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 |
214 OBD_CONNECT_BULK_MBITS;
215 spin_lock(&imp->imp_lock);
216 imp->imp_server_timeout = 1;
217 spin_unlock(&imp->imp_lock);
218 imp->imp_client->cli_request_portal = OUT_PORTAL;
219 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
221 ltd = &lod->lod_mdt_descs;
224 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
227 CERROR("%s: cannot connect to next dev %s (%d)\n",
228 obd->obd_name, osp, rc);
229 GOTO(out_cleanup, rc);
233 /* Allocate ost descriptor and fill it */
234 OBD_ALLOC_PTR(tgt_desc);
236 GOTO(out_cleanup, rc = -ENOMEM);
238 tgt_desc->ltd_tgt = dt_dev;
239 tgt_desc->ltd_exp = exp;
240 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
241 tgt_desc->ltd_gen = gen;
242 tgt_desc->ltd_index = index;
243 tgt_desc->ltd_active = active;
245 down_write(<d->ltd_rw_sem);
246 mutex_lock(<d->ltd_mutex);
247 rc = ltd_add_tgt(ltd, tgt_desc);
251 rc = lu_qos_add_tgt(<d->ltd_qos, tgt_desc);
253 GOTO(out_del_tgt, rc);
255 rc = lu_tgt_pool_add(<d->ltd_tgt_pool, index,
256 ltd->ltd_lov_desc.ld_tgt_count);
258 CERROR("%s: can't set up pool, failed with %d\n",
260 GOTO(out_del_tgt, rc);
263 mutex_unlock(<d->ltd_mutex);
264 up_write(<d->ltd_rw_sem);
266 if (lod->lod_recovery_completed)
267 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
269 if (!for_ost && lod->lod_initialized) {
270 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
272 CERROR("%s: cannot start llog on %s:rc = %d\n",
273 lod2obd(lod)->obd_name, osp, rc);
278 rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
280 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
281 "index = %u, rc = %d\n", osp, type, index, rc);
282 GOTO(out_fini_llog, rc);
286 lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
287 &tgt_desc->ltd_recovery_task);
289 down_write(<d->ltd_rw_sem);
290 mutex_lock(<d->ltd_mutex);
291 lu_tgt_pool_remove(<d->ltd_tgt_pool, index);
293 ltd_del_tgt(ltd, tgt_desc);
295 mutex_unlock(<d->ltd_mutex);
296 up_write(<d->ltd_rw_sem);
297 OBD_FREE_PTR(tgt_desc);
299 /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
300 * objects from the MDT stack. See LU-7184. */
301 lcfg = &lod_env_info(env)->lti_lustre_cfg;
302 memset(lcfg, 0, sizeof(*lcfg));
303 lcfg->lcfg_version = LUSTRE_CFG_VERSION;
304 lcfg->lcfg_command = LCFG_CLEANUP;
305 lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
314 * Schedule target removal from the target table.
316 * Mark the device as dead. The device is not removed here because it may
317 * still be in use. The device will be removed in lod_putref() when the
318 * last reference is released.
320 * \param[in] env execution environment for this thread
321 * \param[in] lod LOD device the target table belongs to
322 * \param[in] ltd target table
323 * \param[in] tgt target
325 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
326 struct lod_tgt_descs *ltd, struct lu_tgt_desc *tgt)
328 lfsck_del_target(env, lod->lod_child, tgt->ltd_tgt, tgt->ltd_index,
331 if (!tgt->ltd_reap) {
333 ltd->ltd_death_row++;
338 * Schedule removal of all the targets from the given target table.
340 * See more details in the description for __lod_del_device()
342 * \param[in] env execution environment for this thread
343 * \param[in] lod LOD device the target table belongs to
344 * \param[in] ltd target table
348 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
349 struct lod_tgt_descs *ltd)
351 struct lu_tgt_desc *tgt;
353 if (ltd->ltd_tgts_size <= 0)
357 mutex_lock(<d->ltd_mutex);
358 ltd_foreach_tgt(ltd, tgt)
359 __lod_del_device(env, lod, ltd, tgt);
360 mutex_unlock(<d->ltd_mutex);
361 lod_putref(lod, ltd);
363 lu_tgt_descs_fini(ltd);
369 * Remove device by name.
371 * Remove a device identified by \a osp from the target table. Given
372 * the device can be in use, the real deletion happens in lod_putref().
374 * \param[in] env execution environment for this thread
375 * \param[in] lod LOD device to be connected to the new OSP
376 * \param[in] ltd target table
377 * \param[in] osp name of OSP device to be removed
378 * \param[in] idx index of the target
379 * \param[in] gen generation number, not used currently
381 * \retval 0 if the device was scheduled for removal
382 * \retval -EINVAL if no device was found
384 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
385 struct lod_tgt_descs *ltd, char *osp, unsigned int idx,
388 struct obd_device *obd;
389 struct lu_tgt_desc *tgt;
390 struct obd_uuid uuid;
395 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
397 obd_str2uuid(&uuid, osp);
399 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
400 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
402 CERROR("can't find %s device\n", osp);
407 CERROR("%s: request to remove OBD %s with invalid generation %d"
408 "\n", obd->obd_name, osp, gen);
412 obd_str2uuid(&uuid, osp);
415 mutex_lock(<d->ltd_mutex);
416 tgt = LTD_TGT(ltd, idx);
417 /* check that the index is allocated in the bitmap */
418 if (!test_bit(idx, ltd->ltd_tgt_bitmap) || !tgt) {
419 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
420 GOTO(out, rc = -EINVAL);
423 /* check that the UUID matches */
424 if (!obd_uuid_equals(&uuid, &tgt->ltd_uuid)) {
425 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
426 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), idx, osp);
427 GOTO(out, rc = -EINVAL);
430 __lod_del_device(env, lod, ltd, tgt);
433 mutex_unlock(<d->ltd_mutex);
434 lod_putref(lod, ltd);
439 * Resize per-thread storage to hold specified size.
441 * A helper function to resize per-thread temporary storage. This storage
442 * is used to process LOV/LVM EAs and may be quite large. We do not want to
443 * allocate/release it every time, so instead we put it into the env and
444 * reallocate on demand. The memory is released when the correspondent thread
447 * \param[in] info LOD-specific storage in the environment
448 * \param[in] size new size to grow the buffer to
450 * \retval 0 on success, -ENOMEM if reallocation failed
452 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
454 __u32 round = size_roundup_power2(size);
456 if (info->lti_ea_store) {
457 LASSERT(info->lti_ea_store_size);
458 LASSERT(info->lti_ea_store_size < round);
459 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
460 info->lti_ea_store_size, round);
461 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
462 info->lti_ea_store = NULL;
463 info->lti_ea_store_size = 0;
466 OBD_ALLOC_LARGE(info->lti_ea_store, round);
467 if (info->lti_ea_store == NULL)
469 info->lti_ea_store_size = round;
474 static void lod_free_comp_buffer(struct lod_layout_component *entries,
475 __u16 count, __u32 bufsize)
477 struct lod_layout_component *entry;
480 for (i = 0; i < count; i++) {
482 if (entry->llc_pool != NULL)
483 lod_set_pool(&entry->llc_pool, NULL);
484 if (entry->llc_ostlist.op_array)
485 OBD_FREE(entry->llc_ostlist.op_array,
486 entry->llc_ostlist.op_size);
487 LASSERT(entry->llc_stripe == NULL);
488 LASSERT(entry->llc_stripes_allocated == 0);
492 OBD_FREE_LARGE(entries, bufsize);
495 void lod_free_def_comp_entries(struct lod_default_striping *lds)
497 lod_free_comp_buffer(lds->lds_def_comp_entries,
498 lds->lds_def_comp_size_cnt,
500 sizeof(*lds->lds_def_comp_entries) *
501 lds->lds_def_comp_size_cnt));
502 lds->lds_def_comp_entries = NULL;
503 lds->lds_def_comp_cnt = 0;
504 lds->lds_def_striping_is_composite = 0;
505 lds->lds_def_comp_size_cnt = 0;
509 * Resize per-thread storage to hold default striping component entries
511 * A helper function to resize per-thread temporary storage. This storage
512 * is used to hold default LOV/LVM EAs and may be quite large. We do not want
513 * to allocate/release it every time, so instead we put it into the env and
514 * reallocate it on demand. The memory is released when the correspondent
515 * thread is finished.
517 * \param[in,out] lds default striping
518 * \param[in] count new component count to grow the buffer to
520 * \retval 0 on success, -ENOMEM if reallocation failed
522 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
524 struct lod_layout_component *entries;
525 __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
527 __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
528 lds->lds_def_comp_size_cnt);
533 OBD_ALLOC_LARGE(entries, new);
537 if (lds->lds_def_comp_entries != NULL) {
538 CDEBUG(D_INFO, "default striping component size %d is not "
539 "enough, need %d\n", old, new);
540 lod_free_def_comp_entries(lds);
543 lds->lds_def_comp_entries = entries;
544 lds->lds_def_comp_size_cnt = count;
549 void lod_free_comp_entries(struct lod_object *lo)
551 if (lo->ldo_mirrors) {
552 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, lo->ldo_mirror_count);
553 lo->ldo_mirrors = NULL;
554 lo->ldo_mirror_count = 0;
556 lod_free_comp_buffer(lo->ldo_comp_entries,
558 sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
559 lo->ldo_comp_entries = NULL;
560 lo->ldo_comp_cnt = 0;
561 lo->ldo_is_composite = 0;
564 int lod_alloc_comp_entries(struct lod_object *lo,
565 int mirror_count, int comp_count)
567 LASSERT(comp_count != 0);
568 LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
570 if (mirror_count > 0) {
571 OBD_ALLOC_PTR_ARRAY(lo->ldo_mirrors, mirror_count);
572 if (!lo->ldo_mirrors)
575 lo->ldo_mirror_count = mirror_count;
578 OBD_ALLOC_LARGE(lo->ldo_comp_entries,
579 sizeof(*lo->ldo_comp_entries) * comp_count);
580 if (lo->ldo_comp_entries == NULL) {
581 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, mirror_count);
582 lo->ldo_mirror_count = 0;
586 lo->ldo_comp_cnt = comp_count;
590 int lod_fill_mirrors(struct lod_object *lo)
592 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
593 struct lod_layout_component *lod_comp;
594 bool found_preferred = false;
596 __u16 mirror_id = 0xffff;
600 LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
602 if (!lo->ldo_is_composite)
605 lod_comp = &lo->ldo_comp_entries[0];
607 for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
608 int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
609 int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
613 /* calculate component preference over all used OSTs */
614 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
615 int idx = lod_comp->llc_ost_indices[j];
616 struct obd_statfs *osfs = &OST_TGT(lod,idx)->ltd_statfs;
618 if (osfs->os_state & OS_STATFS_NONROT)
622 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
623 lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
624 lo->ldo_mirrors[mirror_idx].lme_prefer |= preferred;
625 lo->ldo_mirrors[mirror_idx].lme_preference += pref;
626 lo->ldo_mirrors[mirror_idx].lme_end = i;
630 if (mirror_idx >= 0 && preferred &&
631 !lo->ldo_mirrors[mirror_idx].lme_stale)
632 found_preferred = true;
636 if (mirror_idx >= lo->ldo_mirror_count)
639 mirror_id = mirror_id_of(lod_comp->llc_id);
641 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
642 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
643 lo->ldo_mirrors[mirror_idx].lme_prefer = preferred;
644 lo->ldo_mirrors[mirror_idx].lme_preference = pref;
645 lo->ldo_mirrors[mirror_idx].lme_start = i;
646 lo->ldo_mirrors[mirror_idx].lme_end = i;
648 if (mirror_idx != lo->ldo_mirror_count - 1)
651 if (!found_preferred && mirror_idx > 0) {
655 * if no explicited preferred found, then find a mirror
656 * with higher number of non-rotational OSTs
659 for (i = 0; i <= mirror_idx; i++) {
660 if (lo->ldo_mirrors[i].lme_stale)
662 if (lo->ldo_mirrors[i].lme_preference > pref) {
663 pref = lo->ldo_mirrors[i].lme_preference;
669 lo->ldo_mirrors[best].lme_prefer = 1;
676 * Generate on-disk lov_mds_md structure for each layout component based on
677 * the information in lod_object->ldo_comp_entries[i].
679 * \param[in] env execution environment for this thread
680 * \param[in] lo LOD object
681 * \param[in] comp_idx index of ldo_comp_entries
682 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
683 * \param[in|out] lmm_size buffer size/lmm size
684 * \param[in] is_dir generate lov ea for dir or file? For dir case,
685 * the stripe info is from the default stripe
686 * template, which is collected in lod_ah_init(),
687 * either from parent object or root object; for
688 * file case, it's from the @lo object
690 * \retval 0 if on disk structure is created successfully
691 * \retval negative error number on failure
693 static int lod_gen_component_ea(const struct lu_env *env,
694 struct lod_object *lo, int comp_idx,
695 struct lov_mds_md *lmm, int *lmm_size,
698 struct lod_thread_info *info = lod_env_info(env);
699 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
700 struct lod_device *lod;
701 struct lov_ost_data_v1 *objs;
702 struct lod_layout_component *lod_comp;
711 &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
713 lod_comp = &lo->ldo_comp_entries[comp_idx];
715 magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
716 if (lod_comp->llc_pattern == 0) /* default striping */
717 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
719 lmm->lmm_magic = cpu_to_le32(magic);
720 lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
721 fid_to_lmm_oi(fid, &lmm->lmm_oi);
722 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
723 lmm->lmm_oi.oi.oi_id++;
724 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
726 lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
727 lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
729 * for dir and uninstantiated component, lmm_layout_gen stores
730 * default stripe offset.
732 lmm->lmm_layout_gen =
733 (is_dir || !lod_comp_inited(lod_comp)) ?
734 cpu_to_le16(lod_comp->llc_stripe_offset) :
735 cpu_to_le16(lod_comp->llc_layout_gen);
737 if (magic == LOV_MAGIC_V1) {
738 objs = &lmm->lmm_objects[0];
740 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
741 size_t cplen = strlcpy(v3->lmm_pool_name,
743 sizeof(v3->lmm_pool_name));
744 if (cplen >= sizeof(v3->lmm_pool_name))
746 objs = &v3->lmm_objects[0];
748 stripe_count = lod_comp_entry_stripe_count(lo, comp_idx, is_dir);
749 if (stripe_count == 0 && !is_dir &&
750 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
751 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
754 if (!is_dir && lo->ldo_is_composite)
755 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
757 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
760 /* generate ost_idx of this component stripe */
761 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
762 for (i = 0; i < stripe_count; i++) {
763 struct dt_object *object;
764 __u32 ost_idx = (__u32)-1UL;
765 int type = LU_SEQ_RANGE_OST;
767 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
768 object = lod_comp->llc_stripe[i];
769 /* instantiated component */
770 info->lti_fid = *lu_object_fid(&object->do_lu);
772 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
774 if (cfs_fail_val == 0)
775 cfs_fail_val = info->lti_fid.f_oid;
777 info->lti_fid.f_oid = cfs_fail_val;
780 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
783 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
784 objs[i].l_ost_gen = cpu_to_le32(0);
785 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
788 rc = lod_fld_lookup(env, lod, &info->lti_fid,
791 CERROR("%s: Can not locate "DFID": rc = %d\n",
792 lod2obd(lod)->obd_name,
793 PFID(&info->lti_fid), rc);
796 } else if (lod_comp->llc_ostlist.op_array &&
797 lod_comp->llc_ostlist.op_count) {
798 /* user specified ost list */
799 ost_idx = lod_comp->llc_ostlist.op_array[i];
802 * with un-instantiated or with no specified ost list
803 * component, its l_ost_idx does not matter.
805 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
808 if (lmm_size != NULL)
809 *lmm_size = lov_mds_md_size(stripe_count, magic);
814 * Generate on-disk lov_mds_md structure based on the information in
815 * the lod_object->ldo_comp_entries.
817 * \param[in] env execution environment for this thread
818 * \param[in] lo LOD object
819 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
820 * \param[in|out] lmm_size buffer size/lmm size
821 * \param[in] is_dir generate lov ea for dir or file? For dir case,
822 * the stripe info is from the default stripe
823 * template, which is collected in lod_ah_init(),
824 * either from parent object or root object; for
825 * file case, it's from the @lo object
827 * \retval 0 if on disk structure is created successfully
828 * \retval negative error number on failure
830 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
831 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
833 struct lov_comp_md_entry_v1 *lcme;
834 struct lov_comp_md_v1 *lcm;
835 struct lod_layout_component *comp_entries;
836 __u16 comp_cnt, mirror_cnt;
837 bool is_composite, is_foreign = false;
838 int i, rc = 0, offset;
842 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
843 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
844 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
846 lo->ldo_def_striping->lds_def_striping_is_composite;
848 comp_cnt = lo->ldo_comp_cnt;
849 mirror_cnt = lo->ldo_mirror_count;
850 comp_entries = lo->ldo_comp_entries;
851 is_composite = lo->ldo_is_composite;
852 is_foreign = lo->ldo_is_foreign;
855 LASSERT(lmm_size != NULL);
858 struct lov_foreign_md *lfm;
860 lfm = (struct lov_foreign_md *)lmm;
861 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
862 /* need to store little-endian */
863 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
864 __swab32s(&lfm->lfm_magic);
865 __swab32s(&lfm->lfm_length);
866 __swab32s(&lfm->lfm_type);
867 __swab32s(&lfm->lfm_flags);
869 *lmm_size = lo->ldo_foreign_lov_size;
873 LASSERT(comp_cnt != 0 && comp_entries != NULL);
876 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
880 lcm = (struct lov_comp_md_v1 *)lmm;
881 memset(lcm, 0, sizeof(*lcm));
883 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
884 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
885 lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
886 lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
888 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
889 LASSERT(offset % sizeof(__u64) == 0);
891 for (i = 0; i < comp_cnt; i++) {
892 struct lod_layout_component *lod_comp;
893 struct lov_mds_md *sub_md;
896 lod_comp = &comp_entries[i];
897 lcme = &lcm->lcm_entries[i];
899 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
900 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
902 /* component could be un-inistantiated */
903 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
904 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
905 lcme->lcme_timestamp =
906 cpu_to_le64(lod_comp->llc_timestamp);
907 if (lod_comp->llc_flags & LCME_FL_EXTENSION && !is_dir)
908 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
910 lcme->lcme_extent.e_start =
911 cpu_to_le64(lod_comp->llc_extent.e_start);
912 lcme->lcme_extent.e_end =
913 cpu_to_le64(lod_comp->llc_extent.e_end);
914 lcme->lcme_offset = cpu_to_le32(offset);
916 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
917 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
920 lcme->lcme_size = cpu_to_le32(size);
922 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
923 "offset:%d lmm_size:%d\n", offset, *lmm_size);
925 lcm->lcm_size = cpu_to_le32(offset);
926 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
928 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
939 * Fill lti_ea_store buffer in the environment with a value for the given
940 * EA. The buffer is reallocated if the value doesn't fit.
942 * \param[in,out] env execution environment for this thread
943 * .lti_ea_store buffer is filled with EA's value
944 * \param[in] lo LOD object
945 * \param[in] name name of the EA
947 * \retval > 0 if EA is fetched successfully
948 * \retval 0 if EA is empty
949 * \retval negative error number on failure
951 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
954 struct lod_thread_info *info = lod_env_info(env);
955 struct dt_object *next = dt_object_child(&lo->ldo_obj);
961 if (unlikely(info->lti_ea_store == NULL)) {
962 /* just to enter in allocation block below */
966 info->lti_buf.lb_buf = info->lti_ea_store;
967 info->lti_buf.lb_len = info->lti_ea_store_size;
968 rc = dt_xattr_get(env, next, &info->lti_buf, name);
971 /* if object is not striped or inaccessible */
972 if (rc == -ENODATA || rc == -ENOENT)
976 /* EA doesn't fit, reallocate new buffer */
977 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
978 if (rc == -ENODATA || rc == -ENOENT)
984 rc = lod_ea_store_resize(info, rc);
994 * Verify the target index is present in the current configuration.
996 * \param[in] md LOD device where the target table is stored
997 * \param[in] idx target's index
999 * \retval 0 if the index is present
1000 * \retval -EINVAL if not
1002 int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1004 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1005 !test_bit(idx, md->lod_ost_bitmap))) {
1006 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1007 md->lod_ost_descs.ltd_tgts_size);
1011 if (unlikely(OST_TGT(md, idx) == NULL)) {
1012 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1013 lod2obd(md)->obd_name, idx);
1017 if (unlikely(OST_TGT(md, idx)->ltd_tgt == NULL)) {
1018 CERROR("%s: invalid lod device, for idx: %d\n",
1019 lod2obd(md)->obd_name , idx);
1027 * Instantiate objects for stripes.
1029 * Allocate and initialize LU-objects representing the stripes. The number
1030 * of the stripes (ldo_stripe_count) must be initialized already. The caller
1031 * must ensure nobody else is calling the function on the object at the same
1032 * time. FLDB service must be running to be able to map a FID to the targets
1033 * and find appropriate device representing that target.
1035 * \param[in] env execution environment for this thread
1036 * \param[in,out] lo LOD object
1037 * \param[in] objs an array of IDs to creates the objects from
1038 * \param[in] comp_idx index of ldo_comp_entries
1040 * \retval 0 if the objects are instantiated successfully
1041 * \retval negative error number on failure
1043 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1044 struct lov_ost_data_v1 *objs, int comp_idx)
1046 struct lod_layout_component *lod_comp;
1047 struct lod_thread_info *info = lod_env_info(env);
1048 struct lod_device *md;
1049 struct lu_object *o, *n;
1050 struct lu_device *nd;
1051 struct dt_object **stripe = NULL;
1052 __u32 *ost_indices = NULL;
1058 LASSERT(lo != NULL);
1059 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1061 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1062 lod_comp = &lo->ldo_comp_entries[comp_idx];
1064 LASSERT(lod_comp->llc_stripe == NULL);
1065 LASSERT(lod_comp->llc_stripe_count > 0);
1066 LASSERT(lod_comp->llc_stripe_size > 0);
1068 stripe_len = lod_comp->llc_stripe_count;
1069 OBD_ALLOC_PTR_ARRAY(stripe, stripe_len);
1072 OBD_ALLOC_PTR_ARRAY(ost_indices, stripe_len);
1074 GOTO(out, rc = -ENOMEM);
1076 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1077 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1080 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1081 idx = le32_to_cpu(objs[i].l_ost_idx);
1082 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1085 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1086 PFID(&info->lti_fid));
1087 lod_getref(&md->lod_ost_descs);
1089 rc = validate_lod_and_idx(md, idx);
1090 if (unlikely(rc != 0)) {
1091 lod_putref(md, &md->lod_ost_descs);
1095 nd = &OST_TGT(md, idx)->ltd_tgt->dd_lu_dev;
1096 lod_putref(md, &md->lod_ost_descs);
1098 /* In the function below, .hs_keycmp resolves to
1099 * u_obj_hop_keycmp() */
1100 /* coverity[overrun-buffer-val] */
1101 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1103 GOTO(out, rc = PTR_ERR(o));
1105 n = lu_object_locate(o->lo_header, nd->ld_type);
1108 stripe[i] = container_of(n, struct dt_object, do_lu);
1109 ost_indices[i] = idx;
1114 for (i = 0; i < stripe_len; i++)
1115 if (stripe[i] != NULL)
1116 dt_object_put(env, stripe[i]);
1118 OBD_FREE_PTR_ARRAY(stripe, stripe_len);
1119 lod_comp->llc_stripe_count = 0;
1121 OBD_FREE_PTR_ARRAY(ost_indices, stripe_len);
1123 lod_comp->llc_stripe = stripe;
1124 lod_comp->llc_ost_indices = ost_indices;
1125 lod_comp->llc_stripes_allocated = stripe_len;
1132 * Instantiate objects for striping.
1134 * Parse striping information in \a buf and instantiate the objects
1135 * representing the stripes.
1137 * \param[in] env execution environment for this thread
1138 * \param[in] lo LOD object
1139 * \param[in] buf buffer storing LOV EA to parse
1141 * \retval 0 if parsing and objects creation succeed
1142 * \retval negative error number on failure
1144 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1145 const struct lu_buf *buf)
1147 struct lov_mds_md_v1 *lmm;
1148 struct lov_comp_md_v1 *comp_v1 = NULL;
1149 struct lov_foreign_md *foreign = NULL;
1150 struct lov_ost_data_v1 *objs;
1151 __u32 magic, pattern;
1152 __u16 mirror_cnt = 0;
1158 LASSERT(buf->lb_buf);
1159 LASSERT(buf->lb_len);
1160 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1162 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1163 magic = le32_to_cpu(lmm->lmm_magic);
1165 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1166 magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN &&
1167 magic != LOV_MAGIC_SEL)
1168 GOTO(out, rc = -EINVAL);
1170 lod_striping_free_nolock(env, lo);
1172 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1173 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1174 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1176 GOTO(out, rc = -EINVAL);
1177 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1178 lo->ldo_is_composite = 1;
1179 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1181 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1184 lo->ldo_flr_state = LCM_FL_NONE;
1185 } else if (magic == LOV_MAGIC_FOREIGN) {
1188 foreign = (struct lov_foreign_md *)buf->lb_buf;
1189 length = offsetof(typeof(*foreign), lfm_value);
1190 if (buf->lb_len < length ||
1191 buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1193 "buf len %zu too small for lov_foreign_md\n",
1195 GOTO(out, rc = -EINVAL);
1198 /* just cache foreign LOV EA raw */
1199 rc = lod_alloc_foreign_lov(lo, length);
1202 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1206 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1207 lo->ldo_is_composite = 0;
1210 rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1214 for (i = 0; i < comp_cnt; i++) {
1215 struct lod_layout_component *lod_comp;
1216 struct lu_extent *ext;
1219 lod_comp = &lo->ldo_comp_entries[i];
1220 if (lo->ldo_is_composite) {
1221 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1222 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1224 ext = &comp_v1->lcm_entries[i].lcme_extent;
1225 lod_comp->llc_extent.e_start =
1226 le64_to_cpu(ext->e_start);
1227 if (lod_comp->llc_extent.e_start &
1228 (LOV_MIN_STRIPE_SIZE - 1)) {
1230 "extent start %llu is not a multiple of min size %u\n",
1231 lod_comp->llc_extent.e_start,
1232 LOV_MIN_STRIPE_SIZE);
1233 GOTO(out, rc = -EINVAL);
1236 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1237 if (lod_comp->llc_extent.e_end != LUSTRE_EOF &&
1238 lod_comp->llc_extent.e_end &
1239 (LOV_MIN_STRIPE_SIZE - 1)) {
1241 "extent end %llu is not a multiple of min size %u\n",
1242 lod_comp->llc_extent.e_end,
1243 LOV_MIN_STRIPE_SIZE);
1244 GOTO(out, rc = -EINVAL);
1247 lod_comp->llc_flags =
1248 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1249 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1250 lod_comp->llc_timestamp = le64_to_cpu(
1251 comp_v1->lcm_entries[i].lcme_timestamp);
1253 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1254 if (lod_comp->llc_id == LCME_ID_INVAL)
1255 GOTO(out, rc = -EINVAL);
1257 if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
1258 comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
1259 struct lod_device *d =
1260 lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1262 CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
1263 lod2obd(d)->obd_name,
1264 lod_comp->llc_flags, lod_comp->llc_id, i,
1265 PFID(lod_object_fid(lo)),
1266 le32_to_cpu(comp_v1->lcm_magic));
1269 lod_comp_set_init(lod_comp);
1272 pattern = le32_to_cpu(lmm->lmm_pattern);
1273 if (!lov_pattern_supported(lov_pattern(pattern)))
1274 GOTO(out, rc = -EINVAL);
1276 lod_comp->llc_pattern = pattern;
1277 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1278 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1279 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1281 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1282 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1284 lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1285 objs = &v3->lmm_objects[0];
1287 lod_set_pool(&lod_comp->llc_pool, NULL);
1288 objs = &lmm->lmm_objects[0];
1292 * If uninstantiated template component has valid l_ost_idx,
1293 * then user has specified ost list for this component.
1295 if (!lod_comp_inited(lod_comp)) {
1298 if (objs[0].l_ost_idx != (__u32)-1UL) {
1301 stripe_count = lod_comp_entry_stripe_count(
1303 if (stripe_count == 0 &&
1304 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1305 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1306 GOTO(out, rc = -E2BIG);
1308 * load the user specified ost list, when this
1309 * component is instantiated later, it will be
1310 * used in lod_alloc_ost_list().
1312 lod_comp->llc_ostlist.op_count = stripe_count;
1313 lod_comp->llc_ostlist.op_size =
1314 stripe_count * sizeof(__u32);
1315 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1316 lod_comp->llc_ostlist.op_size);
1317 if (!lod_comp->llc_ostlist.op_array)
1318 GOTO(out, rc = -ENOMEM);
1320 for (j = 0; j < stripe_count; j++)
1321 lod_comp->llc_ostlist.op_array[j] =
1322 le32_to_cpu(objs[j].l_ost_idx);
1325 * this component OST objects starts from the
1326 * first ost_idx, lod_alloc_ost_list() will
1329 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1332 * for uninstantiated component,
1333 * lmm_layout_gen stores default stripe offset.
1335 lod_comp->llc_stripe_offset =
1336 lmm->lmm_layout_gen;
1340 /* skip un-instantiated component object initialization */
1341 if (!lod_comp_inited(lod_comp))
1344 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1345 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1346 rc = lod_initialize_objects(env, lo, objs, i);
1352 rc = lod_fill_mirrors(lo);
1358 lod_striping_free_nolock(env, lo);
1363 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1364 * is already cached.
1366 * \param[in] lo LOD object
1368 * \retval True if the striping is cached, otherwise
1371 static bool lod_striping_loaded(struct lod_object *lo)
1373 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1374 lo->ldo_comp_cached)
1377 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1378 if (lo->ldo_dir_stripe_loaded)
1381 /* Never load LMV stripe for slaves of striped dir */
1382 if (lo->ldo_dir_slave_stripe)
1390 * A generic function to initialize the stripe objects.
1392 * A protected version of lod_striping_load_locked() - load the striping
1393 * information from storage, parse that and instantiate LU objects to
1394 * represent the stripes. The LOD object \a lo supplies a pointer to the
1395 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1396 * return an array of references to the newly instantiated objects.
1398 * \param[in] env execution environment for this thread
1399 * \param[in,out] lo LOD object, where striping is stored and
1400 * which gets an array of references
1402 * \retval 0 if parsing and object creation succeed
1403 * \retval negative error number on failure
1405 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1407 struct lod_thread_info *info = lod_env_info(env);
1408 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1409 struct lu_buf *buf = &info->lti_buf;
1414 if (!dt_object_exists(next))
1417 if (lod_striping_loaded(lo))
1420 mutex_lock(&lo->ldo_layout_mutex);
1421 if (lod_striping_loaded(lo))
1422 GOTO(unlock, rc = 0);
1424 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1425 rc = lod_get_lov_ea(env, lo);
1430 * there is LOV EA (striping information) in this object
1431 * let's parse it and create in-core objects for the stripes
1433 buf->lb_buf = info->lti_ea_store;
1434 buf->lb_len = info->lti_ea_store_size;
1435 rc = lod_parse_striping(env, lo, buf);
1437 lo->ldo_comp_cached = 1;
1438 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1439 rc = lod_get_lmv_ea(env, lo);
1440 if (rc > sizeof(struct lmv_foreign_md)) {
1441 struct lmv_foreign_md *lfm = info->lti_ea_store;
1443 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
1444 lo->ldo_foreign_lmv = info->lti_ea_store;
1445 lo->ldo_foreign_lmv_size =
1446 info->lti_ea_store_size;
1447 info->lti_ea_store = NULL;
1448 info->lti_ea_store_size = 0;
1450 lo->ldo_dir_stripe_loaded = 1;
1451 lo->ldo_dir_is_foreign = 1;
1452 GOTO(unlock, rc = 0);
1456 if (rc < (int)sizeof(struct lmv_mds_md_v1)) {
1457 /* Let's set stripe_loaded to avoid further
1458 * stripe loading especially for non-stripe directory,
1459 * which can hurt performance. (See LU-9840)
1462 lo->ldo_dir_stripe_loaded = 1;
1463 GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1465 buf->lb_buf = info->lti_ea_store;
1466 buf->lb_len = info->lti_ea_store_size;
1467 if (rc == sizeof(struct lmv_mds_md_v1)) {
1468 rc = lod_load_lmv_shards(env, lo, buf, true);
1469 if (buf->lb_buf != info->lti_ea_store) {
1470 OBD_FREE_LARGE(info->lti_ea_store,
1471 info->lti_ea_store_size);
1472 info->lti_ea_store = buf->lb_buf;
1473 info->lti_ea_store_size = buf->lb_len;
1481 * there is LMV EA (striping information) in this object
1482 * let's parse it and create in-core objects for the stripes
1484 rc = lod_parse_dir_striping(env, lo, buf);
1486 lo->ldo_dir_stripe_loaded = 1;
1490 mutex_unlock(&lo->ldo_layout_mutex);
1495 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1496 const struct lu_buf *buf)
1502 mutex_lock(&lo->ldo_layout_mutex);
1503 rc = lod_parse_striping(env, lo, buf);
1504 mutex_unlock(&lo->ldo_layout_mutex);
1510 * Verify lov_user_md_v1/v3 striping.
1512 * Check the validity of all fields including the magic, stripe size,
1513 * stripe count, stripe offset and that the pool is present. Also check
1514 * that each target index points to an existing target. The additional
1515 * \a is_from_disk turns additional checks. In some cases zero fields
1516 * are allowed (like pattern=0).
1518 * \param[in] d LOD device
1519 * \param[in] buf buffer with LOV EA to verify
1520 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1521 * 1 - from disk, do not allow
1523 * \retval 0 if the striping is valid
1524 * \retval -EINVAL if striping is invalid
1526 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1529 struct lov_user_md_v1 *lum;
1530 struct lov_user_md_v3 *lum3;
1531 struct pool_desc *pool = NULL;
1535 __u16 stripe_offset;
1542 if (buf->lb_len < sizeof(*lum)) {
1543 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1545 GOTO(out, rc = -EINVAL);
1548 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1549 if (magic != LOV_USER_MAGIC_V1 &&
1550 magic != LOV_USER_MAGIC_V3 &&
1551 magic != LOV_USER_MAGIC_SPECIFIC) {
1552 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1553 le32_to_cpu(lum->lmm_magic));
1554 GOTO(out, rc = -EINVAL);
1557 /* the user uses "0" for default stripe pattern normally. */
1558 if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1559 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1561 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1562 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1563 le32_to_cpu(lum->lmm_pattern));
1564 GOTO(out, rc = -EINVAL);
1567 /* a released lum comes from creating orphan on hsm release,
1568 * doesn't make sense to verify it. */
1569 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1572 /* 64kB is the largest common page size we see (ia64), and matches the
1574 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1575 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1576 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1577 stripe_size, LOV_MIN_STRIPE_SIZE);
1578 GOTO(out, rc = -EINVAL);
1581 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1582 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1583 lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1584 /* if offset is not within valid range [0, osts_size) */
1585 if (stripe_offset >= d->lod_ost_descs.ltd_tgts_size) {
1586 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1587 stripe_offset, d->lod_ost_descs.ltd_tgts_size);
1588 GOTO(out, rc = -EINVAL);
1591 /* if lmm_stripe_offset is *not* in bitmap */
1592 if (!test_bit(stripe_offset, d->lod_ost_bitmap)) {
1593 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1595 GOTO(out, rc = -EINVAL);
1599 if (magic == LOV_USER_MAGIC_V1)
1600 lum_size = offsetof(struct lov_user_md_v1,
1602 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1603 lum_size = offsetof(struct lov_user_md_v3,
1606 GOTO(out, rc = -EINVAL);
1608 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1609 if (buf->lb_len < lum_size) {
1610 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1611 "magic %#x and stripe_count %u\n",
1612 buf->lb_len, lum_size, magic, stripe_count);
1613 GOTO(out, rc = -EINVAL);
1616 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1620 /* In the function below, .hs_keycmp resolves to
1621 * pool_hashkey_keycmp() */
1622 /* coverity[overrun-buffer-val] */
1623 pool = lod_find_pool(d, lum3->lmm_pool_name);
1627 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1628 rc = lod_check_index_in_pool(stripe_offset, pool);
1630 GOTO(out, rc = -EINVAL);
1633 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1634 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1635 stripe_count, pool_tgt_count(pool));
1636 GOTO(out, rc = -EINVAL);
1641 lod_pool_putref(pool);
1647 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1649 LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1650 LOV_USER_MAGIC_COMP_V1, "Wrong magic %x\n",
1651 le32_to_cpu(comp->lcm_magic));
1652 LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1653 "bad index %d, max = %d\n",
1654 i, le16_to_cpu(comp->lcm_entry_count));
1656 return &comp->lcm_entries[i];
1659 #define for_each_comp_entry_v1(comp, entry) \
1660 for (entry = comp_entry_v1(comp, 0); \
1661 entry <= comp_entry_v1(comp, \
1662 le16_to_cpu(comp->lcm_entry_count) - 1); \
1665 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
1666 struct lov_comp_md_entry_v1 *dom_ent)
1668 struct lov_comp_md_entry_v1 *ent;
1670 __u32 dom_off, dom_size, comp_size, off;
1672 unsigned int size, shift;
1674 entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1675 LASSERT(entries > 0);
1676 comp_v1->lcm_entry_count = cpu_to_le16(entries);
1678 comp_size = le32_to_cpu(comp_v1->lcm_size);
1679 dom_off = le32_to_cpu(dom_ent->lcme_offset);
1680 dom_size = le32_to_cpu(dom_ent->lcme_size);
1682 /* all entries offsets are shifted by entry size at least */
1683 shift = sizeof(*dom_ent);
1684 for_each_comp_entry_v1(comp_v1, ent) {
1685 off = le32_to_cpu(ent->lcme_offset);
1686 if (off == dom_off) {
1687 /* Entry deletion creates two holes in layout data:
1688 * - hole in entries array
1689 * - hole in layout data at dom_off with dom_size
1691 * First memmove is one entry shift from next entry
1692 * start with size up to dom_off in blob
1695 src = (void *)(ent + 1);
1696 size = (unsigned long)((void *)comp_v1 + dom_off - src);
1697 memmove(dst, src, size);
1698 /* take 'off' from just moved entry */
1699 off = le32_to_cpu(ent->lcme_offset);
1700 /* second memmove is blob tail after 'off' up to
1703 dst = (void *)comp_v1 + dom_off - sizeof(*ent);
1704 src = (void *)comp_v1 + off;
1705 size = (unsigned long)(comp_size - off);
1706 memmove(dst, src, size);
1707 /* all entries offsets after DoM entry are shifted by
1708 * dom_size additionally
1712 ent->lcme_offset = cpu_to_le32(off - shift);
1714 comp_v1->lcm_size = cpu_to_le32(comp_size - shift);
1716 /* notify a caller to re-check entry */
1720 void lod_dom_stripesize_recalc(struct lod_device *d)
1722 __u64 threshold_mb = d->lod_dom_threshold_free_mb;
1723 __u32 max_size = d->lod_dom_stripesize_max_kb;
1724 __u32 def_size = d->lod_dom_stripesize_cur_kb;
1726 /* use maximum allowed value if free space is above threshold */
1727 if (d->lod_lsfs_free_mb >= threshold_mb) {
1728 def_size = max_size;
1729 } else if (!d->lod_lsfs_free_mb || max_size <= LOD_DOM_MIN_SIZE_KB) {
1732 /* recalc threshold like it would be with def_size as max */
1733 threshold_mb = mult_frac(threshold_mb, def_size, max_size);
1734 if (d->lod_lsfs_free_mb < threshold_mb)
1735 def_size = rounddown(def_size / 2, LOD_DOM_MIN_SIZE_KB);
1736 else if (d->lod_lsfs_free_mb > threshold_mb * 2)
1737 def_size = max_t(unsigned int, def_size * 2,
1738 LOD_DOM_MIN_SIZE_KB);
1741 if (d->lod_dom_stripesize_cur_kb != def_size) {
1742 CDEBUG(D_LAYOUT, "Change default DOM stripe size %d->%d\n",
1743 d->lod_dom_stripesize_cur_kb, def_size);
1744 d->lod_dom_stripesize_cur_kb = def_size;
1748 static __u32 lod_dom_stripesize_limit(const struct lu_env *env,
1749 struct lod_device *d)
1753 /* set bfree as fraction of total space */
1754 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_SPOOF)) {
1755 spin_lock(&d->lod_lsfs_lock);
1756 d->lod_lsfs_free_mb = mult_frac(d->lod_lsfs_total_mb,
1757 min_t(int, cfs_fail_val, 100), 100);
1758 GOTO(recalc, rc = 0);
1761 if (d->lod_lsfs_age < ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE) {
1762 struct obd_statfs sfs;
1764 spin_lock(&d->lod_lsfs_lock);
1765 if (d->lod_lsfs_age > ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE)
1766 GOTO(unlock, rc = 0);
1768 d->lod_lsfs_age = ktime_get_seconds();
1769 spin_unlock(&d->lod_lsfs_lock);
1770 rc = dt_statfs(env, d->lod_child, &sfs);
1773 "%s: failed to get OSD statfs: rc = %d\n",
1774 lod2obd(d)->obd_name, rc);
1777 /* udpate local OSD cached statfs data */
1778 spin_lock(&d->lod_lsfs_lock);
1779 d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
1780 d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
1782 lod_dom_stripesize_recalc(d);
1784 spin_unlock(&d->lod_lsfs_lock);
1787 return d->lod_dom_stripesize_cur_kb << 10;
1790 int lod_dom_stripesize_choose(const struct lu_env *env, struct lod_device *d,
1791 struct lov_comp_md_v1 *comp_v1,
1792 struct lov_comp_md_entry_v1 *dom_ent,
1795 struct lov_comp_md_entry_v1 *ent;
1796 struct lu_extent *dom_ext, *ext;
1797 struct lov_user_md_v1 *lum;
1798 __u32 max_stripe_size;
1801 bool dom_next_entry = false;
1803 dom_ext = &dom_ent->lcme_extent;
1804 dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1805 max_stripe_size = lod_dom_stripesize_limit(env, d);
1807 /* Check stripe size againts current per-MDT limit */
1808 if (stripe_size <= max_stripe_size)
1811 lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1812 CDEBUG(D_LAYOUT, "overwrite DoM component size %u with MDT limit %u\n",
1813 stripe_size, max_stripe_size);
1814 lum->lmm_stripe_size = cpu_to_le32(max_stripe_size);
1816 /* In common case the DoM stripe is first entry in a mirror and
1817 * can be deleted only if it is not single entry in layout or
1818 * mirror, otherwise error should be returned.
1820 for_each_comp_entry_v1(comp_v1, ent) {
1824 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1828 ext = &ent->lcme_extent;
1829 if (ext->e_start != dom_ext->e_end)
1832 /* Found next component after the DoM one with the same
1833 * mirror_id and adjust its start with DoM component end.
1835 * NOTE: we are considering here that there can be only one
1836 * DoM component in a file, all replicas are located on OSTs
1837 * always and don't need adjustment since use own layouts.
1839 ext->e_start = cpu_to_le64(max_stripe_size);
1840 dom_next_entry = true;
1844 if (max_stripe_size == 0) {
1845 /* DoM component size is zero due to server setting, remove
1846 * it from the layout but only if next component exists in
1847 * the same mirror. That must be checked prior calling the
1848 * lod_erase_dom_stripe().
1850 if (!dom_next_entry)
1853 rc = lod_erase_dom_stripe(comp_v1, dom_ent);
1855 /* Update DoM extent end finally */
1856 dom_ext->e_end = cpu_to_le64(max_stripe_size);
1863 * Verify LOV striping.
1865 * \param[in] d LOD device
1866 * \param[in] buf buffer with LOV EA to verify
1867 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1868 * 1 - from disk, do not allow
1869 * \param[in] start extent start for composite layout
1871 * \retval 0 if the striping is valid
1872 * \retval -EINVAL if striping is invalid
1874 int lod_verify_striping(const struct lu_env *env, struct lod_device *d,
1875 struct lod_object *lo, const struct lu_buf *buf,
1878 struct lov_user_md_v1 *lum;
1879 struct lov_comp_md_v1 *comp_v1;
1880 struct lov_comp_md_entry_v1 *ent;
1881 struct lu_extent *ext;
1884 __u32 stripe_size = 0;
1885 __u16 prev_mid = -1, mirror_id = -1;
1891 if (buf->lb_len < sizeof(lum->lmm_magic)) {
1892 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
1898 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1899 /* treat foreign LOV EA/object case first
1900 * XXX is it expected to try setting again a foreign?
1901 * XXX should we care about different current vs new layouts ?
1903 if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
1904 struct lov_foreign_md *lfm = buf->lb_buf;
1906 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
1908 "buf len %zu < min lov_foreign_md size (%zu)\n",
1909 buf->lb_len, offsetof(typeof(*lfm),
1914 if (foreign_size_le(lfm) > buf->lb_len) {
1916 "buf len %zu < this lov_foreign_md size (%zu)\n",
1917 buf->lb_len, foreign_size_le(lfm));
1920 /* Don't do anything with foreign layouts */
1924 /* normal LOV/layout cases */
1926 if (buf->lb_len < sizeof(*lum)) {
1927 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1932 if (magic != LOV_USER_MAGIC_V1 &&
1933 magic != LOV_USER_MAGIC_V3 &&
1934 magic != LOV_USER_MAGIC_SPECIFIC &&
1935 magic != LOV_USER_MAGIC_COMP_V1) {
1936 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1937 le32_to_cpu(lum->lmm_magic));
1941 if (magic != LOV_USER_MAGIC_COMP_V1)
1942 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1944 /* magic == LOV_USER_MAGIC_COMP_V1 */
1945 comp_v1 = buf->lb_buf;
1946 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1947 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1948 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1954 if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1955 CDEBUG(D_LAYOUT, "entry count is zero\n");
1959 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1960 lo->ldo_comp_cnt > 0) {
1961 /* could be called from lustre.lov.add */
1962 __u32 cnt = lo->ldo_comp_cnt;
1964 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
1965 prev_end = ext->e_end;
1970 for_each_comp_entry_v1(comp_v1, ent) {
1971 ext = &ent->lcme_extent;
1973 if (le64_to_cpu(ext->e_start) > le64_to_cpu(ext->e_end) ||
1974 le64_to_cpu(ext->e_start) & (LOV_MIN_STRIPE_SIZE - 1) ||
1975 (le64_to_cpu(ext->e_end) != LUSTRE_EOF &&
1976 le64_to_cpu(ext->e_end) & (LOV_MIN_STRIPE_SIZE - 1))) {
1977 CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
1978 le64_to_cpu(ext->e_start),
1979 le64_to_cpu(ext->e_end));
1984 /* lcme_id contains valid value */
1985 if (le32_to_cpu(ent->lcme_id) == 0 ||
1986 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
1987 CDEBUG(D_LAYOUT, "invalid id %u\n",
1988 le32_to_cpu(ent->lcme_id));
1992 if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
1993 mirror_id = mirror_id_of(
1994 le32_to_cpu(ent->lcme_id));
1996 /* first component must start with 0 */
1997 if (mirror_id != prev_mid &&
1998 le64_to_cpu(ext->e_start) != 0) {
2000 "invalid start:%llu, expect:0\n",
2001 le64_to_cpu(ext->e_start));
2005 prev_mid = mirror_id;
2009 if (le64_to_cpu(ext->e_start) == 0) {
2014 /* the next must be adjacent with the previous one */
2015 if (le64_to_cpu(ext->e_start) != prev_end) {
2017 "invalid start actual:%llu, expect:%llu\n",
2018 le64_to_cpu(ext->e_start), prev_end);
2022 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
2023 tmp.lb_len = le32_to_cpu(ent->lcme_size);
2025 /* Check DoM entry is always the first one */
2027 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
2029 /* DoM component must be the first in a mirror */
2030 if (le64_to_cpu(ext->e_start) > 0) {
2031 CDEBUG(D_LAYOUT, "invalid DoM component "
2032 "with %llu extent start\n",
2033 le64_to_cpu(ext->e_start));
2036 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2037 /* There is just one stripe on MDT and it must
2038 * cover whole component size. */
2039 if (stripe_size != le64_to_cpu(ext->e_end)) {
2040 CDEBUG(D_LAYOUT, "invalid DoM layout "
2041 "stripe size %u != %llu "
2042 "(component size)\n",
2043 stripe_size, prev_end);
2046 /* Check and adjust stripe size by per-MDT limit */
2047 rc = lod_dom_stripesize_choose(env, d, comp_v1, ent,
2049 /* DoM entry was removed, re-check layout from start */
2050 if (rc == -ERESTART)
2055 /* Any stripe count is forbidden on DoM component */
2056 if (lum->lmm_stripe_count) {
2058 "invalid DoM layout stripe count %u, must be 0\n",
2059 le16_to_cpu(lum->lmm_stripe_count));
2063 /* Any pool is forbidden on DoM component */
2064 if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
2065 struct lov_user_md_v3 *v3 = (void *)lum;
2067 if (v3->lmm_pool_name[0] != '\0') {
2069 "DoM component cannot have pool assigned\n");
2075 prev_end = le64_to_cpu(ext->e_end);
2077 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
2081 if (prev_end == LUSTRE_EOF || ext->e_start == prev_end)
2084 /* extent end must be aligned with the stripe_size */
2085 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2086 if (stripe_size && prev_end % stripe_size) {
2087 CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
2088 "stripe_sz: %u, [%llu, %llu)\n",
2089 stripe_size, ext->e_start, prev_end);
2094 /* make sure that the mirror_count is telling the truth */
2095 if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
2102 * set the default stripe size, if unset.
2104 * \param[in,out] val number of bytes per OST stripe
2106 * The minimum stripe size is 64KB to ensure that a single stripe is an
2107 * even multiple of a client PAGE_SIZE (IA64, PPC, etc). Otherwise, it
2108 * is difficult to split dirty pages across OSCs during writes.
2110 void lod_fix_desc_stripe_size(__u64 *val)
2112 if (*val < LOV_MIN_STRIPE_SIZE) {
2114 LCONSOLE_INFO("Increasing default stripe size to "
2115 "minimum value %u\n",
2116 LOV_DESC_STRIPE_SIZE_DEFAULT);
2117 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
2118 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
2119 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
2120 LCONSOLE_WARN("Changing default stripe size to %llu (a "
2121 "multiple of %u)\n",
2122 *val, LOV_MIN_STRIPE_SIZE);
2127 * set the filesystem default number of stripes, if unset.
2129 * \param[in,out] val number of stripes
2131 * A value of "0" means "use the system-wide default stripe count", which
2132 * has either been inherited by now, or falls back to 1 stripe per file.
2133 * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
2134 * and is a valid value, so is left unchanged here.
2136 void lod_fix_desc_stripe_count(__u32 *val)
2143 * set the filesystem default layout pattern
2145 * \param[in,out] val LOV_PATTERN_* layout
2147 * A value of "0" means "use the system-wide default layout type", which
2148 * has either been inherited by now, or falls back to plain RAID0 striping.
2150 void lod_fix_desc_pattern(__u32 *val)
2152 /* from lov_setstripe */
2153 if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
2154 LCONSOLE_WARN("lod: Unknown stripe pattern: %#x\n", *val);
2159 void lod_fix_lmv_desc_pattern(__u32 *val)
2161 if ((*val) && !lmv_is_known_hash_type(*val)) {
2162 LCONSOLE_WARN("lod: Unknown md stripe pattern: %#x\n", *val);
2167 void lod_fix_desc_qos_maxage(__u32 *val)
2169 /* fix qos_maxage */
2171 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2175 * Used to fix insane default striping.
2177 * \param[in] desc striping description
2179 void lod_fix_desc(struct lov_desc *desc)
2181 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2182 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2183 lod_fix_desc_pattern(&desc->ld_pattern);
2184 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2187 static void lod_fix_lmv_desc(struct lmv_desc *desc)
2189 desc->ld_active_tgt_count = 0;
2190 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2191 lod_fix_lmv_desc_pattern(&desc->ld_pattern);
2192 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2196 * Initialize the structures used to store pools and default striping.
2198 * \param[in] lod LOD device
2199 * \param[in] lcfg configuration structure storing default striping.
2201 * \retval 0 if initialization succeeds
2202 * \retval negative error number on failure
2204 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2206 struct obd_device *obd;
2207 struct lov_desc *desc;
2211 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2212 LASSERT(obd != NULL);
2213 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2215 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2216 CERROR("LOD setup requires a descriptor\n");
2220 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2222 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2223 CERROR("descriptor size wrong: %d > %d\n",
2224 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2228 if (desc->ld_magic != LOV_DESC_MAGIC) {
2229 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2230 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2231 obd->obd_name, desc);
2232 lustre_swab_lov_desc(desc);
2234 CERROR("%s: Bad lov desc magic: %#x\n",
2235 obd->obd_name, desc->ld_magic);
2242 desc->ld_active_tgt_count = 0;
2243 lod->lod_ost_descs.ltd_lov_desc = *desc;
2245 /* NB: config doesn't contain lmv_desc, alter it via sysfs. */
2246 lod_fix_lmv_desc(&lod->lod_mdt_descs.ltd_lmv_desc);
2248 lod->lod_sp_me = LUSTRE_SP_CLI;
2250 /* Set up OST pool environment */
2251 lod->lod_pool_count = 0;
2252 rc = lod_pool_hash_init(&lod->lod_pools_hash_body);
2256 INIT_LIST_HEAD(&lod->lod_pool_list);
2257 lod->lod_pool_count = 0;
2258 rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_tgt_pool, 0);
2262 rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool, 0);
2264 GOTO(out_mdt_pool, rc);
2266 rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_tgt_pool, 0);
2268 GOTO(out_mdt_rr_pool, rc);
2270 rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
2272 GOTO(out_ost_pool, rc);
2277 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2279 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2281 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
2283 lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2289 * Release the structures describing the pools.
2291 * \param[in] lod LOD device from which we release the structures
2295 int lod_pools_fini(struct lod_device *lod)
2297 struct obd_device *obd = lod2obd(lod);
2298 struct pool_desc *pool, *tmp;
2301 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2302 /* free pool structs */
2303 CDEBUG(D_INFO, "delete pool %p\n", pool);
2304 /* In the function below, .hs_keycmp resolves to
2305 * pool_hashkey_keycmp() */
2306 /* coverity[overrun-buffer-val] */
2307 lod_pool_del(obd, pool->pool_name);
2310 lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2311 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool);
2312 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2313 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2314 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);