size - off);
1775 memmove(dst, src, size);
1776 /* all entries offsets after DoM entry are shifted by
1777 * dom_size additionally
1781 ent->lcme_offset = cpu_to_le32(off - shift);
1783 comp_v1->lcm_size = cpu_to_le32(comp_size - shift);
1785 /* notify a caller to re-check entry */
1789 void lod_dom_stripesize_recalc(struct lod_device *d)
1791 __u64 threshold_mb = d->lod_dom_threshold_free_mb;
1792 __u32 max_size = d->lod_dom_stripesize_max_kb;
1793 __u32 def_size = d->lod_dom_stripesize_cur_kb;
1795 /* use maximum allowed value if free space is above threshold */
1796 if (d->lod_lsfs_free_mb >= threshold_mb) {
1797 def_size = max_size;
1798 } else if (!d->lod_lsfs_free_mb || max_size <= LOD_DOM_MIN_SIZE_KB) {
1801 /* recalc threshold like it would be with def_size as max */
1802 threshold_mb = mult_frac(threshold_mb, def_size, max_size);
1803 if (d->lod_lsfs_free_mb < threshold_mb)
1804 def_size = rounddown(def_size / 2, LOD_DOM_MIN_SIZE_KB);
1805 else if (d->lod_lsfs_free_mb > threshold_mb * 2)
1806 def_size = max_t(unsigned int, def_size * 2,
1807 LOD_DOM_MIN_SIZE_KB);
1810 if (d->lod_dom_stripesize_cur_kb != def_size) {
1811 CDEBUG(D_LAYOUT, "Change default DOM stripe size %d->%d\n",
1812 d->lod_dom_stripesize_cur_kb, def_size);
1813 d->lod_dom_stripesize_cur_kb = def_size;
1817 static __u32 lod_dom_stripesize_limit(const struct lu_env *env,
1818 struct lod_device *d)
1822 /* set bfree as fraction of total space */
1823 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_SPOOF)) {
1824 spin_lock(&d->lod_lsfs_lock);
1825 d->lod_lsfs_free_mb = mult_frac(d->lod_lsfs_total_mb,
1826 min_t(int, cfs_fail_val, 100), 100);
1827 GOTO(recalc, rc = 0);
1830 if (d->lod_lsfs_age < ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE) {
1831 struct obd_statfs sfs;
1833 spin_lock(&d->lod_lsfs_lock);
1834 if (d->lod_lsfs_age > ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE)
1835 GOTO(unlock, rc = 0);
1837 d->lod_lsfs_age = ktime_get_seconds();
1838 spin_unlock(&d->lod_lsfs_lock);
1839 rc = dt_statfs(env, d->lod_child, &sfs);
1842 "%s: failed to get OSD statfs: rc = %d\n",
1843 lod2obd(d)->obd_name, rc);
1846 /* udpate local OSD cached statfs data */
1847 spin_lock(&d->lod_lsfs_lock);
1848 d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
1849 d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
1851 lod_dom_stripesize_recalc(d);
1853 spin_unlock(&d->lod_lsfs_lock);
1856 return d->lod_dom_stripesize_cur_kb << 10;
1859 int lod_dom_stripesize_choose(const struct lu_env *env, struct lod_device *d,
1860 struct lov_comp_md_v1 *comp_v1,
1861 struct lov_comp_md_entry_v1 *dom_ent,
1864 struct lov_comp_md_entry_v1 *ent;
1865 struct lu_extent *dom_ext, *ext;
1866 struct lov_user_md_v1 *lum;
1867 __u32 max_stripe_size;
1870 bool dom_next_entry = false;
1872 dom_ext = &dom_ent->lcme_extent;
1873 dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1874 max_stripe_size = lod_dom_stripesize_limit(env, d);
1876 /* Check stripe size againts current per-MDT limit */
1877 if (stripe_size <= max_stripe_size)
1880 lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1881 CDEBUG(D_LAYOUT, "overwrite DoM component size %u with MDT limit %u\n",
1882 stripe_size, max_stripe_size);
1883 lum->lmm_stripe_size = cpu_to_le32(max_stripe_size);
1885 /* In common case the DoM stripe is first entry in a mirror and
1886 * can be deleted only if it is not single entry in layout or
1887 * mirror, otherwise error should be returned.
1889 for_each_comp_entry_v1(comp_v1, ent) {
1893 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1897 ext = &ent->lcme_extent;
1898 if (ext->e_start != dom_ext->e_end)
1901 /* Found next component after the DoM one with the same
1902 * mirror_id and adjust its start with DoM component end.
1904 * NOTE: we are considering here that there can be only one
1905 * DoM component in a file, all replicas are located on OSTs
1906 * always and don't need adjustment since use own layouts.
1908 ext->e_start = cpu_to_le64(max_stripe_size);
1909 dom_next_entry = true;
1913 if (max_stripe_size == 0) {
1914 /* DoM component size is zero due to server setting, remove
1915 * it from the layout but only if next component exists in
1916 * the same mirror. That must be checked prior calling the
1917 * lod_erase_dom_stripe().
1919 if (!dom_next_entry)
1922 rc = lod_erase_dom_stripe(comp_v1, dom_ent);
1924 /* Update DoM extent end finally */
1925 dom_ext->e_end = cpu_to_le64(max_stripe_size);
1932 * Verify LOV striping.
1934 * \param[in] d LOD device
1935 * \param[in] buf buffer with LOV EA to verify
1936 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1937 * 1 - from disk, do not allow
1938 * \param[in] start extent start for composite layout
1940 * \retval 0 if the striping is valid
1941 * \retval -EINVAL if striping is invalid
1943 int lod_verify_striping(const struct lu_env *env, struct lod_device *d,
1944 struct lod_object *lo, const struct lu_buf *buf,
1947 struct lov_user_md_v1 *lum;
1948 struct lov_comp_md_v1 *comp_v1;
1949 struct lov_comp_md_entry_v1 *ent;
1950 struct lu_extent *ext;
1953 __u32 stripe_size = 0;
1954 __u16 prev_mid = -1, mirror_id = -1;
1960 if (buf->lb_len < sizeof(lum->lmm_magic)) {
1961 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
1967 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1968 /* treat foreign LOV EA/object case first
1969 * XXX is it expected to try setting again a foreign?
1970 * XXX should we care about different current vs new layouts ?
1972 if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
1973 struct lov_foreign_md *lfm = buf->lb_buf;
1975 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
1977 "buf len %zu < min lov_foreign_md size (%zu)\n",
1978 buf->lb_len, offsetof(typeof(*lfm),
1983 if (foreign_size_le(lfm) > buf->lb_len) {
1985 "buf len %zu < this lov_foreign_md size (%zu)\n",
1986 buf->lb_len, foreign_size_le(lfm));
1989 /* Don't do anything with foreign layouts */
1993 /* normal LOV/layout cases */
1995 if (buf->lb_len < sizeof(*lum)) {
1996 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
2002 case LOV_USER_MAGIC_FOREIGN:
2004 case LOV_USER_MAGIC_V1:
2005 case LOV_USER_MAGIC_V3:
2006 case LOV_USER_MAGIC_SPECIFIC:
2007 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
2008 case LOV_USER_MAGIC_COMP_V1:
2009 case LOV_USER_MAGIC_SEL:
2012 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
2013 le32_to_cpu(lum->lmm_magic));
2017 /* magic == LOV_USER_MAGIC_COMP_V1 */
2018 comp_v1 = buf->lb_buf;
2019 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
2020 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
2021 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
2027 if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
2028 CDEBUG(D_LAYOUT, "entry count is zero\n");
2032 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
2033 lo->ldo_comp_cnt > 0) {
2034 /* could be called from lustre.lov.add */
2035 __u32 cnt = lo->ldo_comp_cnt;
2037 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
2038 prev_end = ext->e_end;
2043 for_each_comp_entry_v1(comp_v1, ent) {
2044 ext = &ent->lcme_extent;
2046 if (le64_to_cpu(ext->e_start) > le64_to_cpu(ext->e_end) ||
2047 le64_to_cpu(ext->e_start) & (LOV_MIN_STRIPE_SIZE - 1) ||
2048 (le64_to_cpu(ext->e_end) != LUSTRE_EOF &&
2049 le64_to_cpu(ext->e_end) & (LOV_MIN_STRIPE_SIZE - 1))) {
2050 CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
2051 le64_to_cpu(ext->e_start),
2052 le64_to_cpu(ext->e_end));
2057 /* lcme_id contains valid value */
2058 if (le32_to_cpu(ent->lcme_id) == 0 ||
2059 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
2060 CDEBUG(D_LAYOUT, "invalid id %u\n",
2061 le32_to_cpu(ent->lcme_id));
2065 if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
2066 mirror_id = mirror_id_of(
2067 le32_to_cpu(ent->lcme_id));
2069 /* first component must start with 0 */
2070 if (mirror_id != prev_mid &&
2071 le64_to_cpu(ext->e_start) != 0) {
2073 "invalid start:%llu, expect:0\n",
2074 le64_to_cpu(ext->e_start));
2078 prev_mid = mirror_id;
2082 if (le64_to_cpu(ext->e_start) == 0) {
2087 /* the next must be adjacent with the previous one */
2088 if (le64_to_cpu(ext->e_start) != prev_end) {
2090 "invalid start actual:%llu, expect:%llu\n",
2091 le64_to_cpu(ext->e_start), prev_end);
2095 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
2096 tmp.lb_len = le32_to_cpu(ent->lcme_size);
2098 /* Check DoM entry is always the first one */
2100 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
2102 /* DoM component must be the first in a mirror */
2103 if (le64_to_cpu(ext->e_start) > 0) {
2104 CDEBUG(D_LAYOUT, "invalid DoM component "
2105 "with %llu extent start\n",
2106 le64_to_cpu(ext->e_start));
2109 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2110 /* There is just one stripe on MDT and it must
2111 * cover whole component size. */
2112 if (stripe_size != le64_to_cpu(ext->e_end)) {
2113 CDEBUG(D_LAYOUT, "invalid DoM layout "
2114 "stripe size %u != %llu "
2115 "(component size)\n",
2116 stripe_size, prev_end);
2119 /* Check and adjust stripe size by per-MDT limit */
2120 rc = lod_dom_stripesize_choose(env, d, comp_v1, ent,
2122 /* DoM entry was removed, re-check layout from start */
2123 if (rc == -ERESTART)
2128 if (le16_to_cpu(lum->lmm_stripe_count) == 1)
2129 lum->lmm_stripe_count = 0;
2130 /* Any stripe count is forbidden on DoM component */
2131 if (lum->lmm_stripe_count > 0) {
2133 "invalid DoM layout stripe count %u, must be 0\n",
2134 le16_to_cpu(lum->lmm_stripe_count));
2138 /* Any pool is forbidden on DoM component */
2139 if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
2140 struct lov_user_md_v3 *v3 = (void *)lum;
2142 if (v3->lmm_pool_name[0] != '\0') {
2144 "DoM component cannot have pool assigned\n");
2150 prev_end = le64_to_cpu(ext->e_end);
2152 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
2156 if (prev_end == LUSTRE_EOF || ext->e_start == prev_end)
2159 /* extent end must be aligned with the stripe_size */
2160 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2161 if (stripe_size && prev_end % stripe_size) {
2162 CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
2163 "stripe_sz: %u, [%llu, %llu)\n",
2164 stripe_size, ext->e_start, prev_end);
2169 /* make sure that the mirror_count is telling the truth */
2170 if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
2177 * set the default stripe size, if unset.
2179 * \param[in,out] val number of bytes per OST stripe
2181 * The minimum stripe size is 64KB to ensure that a single stripe is an
2182 * even multiple of a client PAGE_SIZE (IA64, PPC, etc). Otherwise, it
2183 * is difficult to split dirty pages across OSCs during writes.
2185 void lod_fix_desc_stripe_size(__u64 *val)
2187 if (*val < LOV_MIN_STRIPE_SIZE) {
2189 LCONSOLE_INFO("Increasing default stripe size to "
2190 "minimum value %u\n",
2191 LOV_DESC_STRIPE_SIZE_DEFAULT);
2192 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
2193 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
2194 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
2195 LCONSOLE_WARN("Changing default stripe size to %llu (a "
2196 "multiple of %u)\n",
2197 *val, LOV_MIN_STRIPE_SIZE);
2202 * set the filesystem default number of stripes, if unset.
2204 * \param[in,out] val number of stripes
2206 * A value of "0" means "use the system-wide default stripe count", which
2207 * has either been inherited by now, or falls back to 1 stripe per file.
2208 * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
2209 * and is a valid value, so is left unchanged here.
2211 void lod_fix_desc_stripe_count(__u32 *val)
2218 * set the filesystem default layout pattern
2220 * \param[in,out] val LOV_PATTERN_* layout
2222 * A value of "0" means "use the system-wide default layout type", which
2223 * has either been inherited by now, or falls back to plain RAID0 striping.
2225 void lod_fix_desc_pattern(__u32 *val)
2227 /* from lov_setstripe */
2228 if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
2229 LCONSOLE_WARN("lod: Unknown stripe pattern: %#x\n", *val);
2234 void lod_fix_lmv_desc_pattern(__u32 *val)
2236 if ((*val) && !lmv_is_known_hash_type(*val)) {
2237 LCONSOLE_WARN("lod: Unknown md stripe pattern: %#x\n", *val);
2242 void lod_fix_desc_qos_maxage(__u32 *val)
2244 /* fix qos_maxage */
2246 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2250 * Used to fix insane default striping.
2252 * \param[in] desc striping description
2254 void lod_fix_desc(struct lov_desc *desc)
2256 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2257 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2258 lod_fix_desc_pattern(&desc->ld_pattern);
2259 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2262 static void lod_fix_lmv_desc(struct lmv_desc *desc)
2264 desc->ld_active_tgt_count = 0;
2265 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2266 lod_fix_lmv_desc_pattern(&desc->ld_pattern);
2267 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2271 * Initialize the structures used to store pools and default striping.
2273 * \param[in] lod LOD device
2274 * \param[in] lcfg configuration structure storing default striping.
2276 * \retval 0 if initialization succeeds
2277 * \retval negative error number on failure
2279 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2281 struct obd_device *obd;
2282 struct lov_desc *desc;
2286 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2287 LASSERT(obd != NULL);
2288 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2290 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2291 CERROR("LOD setup requires a descriptor\n");
2295 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2297 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2298 CERROR("descriptor size wrong: %d > %d\n",
2299 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2303 if (desc->ld_magic != LOV_DESC_MAGIC) {
2304 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2305 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2306 obd->obd_name, desc);
2307 lustre_swab_lov_desc(desc);
2309 CERROR("%s: Bad lov desc magic: %#x\n",
2310 obd->obd_name, desc->ld_magic);
2317 desc->ld_active_tgt_count = 0;
2318 lod->lod_ost_descs.ltd_lov_desc = *desc;
2320 /* NB: config doesn't contain lmv_desc, alter it via sysfs. */
2321 lod_fix_lmv_desc(&lod->lod_mdt_descs.ltd_lmv_desc);
2323 lod->lod_sp_me = LUSTRE_SP_CLI;
2325 /* Set up OST pool environment */
2326 lod->lod_pool_count = 0;
2327 rc = lod_pool_hash_init(&lod->lod_pools_hash_body);
2331 INIT_LIST_HEAD(&lod->lod_pool_list);
2332 lod->lod_pool_count = 0;
2333 rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_tgt_pool, 0);
2337 rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool, 0);
2339 GOTO(out_mdt_pool, rc);
2341 rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_tgt_pool, 0);
2343 GOTO(out_mdt_rr_pool, rc);
2345 rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
2347 GOTO(out_ost_pool, rc);
2352 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2354 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2356 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
2358 lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2364 * Release the structures describing the pools.
2366 * \param[in] lod LOD device from which we release the structures
2370 int lod_pools_fini(struct lod_device *lod)
2372 struct obd_device *obd = lod2obd(lod);
2373 struct pool_desc *pool, *tmp;
2376 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2377 /* free pool structs */
2378 CDEBUG(D_INFO, "delete pool %p\n", pool);
2379 /* In the function below, .hs_keycmp resolves to
2380 * pool_hashkey_keycmp() */
2381 /* coverity[overrun-buffer-val] */
2382 lod_pool_del(obd, pool->pool_name);
2385 lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2386 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool);
2387 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2388 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2389 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);