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 |
180 data->ocd_version = LUSTRE_VERSION_CODE;
181 data->ocd_index = index;
183 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
185 data->ocd_connect_flags |= OBD_CONNECT_AT |
188 #ifdef HAVE_LRU_RESIZE_SUPPORT
189 OBD_CONNECT_LRU_RESIZE |
192 OBD_CONNECT_REQPORTAL |
193 OBD_CONNECT_SKIP_ORPHAN |
195 OBD_CONNECT_LVB_TYPE |
196 OBD_CONNECT_VERSION |
197 OBD_CONNECT_PINGLESS |
199 OBD_CONNECT_BULK_MBITS;
200 data->ocd_connect_flags2 = OBD_CONNECT2_REPLAY_CREATE;
202 data->ocd_group = tgt_index;
203 ltd = &lod->lod_ost_descs;
205 struct obd_import *imp = obd->u.cli.cl_import;
208 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
209 data->ocd_connect_flags |= OBD_CONNECT_ACL |
211 OBD_CONNECT_MDS_MDS |
212 OBD_CONNECT_MULTIMODRPCS |
217 OBD_CONNECT_BULK_MBITS;
218 spin_lock(&imp->imp_lock);
219 imp->imp_server_timeout = 1;
220 spin_unlock(&imp->imp_lock);
221 imp->imp_client->cli_request_portal = OUT_PORTAL;
222 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
224 ltd = &lod->lod_mdt_descs;
227 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
230 CERROR("%s: cannot connect to next dev %s (%d)\n",
231 obd->obd_name, osp, rc);
232 GOTO(out_cleanup, rc);
236 /* Allocate ost descriptor and fill it */
237 OBD_ALLOC_PTR(tgt_desc);
239 GOTO(out_cleanup, rc = -ENOMEM);
241 tgt_desc->ltd_tgt = dt_dev;
242 tgt_desc->ltd_exp = exp;
243 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
244 tgt_desc->ltd_gen = gen;
245 tgt_desc->ltd_index = index;
246 tgt_desc->ltd_active = active;
248 down_write(<d->ltd_rw_sem);
249 mutex_lock(<d->ltd_mutex);
250 rc = ltd_add_tgt(ltd, tgt_desc);
254 rc = lu_qos_add_tgt(<d->ltd_qos, tgt_desc);
256 GOTO(out_del_tgt, rc);
258 rc = lu_tgt_pool_add(<d->ltd_tgt_pool, index,
259 ltd->ltd_lov_desc.ld_tgt_count);
261 CERROR("%s: can't set up pool, failed with %d\n",
263 GOTO(out_del_tgt, rc);
266 mutex_unlock(<d->ltd_mutex);
267 up_write(<d->ltd_rw_sem);
269 if (lod->lod_recovery_completed)
270 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
272 if (!for_ost && lod->lod_initialized) {
273 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
275 CERROR("%s: cannot start llog on %s:rc = %d\n",
276 lod2obd(lod)->obd_name, osp, rc);
281 rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
283 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
284 "index = %u, rc = %d\n", osp, type, index, rc);
285 GOTO(out_fini_llog, rc);
289 lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
290 &tgt_desc->ltd_recovery_task);
292 down_write(<d->ltd_rw_sem);
293 mutex_lock(<d->ltd_mutex);
294 lu_tgt_pool_remove(<d->ltd_tgt_pool, index);
296 ltd_del_tgt(ltd, tgt_desc);
298 mutex_unlock(<d->ltd_mutex);
299 up_write(<d->ltd_rw_sem);
300 OBD_FREE_PTR(tgt_desc);
302 /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
303 * objects from the MDT stack. See LU-7184. */
304 lcfg = &lod_env_info(env)->lti_lustre_cfg;
305 memset(lcfg, 0, sizeof(*lcfg));
306 lcfg->lcfg_version = LUSTRE_CFG_VERSION;
307 lcfg->lcfg_command = LCFG_CLEANUP;
308 lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
317 * Schedule target removal from the target table.
319 * Mark the device as dead. The device is not removed here because it may
320 * still be in use. The device will be removed in lod_putref() when the
321 * last reference is released.
323 * \param[in] env execution environment for this thread
324 * \param[in] lod LOD device the target table belongs to
325 * \param[in] ltd target table
326 * \param[in] tgt target
328 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
329 struct lod_tgt_descs *ltd, struct lu_tgt_desc *tgt)
331 lfsck_del_target(env, lod->lod_child, tgt->ltd_tgt, tgt->ltd_index,
334 if (!tgt->ltd_reap) {
336 ltd->ltd_death_row++;
341 * Schedule removal of all the targets from the given target table.
343 * See more details in the description for __lod_del_device()
345 * \param[in] env execution environment for this thread
346 * \param[in] lod LOD device the target table belongs to
347 * \param[in] ltd target table
351 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
352 struct lod_tgt_descs *ltd)
354 struct lu_tgt_desc *tgt;
356 if (ltd->ltd_tgts_size <= 0)
360 mutex_lock(<d->ltd_mutex);
361 ltd_foreach_tgt(ltd, tgt)
362 __lod_del_device(env, lod, ltd, tgt);
363 mutex_unlock(<d->ltd_mutex);
364 lod_putref(lod, ltd);
366 lu_tgt_descs_fini(ltd);
372 * Remove device by name.
374 * Remove a device identified by \a osp from the target table. Given
375 * the device can be in use, the real deletion happens in lod_putref().
377 * \param[in] env execution environment for this thread
378 * \param[in] lod LOD device to be connected to the new OSP
379 * \param[in] ltd target table
380 * \param[in] osp name of OSP device to be removed
381 * \param[in] idx index of the target
382 * \param[in] gen generation number, not used currently
384 * \retval 0 if the device was scheduled for removal
385 * \retval -EINVAL if no device was found
387 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
388 struct lod_tgt_descs *ltd, char *osp, unsigned int idx,
391 struct obd_device *obd;
392 struct lu_tgt_desc *tgt;
393 struct obd_uuid uuid;
398 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
400 obd_str2uuid(&uuid, osp);
402 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
403 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
405 CERROR("can't find %s device\n", osp);
410 CERROR("%s: request to remove OBD %s with invalid generation %d"
411 "\n", obd->obd_name, osp, gen);
415 obd_str2uuid(&uuid, osp);
418 mutex_lock(<d->ltd_mutex);
419 tgt = LTD_TGT(ltd, idx);
420 /* check that the index is allocated in the bitmap */
421 if (!test_bit(idx, ltd->ltd_tgt_bitmap) || !tgt) {
422 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
423 GOTO(out, rc = -EINVAL);
426 /* check that the UUID matches */
427 if (!obd_uuid_equals(&uuid, &tgt->ltd_uuid)) {
428 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
429 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), idx, osp);
430 GOTO(out, rc = -EINVAL);
433 __lod_del_device(env, lod, ltd, tgt);
436 mutex_unlock(<d->ltd_mutex);
437 lod_putref(lod, ltd);
442 * Resize per-thread storage to hold specified size.
444 * A helper function to resize per-thread temporary storage. This storage
445 * is used to process LOV/LVM EAs and may be quite large. We do not want to
446 * allocate/release it every time, so instead we put it into the env and
447 * reallocate on demand. The memory is released when the correspondent thread
450 * \param[in] info LOD-specific storage in the environment
451 * \param[in] size new size to grow the buffer to
453 * \retval 0 on success, -ENOMEM if reallocation failed
455 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
457 __u32 round = size_roundup_power2(size);
459 if (info->lti_ea_store) {
460 LASSERT(info->lti_ea_store_size);
461 LASSERT(info->lti_ea_store_size < round);
462 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
463 info->lti_ea_store_size, round);
464 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
465 info->lti_ea_store = NULL;
466 info->lti_ea_store_size = 0;
469 OBD_ALLOC_LARGE(info->lti_ea_store, round);
470 if (info->lti_ea_store == NULL)
472 info->lti_ea_store_size = round;
477 static void lod_free_comp_buffer(struct lod_layout_component *entries,
478 __u16 count, __u32 bufsize)
480 struct lod_layout_component *entry;
483 for (i = 0; i < count; i++) {
485 if (entry->llc_magic == LOV_MAGIC_FOREIGN)
487 if (entry->llc_pool != NULL)
488 lod_set_pool(&entry->llc_pool, NULL);
489 if (entry->llc_ostlist.op_array)
490 OBD_FREE(entry->llc_ostlist.op_array,
491 entry->llc_ostlist.op_size);
492 LASSERT(entry->llc_stripe == NULL);
493 LASSERT(entry->llc_stripes_allocated == 0);
497 OBD_FREE_LARGE(entries, bufsize);
500 void lod_free_def_comp_entries(struct lod_default_striping *lds)
502 lod_free_comp_buffer(lds->lds_def_comp_entries,
503 lds->lds_def_comp_size_cnt,
505 sizeof(*lds->lds_def_comp_entries) *
506 lds->lds_def_comp_size_cnt));
507 lds->lds_def_comp_entries = NULL;
508 lds->lds_def_comp_cnt = 0;
509 lds->lds_def_striping_is_composite = 0;
510 lds->lds_def_comp_size_cnt = 0;
514 * Resize per-thread storage to hold default striping component entries
516 * A helper function to resize per-thread temporary storage. This storage
517 * is used to hold default LOV/LVM EAs and may be quite large. We do not want
518 * to allocate/release it every time, so instead we put it into the env and
519 * reallocate it on demand. The memory is released when the correspondent
520 * thread is finished.
522 * \param[in,out] lds default striping
523 * \param[in] count new component count to grow the buffer to
525 * \retval 0 on success, -ENOMEM if reallocation failed
527 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
529 struct lod_layout_component *entries;
530 __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
532 __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
533 lds->lds_def_comp_size_cnt);
538 OBD_ALLOC_LARGE(entries, new);
542 if (lds->lds_def_comp_entries != NULL) {
543 CDEBUG(D_INFO, "default striping component size %d is not "
544 "enough, need %d\n", old, new);
545 lod_free_def_comp_entries(lds);
548 lds->lds_def_comp_entries = entries;
549 lds->lds_def_comp_size_cnt = count;
554 void lod_free_comp_entries(struct lod_object *lo)
556 if (lo->ldo_mirrors) {
557 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, lo->ldo_mirror_count);
558 lo->ldo_mirrors = NULL;
559 lo->ldo_mirror_count = 0;
561 lod_free_comp_buffer(lo->ldo_comp_entries,
563 sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
564 lo->ldo_comp_entries = NULL;
565 lo->ldo_comp_cnt = 0;
566 lo->ldo_is_composite = 0;
569 int lod_alloc_comp_entries(struct lod_object *lo,
570 int mirror_count, int comp_count)
572 LASSERT(comp_count != 0);
573 LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
575 if (mirror_count > 0) {
576 OBD_ALLOC_PTR_ARRAY(lo->ldo_mirrors, mirror_count);
577 if (!lo->ldo_mirrors)
580 lo->ldo_mirror_count = mirror_count;
583 OBD_ALLOC_LARGE(lo->ldo_comp_entries,
584 sizeof(*lo->ldo_comp_entries) * comp_count);
585 if (lo->ldo_comp_entries == NULL) {
586 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, mirror_count);
587 lo->ldo_mirrors = NULL;
588 lo->ldo_mirror_count = 0;
592 lo->ldo_comp_cnt = comp_count;
593 lo->ldo_is_foreign = 0;
597 int lod_fill_mirrors(struct lod_object *lo)
599 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
600 struct lod_layout_component *lod_comp;
601 bool found_preferred = false;
603 __u16 mirror_id = 0xffff;
607 LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
609 if (!lo->ldo_is_composite)
612 lod_comp = &lo->ldo_comp_entries[0];
614 for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
615 bool stale = lod_comp->llc_flags & LCME_FL_STALE;
616 bool preferred = lod_comp->llc_flags & LCME_FL_PREF_WR;
617 bool mirror_hsm = lod_is_hsm(lod_comp);
618 bool init = (lod_comp->llc_stripe != NULL) &&
619 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
620 !(lod_comp->llc_pattern & LOV_PATTERN_MDT);
624 /* calculate component preference over all used OSTs */
625 for (j = 0; init && j < lod_comp->llc_stripes_allocated; j++) {
626 __u32 idx = lod_comp->llc_ost_indices[j];
627 struct lod_tgt_desc *ltd;
629 if (lod_comp->llc_stripe[j] == NULL)
632 if (unlikely(idx >= lod->lod_ost_descs.ltd_tgts_size)) {
633 CERROR("%s: "DFID" OST idx %u > max %u\n",
634 lod2obd(lod)->obd_name,
635 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
636 idx, lod->lod_ost_descs.ltd_tgts_size);
639 ltd = OST_TGT(lod, idx);
640 if (unlikely(!ltd)) {
641 CERROR("%s: "DFID" OST idx %u is NULL\n",
642 lod2obd(lod)->obd_name,
643 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
648 if (ltd->ltd_statfs.os_state & OS_STATFS_NONROT)
652 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
653 /* Currently HSM mirror does not support PFL. */
654 if (lo->ldo_mirrors[mirror_idx].lme_hsm)
656 lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
657 lo->ldo_mirrors[mirror_idx].lme_prefer |= preferred;
658 lo->ldo_mirrors[mirror_idx].lme_preference += pref;
659 lo->ldo_mirrors[mirror_idx].lme_end = i;
663 if (mirror_idx >= 0 && preferred &&
664 !lo->ldo_mirrors[mirror_idx].lme_stale)
665 found_preferred = true;
669 if (mirror_idx >= lo->ldo_mirror_count)
672 if (mirror_hsm && (lod_comp->llc_extent.e_start != 0 ||
673 lod_comp->llc_extent.e_end != LUSTRE_EOF))
676 mirror_id = mirror_id_of(lod_comp->llc_id);
678 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
679 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
680 lo->ldo_mirrors[mirror_idx].lme_prefer = preferred;
681 lo->ldo_mirrors[mirror_idx].lme_hsm = mirror_hsm;
682 lo->ldo_mirrors[mirror_idx].lme_preference = pref;
683 lo->ldo_mirrors[mirror_idx].lme_start = i;
684 lo->ldo_mirrors[mirror_idx].lme_end = i;
686 if (mirror_idx != lo->ldo_mirror_count - 1)
689 if (!found_preferred && mirror_idx > 0) {
693 * if no explicited preferred found, then find a mirror
694 * with higher number of non-rotational OSTs
697 for (i = 0; i <= mirror_idx; i++) {
698 if (lo->ldo_mirrors[i].lme_stale)
700 if (lo->ldo_mirrors[i].lme_preference > pref) {
701 pref = lo->ldo_mirrors[i].lme_preference;
707 lo->ldo_mirrors[best].lme_prefer = 1;
714 * Generate on-disk lov_mds_md structure for each layout component based on
715 * the information in lod_object->ldo_comp_entries[i].
717 * \param[in] env execution environment for this thread
718 * \param[in] lo LOD object
719 * \param[in] comp_idx index of ldo_comp_entries
720 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
721 * \param[in|out] lmm_size buffer size/lmm size
722 * \param[in] is_dir generate lov ea for dir or file? For dir case,
723 * the stripe info is from the default stripe
724 * template, which is collected in lod_ah_init(),
725 * either from parent object or root object; for
726 * file case, it's from the @lo object
728 * \retval 0 if on disk structure is created successfully
729 * \retval negative error number on failure
731 static int lod_gen_component_ea(const struct lu_env *env,
732 struct lod_object *lo, int comp_idx,
733 struct lov_mds_md *lmm, int *lmm_size,
736 struct lod_thread_info *info = lod_env_info(env);
737 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
738 struct lod_device *lod;
739 struct lov_ost_data_v1 *objs;
740 struct lod_layout_component *lod_comp;
749 &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
751 lod_comp = &lo->ldo_comp_entries[comp_idx];
753 magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
754 if (lod_comp->llc_pattern == 0) /* default striping */
755 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
757 lmm->lmm_magic = cpu_to_le32(magic);
758 lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
759 fid_to_lmm_oi(fid, &lmm->lmm_oi);
760 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
761 lmm->lmm_oi.oi.oi_id++;
762 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
764 lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
765 lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
767 * for dir and uninstantiated component, lmm_layout_gen stores
768 * default stripe offset.
770 lmm->lmm_layout_gen =
771 (is_dir || !lod_comp_inited(lod_comp)) ?
772 cpu_to_le16(lod_comp->llc_stripe_offset) :
773 cpu_to_le16(lod_comp->llc_layout_gen);
775 if (magic == LOV_MAGIC_V1) {
776 objs = &lmm->lmm_objects[0];
778 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
779 size_t cplen = strlcpy(v3->lmm_pool_name,
781 sizeof(v3->lmm_pool_name));
782 if (cplen >= sizeof(v3->lmm_pool_name))
784 objs = &v3->lmm_objects[0];
786 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
787 stripe_count = lod_comp_entry_stripe_count(lo, comp_idx, is_dir);
788 if (stripe_count == 0 && !is_dir &&
789 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
790 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
791 /* Try again if all active targets are disconnected.
792 * It is possible when MDS does failover. */
793 if (!lod->lod_ost_active_count &&
799 if (!is_dir && lo->ldo_is_composite)
800 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
802 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
805 /* generate ost_idx of this component stripe */
806 for (i = 0; i < stripe_count; i++) {
807 struct dt_object *object;
808 __u32 ost_idx = (__u32)-1UL;
809 int type = LU_SEQ_RANGE_OST;
811 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
812 object = lod_comp->llc_stripe[i];
813 /* instantiated component */
814 info->lti_fid = *lu_object_fid(&object->do_lu);
816 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
818 if (cfs_fail_val == 0)
819 cfs_fail_val = info->lti_fid.f_oid;
821 info->lti_fid.f_oid = cfs_fail_val;
824 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
827 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
828 objs[i].l_ost_gen = cpu_to_le32(0);
829 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
832 rc = lod_fld_lookup(env, lod, &info->lti_fid,
835 CERROR("%s: Can not locate "DFID": rc = %d\n",
836 lod2obd(lod)->obd_name,
837 PFID(&info->lti_fid), rc);
840 } else if (lod_comp->llc_ostlist.op_array &&
841 lod_comp->llc_ostlist.op_count) {
842 /* user specified ost list */
843 ost_idx = lod_comp->llc_ostlist.op_array[i];
846 * with un-instantiated or with no specified ost list
847 * component, its l_ost_idx does not matter.
849 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
852 if (lmm_size != NULL)
853 *lmm_size = lov_mds_md_size(stripe_count, magic);
858 * Generate on-disk lov_hsm_md structure based on the information in
859 * the lod_object->ldo_comp_entries.
861 static int lod_gen_component_ea_foreign(const struct lu_env *env,
862 struct lod_object *lo,
863 struct lod_layout_component *lod_comp,
864 void *lmm, int *lmm_size)
866 struct lov_foreign_md *lfm = (struct lov_foreign_md *)lmm;
870 lfm->lfm_magic = cpu_to_le32(LOV_MAGIC_FOREIGN);
871 lfm->lfm_length = cpu_to_le32(lod_comp->llc_length);
872 lfm->lfm_type = cpu_to_le32(lod_comp->llc_type);
873 lfm->lfm_flags = cpu_to_le32(lod_comp->llc_foreign_flags);
875 if (lov_hsm_type_supported(lod_comp->llc_type)) {
876 if (lod_comp->llc_length != sizeof(struct lov_hsm_base))
879 lov_foreign_hsm_to_le(lfm, &lod_comp->llc_hsm);
883 *lmm_size = lov_foreign_md_size(lod_comp->llc_length);
889 * Generate on-disk lov_mds_md structure based on the information in
890 * the lod_object->ldo_comp_entries.
892 * \param[in] env execution environment for this thread
893 * \param[in] lo LOD object
894 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
895 * \param[in|out] lmm_size buffer size/lmm size
896 * \param[in] is_dir generate lov ea for dir or file? For dir case,
897 * the stripe info is from the default stripe
898 * template, which is collected in lod_ah_init(),
899 * either from parent object or root object; for
900 * file case, it's from the @lo object
902 * \retval 0 if on disk structure is created successfully
903 * \retval negative error number on failure
905 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
906 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
908 struct lov_comp_md_entry_v1 *lcme;
909 struct lov_comp_md_v1 *lcm;
910 struct lod_layout_component *comp_entries;
911 __u16 comp_cnt, mirror_cnt;
912 bool is_composite, is_foreign = false;
913 int i, rc = 0, offset;
917 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
918 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
919 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
921 lo->ldo_def_striping->lds_def_striping_is_composite;
923 comp_cnt = lo->ldo_comp_cnt;
924 mirror_cnt = lo->ldo_mirror_count;
925 comp_entries = lo->ldo_comp_entries;
926 is_composite = lo->ldo_is_composite;
927 is_foreign = lo->ldo_is_foreign;
930 LASSERT(lmm_size != NULL);
933 struct lov_foreign_md *lfm;
935 lfm = (struct lov_foreign_md *)lmm;
936 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
937 /* need to store little-endian */
938 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
939 __swab32s(&lfm->lfm_magic);
940 __swab32s(&lfm->lfm_length);
941 __swab32s(&lfm->lfm_type);
942 __swab32s(&lfm->lfm_flags);
944 *lmm_size = lo->ldo_foreign_lov_size;
948 LASSERT(comp_cnt != 0 && comp_entries != NULL);
951 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
955 lcm = (struct lov_comp_md_v1 *)lmm;
956 memset(lcm, 0, sizeof(*lcm));
958 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
959 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
960 lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
962 lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
964 lcm->lcm_flags = LCM_FL_NONE;
966 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
967 LASSERT(offset % sizeof(__u64) == 0);
969 for (i = 0; i < comp_cnt; i++) {
970 struct lod_layout_component *lod_comp;
971 struct lov_mds_md *sub_md;
974 lod_comp = &comp_entries[i];
975 lcme = &lcm->lcm_entries[i];
977 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
978 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
980 /* component could be un-inistantiated */
981 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
982 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
983 lcme->lcme_timestamp =
984 cpu_to_le64(lod_comp->llc_timestamp);
985 if (lod_comp->llc_flags & LCME_FL_EXTENSION && !is_dir)
986 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
988 lcme->lcme_extent.e_start =
989 cpu_to_le64(lod_comp->llc_extent.e_start);
990 lcme->lcme_extent.e_end =
991 cpu_to_le64(lod_comp->llc_extent.e_end);
992 lcme->lcme_offset = cpu_to_le32(offset);
994 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
995 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN) {
996 if (!lov_hsm_type_supported(lod_comp->llc_type)) {
997 CDEBUG(D_LAYOUT, "Unknown HSM type: %u\n",
999 GOTO(out, rc = -EINVAL);
1001 rc = lod_gen_component_ea_foreign(env, lo, lod_comp,
1004 rc = lod_gen_component_ea(env, lo, i, sub_md,
1009 lcme->lcme_size = cpu_to_le32(size);
1011 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
1012 "offset:%d lmm_size:%d\n", offset, *lmm_size);
1014 lcm->lcm_size = cpu_to_le32(offset);
1015 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
1017 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
1028 * Fill lti_ea_store buffer in the environment with a value for the given
1029 * EA. The buffer is reallocated if the value doesn't fit.
1031 * \param[in,out] env execution environment for this thread
1032 * .lti_ea_store buffer is filled with EA's value
1033 * \param[in] lo LOD object
1034 * \param[in] name name of the EA
1036 * \retval > 0 if EA is fetched successfully
1037 * \retval 0 if EA is empty
1038 * \retval negative error number on failure
1040 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
1043 struct lod_thread_info *info = lod_env_info(env);
1044 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1050 if (unlikely(info->lti_ea_store == NULL)) {
1051 /* just to enter in allocation block below */
1055 info->lti_buf.lb_buf = info->lti_ea_store;
1056 info->lti_buf.lb_len = info->lti_ea_store_size;
1057 rc = dt_xattr_get(env, next, &info->lti_buf, name);
1060 /* if object is not striped or inaccessible */
1061 if (rc == -ENODATA || rc == -ENOENT)
1064 if (rc == -ERANGE) {
1065 /* EA doesn't fit, reallocate new buffer */
1066 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
1067 if (rc == -ENODATA || rc == -ENOENT)
1073 rc = lod_ea_store_resize(info, rc);
1083 * Verify the target index is present in the current configuration.
1085 * \param[in] md LOD device where the target table is stored
1086 * \param[in] idx target's index
1088 * \retval 0 if the index is present
1089 * \retval -EINVAL if not
1091 int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1093 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1094 !test_bit(idx, md->lod_ost_bitmap))) {
1095 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1096 md->lod_ost_descs.ltd_tgts_size);
1100 if (unlikely(OST_TGT(md, idx) == NULL)) {
1101 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1102 lod2obd(md)->obd_name, idx);
1106 if (unlikely(OST_TGT(md, idx)->ltd_tgt == NULL)) {
1107 CERROR("%s: invalid lod device, for idx: %d\n",
1108 lod2obd(md)->obd_name , idx);
1116 * Instantiate objects for stripes.
1118 * Allocate and initialize LU-objects representing the stripes. The number
1119 * of the stripes (llc_stripe_count) must be initialized already. The caller
1120 * must ensure nobody else is calling the function on the object at the same
1121 * time. FLDB service must be running to be able to map a FID to the targets
1122 * and find appropriate device representing that target.
1124 * \param[in] env execution environment for this thread
1125 * \param[in,out] lo LOD object
1126 * \param[in] objs an array of IDs to creates the objects from
1127 * \param[in] comp_idx index of ldo_comp_entries
1129 * \retval 0 if the objects are instantiated successfully
1130 * \retval negative error number on failure
1132 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1133 struct lov_ost_data_v1 *objs, int comp_idx)
1135 struct lod_layout_component *lod_comp;
1136 struct lod_thread_info *info = lod_env_info(env);
1137 struct lod_device *md;
1138 struct lu_object *o, *n;
1139 struct lu_device *nd;
1140 struct dt_object **stripe = NULL;
1141 __u32 *ost_indices = NULL;
1147 LASSERT(lo != NULL);
1148 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1150 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1151 lod_comp = &lo->ldo_comp_entries[comp_idx];
1153 LASSERT(lod_comp->llc_stripe == NULL);
1154 LASSERT(lod_comp->llc_stripe_count > 0);
1155 LASSERT(lod_comp->llc_stripe_size > 0);
1157 stripe_len = lod_comp->llc_stripe_count;
1158 OBD_ALLOC_PTR_ARRAY(stripe, stripe_len);
1161 OBD_ALLOC_PTR_ARRAY(ost_indices, stripe_len);
1163 GOTO(out, rc = -ENOMEM);
1165 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1166 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1169 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1170 idx = le32_to_cpu(objs[i].l_ost_idx);
1171 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1174 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1175 PFID(&info->lti_fid));
1176 lod_getref(&md->lod_ost_descs);
1178 rc = validate_lod_and_idx(md, idx);
1179 if (unlikely(rc != 0)) {
1180 lod_putref(md, &md->lod_ost_descs);
1184 nd = &OST_TGT(md, idx)->ltd_tgt->dd_lu_dev;
1185 lod_putref(md, &md->lod_ost_descs);
1187 /* In the function below, .hs_keycmp resolves to
1188 * u_obj_hop_keycmp() */
1189 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1191 GOTO(out, rc = PTR_ERR(o));
1193 n = lu_object_locate(o->lo_header, nd->ld_type);
1196 stripe[i] = container_of(n, struct dt_object, do_lu);
1197 ost_indices[i] = idx;
1202 for (i = 0; i < stripe_len; i++)
1203 if (stripe[i] != NULL)
1204 dt_object_put(env, stripe[i]);
1206 OBD_FREE_PTR_ARRAY(stripe, stripe_len);
1207 lod_comp->llc_stripe_count = 0;
1209 OBD_FREE_PTR_ARRAY(ost_indices, stripe_len);
1211 lod_comp->llc_stripe = stripe;
1212 lod_comp->llc_ost_indices = ost_indices;
1213 lod_comp->llc_stripes_allocated = stripe_len;
1219 int lod_init_comp_foreign(struct lod_layout_component *lod_comp, void *lmm)
1221 struct lov_foreign_md *lfm;
1223 lfm = (struct lov_foreign_md *)lmm;
1224 lod_comp->llc_length = le32_to_cpu(lfm->lfm_length);
1225 lod_comp->llc_type = le32_to_cpu(lfm->lfm_type);
1227 if (!lov_hsm_type_supported(lod_comp->llc_type)) {
1229 "Unsupport HSM type: %u length: %u flags: %08X\n",
1230 lod_comp->llc_type, lod_comp->llc_length,
1231 le32_to_cpu(lfm->lfm_flags));
1236 * Currently it only stores the file FID as the field @lhm_archive_uuid
1237 * which is used to be the identifier within HSM backend for the archive
1239 * Thus the length of foreign layout value (HSM is a kind of foreign
1240 * layout type) is: sizeof(lhm_archive_id) + sizeof(lhm_archive_ver) +
1242 * It should fix to support other kinds of identifier for different HSM
1243 * solutions such as S3.
1245 if (lod_comp->llc_length != sizeof(struct lov_hsm_base)) {
1246 CDEBUG(D_LAYOUT, "Invalid HSM len: %u, should be %zu\n",
1247 lod_comp->llc_length, sizeof(struct lov_hsm_base));
1251 lod_comp->llc_foreign_flags = le32_to_cpu(lfm->lfm_flags);
1252 lov_foreign_hsm_to_cpu(&lod_comp->llc_hsm, lfm);
1257 * Instantiate objects for striping.
1259 * Parse striping information in \a buf and instantiate the objects
1260 * representing the stripes.
1262 * \param[in] env execution environment for this thread
1263 * \param[in] lo LOD object
1264 * \param[in] buf buffer storing LOV EA to parse
1265 * \param[in] lvf verify flags when parsing the layout
1267 * \retval 0 if parsing and objects creation succeed
1268 * \retval negative error number on failure
1270 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1271 const struct lu_buf *buf, enum layout_verify_flags lvf)
1273 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1274 struct lov_mds_md_v1 *lmm;
1275 struct lov_comp_md_v1 *comp_v1 = NULL;
1276 struct lov_foreign_md *foreign = NULL;
1277 struct lov_ost_data_v1 *objs;
1278 __u32 magic, pattern;
1279 __u16 mirror_cnt = 0;
1283 __u16 mirror_id = MIRROR_ID_NEG;
1285 int stale_mirrors = 0;
1289 LASSERT(buf->lb_buf);
1290 LASSERT(buf->lb_len);
1291 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1293 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1294 magic = le32_to_cpu(lmm->lmm_magic);
1296 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1297 magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN &&
1298 magic != LOV_MAGIC_SEL)
1299 GOTO(out, rc = -EINVAL);
1301 lod_striping_free_nolock(env, lo);
1303 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1304 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1305 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1307 GOTO(out, rc = -EINVAL);
1308 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1309 lo->ldo_is_composite = 1;
1310 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1312 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1315 lo->ldo_flr_state = LCM_FL_NONE;
1316 } else if (magic == LOV_MAGIC_FOREIGN) {
1319 foreign = (struct lov_foreign_md *)buf->lb_buf;
1320 length = offsetof(typeof(*foreign), lfm_value);
1321 if (buf->lb_len < length ||
1322 buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1324 "buf len %zu too small for lov_foreign_md\n",
1326 GOTO(out, rc = -EINVAL);
1329 /* just cache foreign LOV EA raw */
1330 rc = lod_alloc_foreign_lov(lo, length);
1333 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1337 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1338 lo->ldo_is_composite = 0;
1341 rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1345 for (i = 0; i < comp_cnt; i++) {
1346 struct lod_layout_component *lod_comp;
1347 struct lu_extent *ext;
1350 lod_comp = &lo->ldo_comp_entries[i];
1351 if (lo->ldo_is_composite) {
1352 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1353 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1355 ext = &comp_v1->lcm_entries[i].lcme_extent;
1356 lod_comp->llc_extent.e_start =
1357 le64_to_cpu(ext->e_start);
1358 if (lod_comp->llc_extent.e_start &
1359 (LOV_MIN_STRIPE_SIZE - 1)) {
1361 "extent start %llu is not a multiple of min size %u\n",
1362 lod_comp->llc_extent.e_start,
1363 LOV_MIN_STRIPE_SIZE);
1364 GOTO(out, rc = -EINVAL);
1367 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1368 if (lod_comp->llc_extent.e_end != LUSTRE_EOF &&
1369 lod_comp->llc_extent.e_end &
1370 (LOV_MIN_STRIPE_SIZE - 1)) {
1372 "extent end %llu is not a multiple of min size %u\n",
1373 lod_comp->llc_extent.e_end,
1374 LOV_MIN_STRIPE_SIZE);
1375 GOTO(out, rc = -EINVAL);
1378 lod_comp->llc_flags =
1379 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1381 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1382 lod_comp->llc_timestamp = le64_to_cpu(
1383 comp_v1->lcm_entries[i].lcme_timestamp);
1385 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1386 if (lod_comp->llc_id == LCME_ID_INVAL)
1387 GOTO(out, rc = -EINVAL);
1389 if (lvf & LVF_ALL_STALE) {
1390 if (mirror_id_of(lod_comp->llc_id) ==
1392 /* remaining comps in the mirror */
1393 stale |= lod_comp->llc_flags &
1397 * new mirror, check last mirror's
1404 mirror_id_of(lod_comp->llc_id);
1406 /* the first comp of the new mirror */
1407 stale = lod_comp->llc_flags &
1412 if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
1413 comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
1414 CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
1415 lod2obd(d)->obd_name,
1416 lod_comp->llc_flags, lod_comp->llc_id, i,
1417 PFID(lod_object_fid(lo)),
1418 le32_to_cpu(comp_v1->lcm_magic));
1421 lod_comp->llc_magic = le32_to_cpu(lmm->lmm_magic);
1422 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN) {
1423 rc = lod_init_comp_foreign(lod_comp, lmm);
1429 lod_comp->llc_magic = le32_to_cpu(lmm->lmm_magic);
1430 lod_comp_set_init(lod_comp);
1433 pattern = le32_to_cpu(lmm->lmm_pattern);
1434 if (!lov_pattern_supported(lov_pattern(pattern)))
1435 GOTO(out, rc = -EINVAL);
1437 if (pattern & LOV_PATTERN_MDT) {
1438 if (lod_comp->llc_extent.e_start != 0) {
1439 CERROR("%s: DOM entry must be the first stripe "
1440 "in a mirror\n", lod2obd(d)->obd_name);
1441 GOTO(out, rc = -EINVAL);
1444 dom_size = lod_comp->llc_extent.e_end;
1445 } else if (dom_size != lod_comp->llc_extent.e_end) {
1446 CERROR("%s: DOM entries with different sizes "
1447 "%#llx/%#llx\n", lod2obd(d)->obd_name,
1448 dom_size, lod_comp->llc_extent.e_end);
1449 GOTO(out, rc = -EINVAL);
1453 lod_comp->llc_pattern = pattern;
1454 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1455 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1456 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1458 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1459 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1461 lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1462 objs = &v3->lmm_objects[0];
1464 lod_set_pool(&lod_comp->llc_pool, NULL);
1465 objs = &lmm->lmm_objects[0];
1469 * If uninstantiated template component has valid l_ost_idx,
1470 * then user has specified ost list for this component.
1472 if (!lod_comp_inited(lod_comp)) {
1475 if (objs[0].l_ost_idx != (__u32)-1UL) {
1478 stripe_count = lod_comp_entry_stripe_count(
1480 if (stripe_count == 0 &&
1481 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1482 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1483 GOTO(out, rc = -E2BIG);
1485 * load the user specified ost list, when this
1486 * component is instantiated later, it will be
1487 * used in lod_alloc_ost_list().
1489 lod_comp->llc_ostlist.op_count = stripe_count;
1490 lod_comp->llc_ostlist.op_size =
1491 stripe_count * sizeof(__u32);
1492 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1493 lod_comp->llc_ostlist.op_size);
1494 if (!lod_comp->llc_ostlist.op_array)
1495 GOTO(out, rc = -ENOMEM);
1497 for (j = 0; j < stripe_count; j++)
1498 lod_comp->llc_ostlist.op_array[j] =
1499 le32_to_cpu(objs[j].l_ost_idx);
1502 * this component OST objects starts from the
1503 * first ost_idx, lod_alloc_ost_list() will
1506 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1509 * for uninstantiated component,
1510 * lmm_layout_gen stores default stripe offset.
1512 lod_comp->llc_stripe_offset =
1513 lmm->lmm_layout_gen;
1517 /* skip un-instantiated component object initialization */
1518 if (!lod_comp_inited(lod_comp))
1521 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1522 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1523 rc = lod_initialize_objects(env, lo, objs, i);
1529 if (lo->ldo_is_composite && (lvf & LVF_ALL_STALE)) {
1530 /* check the last mirror stale-ness */
1534 if (mirror_cnt == stale_mirrors) {
1536 CERROR("%s: can not set all stale mirrors for "
1538 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
1544 rc = lod_fill_mirrors(lo);
1550 lod_striping_free_nolock(env, lo);
1555 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1556 * is already cached.
1558 * \param[in] lo LOD object
1560 * \retval True if the striping is cached, otherwise
1563 static bool lod_striping_loaded(struct lod_object *lo)
1565 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1566 lo->ldo_comp_cached)
1569 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1570 if (lo->ldo_dir_stripe_loaded)
1573 /* Never load LMV stripe for slaves of striped dir */
1574 if (lo->ldo_dir_slave_stripe)
1582 * A generic function to initialize the stripe objects.
1584 * A protected version of lod_striping_load_locked() - load the striping
1585 * information from storage, parse that and instantiate LU objects to
1586 * represent the stripes. The LOD object \a lo supplies a pointer to the
1587 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1588 * return an array of references to the newly instantiated objects.
1590 * \param[in] env execution environment for this thread
1591 * \param[in,out] lo LOD object, where striping is stored and
1592 * which gets an array of references
1594 * \retval 0 if parsing and object creation succeed
1595 * \retval negative error number on failure
1597 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1599 struct lod_thread_info *info = lod_env_info(env);
1600 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1601 struct lu_buf *buf = &info->lti_buf;
1606 if (!dt_object_exists(next))
1609 if (lod_striping_loaded(lo))
1612 mutex_lock(&lo->ldo_layout_mutex);
1613 if (lod_striping_loaded(lo))
1614 GOTO(unlock, rc = 0);
1616 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1617 rc = lod_get_lov_ea(env, lo);
1622 * there is LOV EA (striping information) in this object
1623 * let's parse it and create in-core objects for the stripes
1625 buf->lb_buf = info->lti_ea_store;
1626 buf->lb_len = info->lti_ea_store_size;
1627 rc = lod_parse_striping(env, lo, buf, 0);
1629 lo->ldo_comp_cached = 1;
1630 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1631 rc = lod_get_lmv_ea(env, lo);
1632 if (rc > sizeof(struct lmv_foreign_md)) {
1633 struct lmv_foreign_md *lfm = info->lti_ea_store;
1635 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
1636 lo->ldo_foreign_lmv = info->lti_ea_store;
1637 lo->ldo_foreign_lmv_size =
1638 info->lti_ea_store_size;
1639 info->lti_ea_store = NULL;
1640 info->lti_ea_store_size = 0;
1642 lo->ldo_dir_stripe_loaded = 1;
1643 lo->ldo_is_foreign = 1;
1644 GOTO(unlock, rc = 0);
1648 if (rc < (int)sizeof(struct lmv_mds_md_v1)) {
1649 /* Let's set stripe_loaded to avoid further
1650 * stripe loading especially for non-stripe directory,
1651 * which can hurt performance. (See LU-9840)
1654 lo->ldo_dir_stripe_loaded = 1;
1655 GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1657 buf->lb_buf = info->lti_ea_store;
1658 buf->lb_len = info->lti_ea_store_size;
1659 if (rc == sizeof(struct lmv_mds_md_v1)) {
1660 rc = lod_load_lmv_shards(env, lo, buf, true);
1661 if (buf->lb_buf != info->lti_ea_store) {
1662 OBD_FREE_LARGE(info->lti_ea_store,
1663 info->lti_ea_store_size);
1664 info->lti_ea_store = buf->lb_buf;
1665 info->lti_ea_store_size = buf->lb_len;
1673 * there is LMV EA (striping information) in this object
1674 * let's parse it and create in-core objects for the stripes
1676 rc = lod_parse_dir_striping(env, lo, buf);
1678 lo->ldo_dir_stripe_loaded = 1;
1682 mutex_unlock(&lo->ldo_layout_mutex);
1687 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1688 const struct lu_buf *buf, enum layout_verify_flags lvf)
1694 mutex_lock(&lo->ldo_layout_mutex);
1695 rc = lod_parse_striping(env, lo, buf, lvf);
1696 mutex_unlock(&lo->ldo_layout_mutex);
1702 * Verify lov_user_md_v1/v3 striping.
1704 * Check the validity of all fields including the magic, stripe size,
1705 * stripe count, stripe offset and that the pool is present. Also check
1706 * that each target index points to an existing target. The additional
1707 * \a is_from_disk turns additional checks. In some cases zero fields
1708 * are allowed (like pattern=0).
1710 * \param[in] d LOD device
1711 * \param[in] buf buffer with LOV EA to verify
1712 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1713 * 1 - from disk, do not allow
1715 * \retval 0 if the striping is valid
1716 * \retval -EINVAL if striping is invalid
1718 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1721 struct lov_user_md_v1 *lum;
1722 struct lov_user_md_v3 *lum3;
1723 struct pool_desc *pool = NULL;
1727 __u16 stripe_offset;
1734 if (buf->lb_len < sizeof(*lum)) {
1735 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1737 GOTO(out, rc = -EINVAL);
1740 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1741 if (magic != LOV_USER_MAGIC_V1 &&
1742 magic != LOV_USER_MAGIC_V3 &&
1743 magic != LOV_USER_MAGIC_SPECIFIC) {
1744 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1745 le32_to_cpu(lum->lmm_magic));
1746 GOTO(out, rc = -EINVAL);
1749 /* the user uses "0" for default stripe pattern normally. */
1750 if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1751 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1753 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1754 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1755 le32_to_cpu(lum->lmm_pattern));
1756 GOTO(out, rc = -EINVAL);
1759 /* a released lum comes from creating orphan on hsm release,
1760 * doesn't make sense to verify it. */
1761 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1764 /* 64kB is the largest common page size we see (ia64), and matches the
1766 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1767 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1768 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1769 stripe_size, LOV_MIN_STRIPE_SIZE);
1770 GOTO(out, rc = -EINVAL);
1773 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1774 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1775 !(lov_pattern(le32_to_cpu(lum->lmm_pattern)) & LOV_PATTERN_MDT)) {
1776 /* if offset is not within valid range [0, osts_size) */
1777 if (stripe_offset >= d->lod_ost_descs.ltd_tgts_size) {
1778 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1779 stripe_offset, d->lod_ost_descs.ltd_tgts_size);
1780 GOTO(out, rc = -EINVAL);
1783 /* if lmm_stripe_offset is *not* in bitmap */
1784 if (!test_bit(stripe_offset, d->lod_ost_bitmap)) {
1785 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1787 GOTO(out, rc = -EINVAL);
1791 if (magic == LOV_USER_MAGIC_V1)
1792 lum_size = offsetof(struct lov_user_md_v1,
1794 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1795 lum_size = offsetof(struct lov_user_md_v3,
1798 GOTO(out, rc = -EINVAL);
1800 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1801 if (buf->lb_len < lum_size) {
1802 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1803 "magic %#x and stripe_count %u\n",
1804 buf->lb_len, lum_size, magic, stripe_count);
1805 GOTO(out, rc = -EINVAL);
1808 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1812 /* In the function below, .hs_keycmp resolves to
1813 * pool_hashkey_keycmp() */
1814 pool = lod_find_pool(d, lum3->lmm_pool_name);
1818 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1819 rc = lod_check_index_in_pool(stripe_offset, pool);
1821 GOTO(out, rc = -EINVAL);
1824 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1825 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1826 stripe_count, pool_tgt_count(pool));
1827 GOTO(out, rc = -EINVAL);
1832 lod_pool_putref(pool);
1838 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1840 LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1841 LOV_USER_MAGIC_COMP_V1 ||
1842 (le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1843 LOV_USER_MAGIC_SEL, "Wrong magic %x\n",
1844 le32_to_cpu(comp->lcm_magic));
1845 LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1846 "bad index %d, max = %d\n",
1847 i, le16_to_cpu(comp->lcm_entry_count));
1849 return &comp->lcm_entries[i];
1852 #define for_each_comp_entry_v1(comp, entry) \
1853 for (entry = comp_entry_v1(comp, 0); \
1854 entry <= comp_entry_v1(comp, \
1855 le16_to_cpu(comp->lcm_entry_count) - 1); \
1858 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
1859 struct lov_comp_md_entry_v1 *dom_ent)
1861 struct lov_comp_md_entry_v1 *ent;
1863 __u32 dom_off, dom_size, comp_size, off;
1865 unsigned int size, shift;
1867 entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1868 LASSERT(entries > 0);
1869 comp_v1->lcm_entry_count = cpu_to_le16(entries);
1871 comp_size = le32_to_cpu(comp_v1->lcm_size);
1872 dom_off = le32_to_cpu(dom_ent->lcme_offset);
1873 dom_size = le32_to_cpu(dom_ent->lcme_size);
1875 /* all entries offsets are shifted by entry size at least */
1876 shift = sizeof(*dom_ent);
1877 for_each_comp_entry_v1(comp_v1, ent) {
1878 off = le32_to_cpu(ent->lcme_offset);
1879 if (off == dom_off) {
1880 /* Entry deletion creates two holes in layout data:
1881 * - hole in entries array
1882 * - hole in layout data at dom_off with dom_size
1884 * First memmove is one entry shift from next entry
1885 * start with size up to dom_off in blob
1888 src = (void *)(ent + 1);
1889 size = (unsigned long)((void *)comp_v1 + dom_off - src);
1890 memmove(dst, src, size);
1891 /* take 'off' from just moved entry */
1892 off = le32_to_cpu(ent->lcme_offset);
1893 /* second memmove is blob tail after 'off' up to
1896 dst = (void *)comp_v1 + dom_off - sizeof(*ent);
1897 src = (void *)comp_v1 + off;
1898 size = (unsigned long)(comp_size - off);
1899 memmove(dst, src, size);
1900 /* all entries offsets after DoM entry are shifted by
1901 * dom_size additionally
1905 ent->lcme_offset = cpu_to_le32(off - shift);
1907 comp_v1->lcm_size = cpu_to_le32(comp_size - shift);
1909 /* notify a caller to re-check entry */
1913 void lod_dom_stripesize_recalc(struct lod_device *d)
1915 __u64 threshold_mb = d->lod_dom_threshold_free_mb;
1916 __u32 max_size = d->lod_dom_stripesize_max_kb;
1917 __u32 def_size = d->lod_dom_stripesize_cur_kb;
1919 /* use maximum allowed value if free space is above threshold */
1920 if (d->lod_lsfs_free_mb >= threshold_mb) {
1921 def_size = max_size;
1922 } else if (!d->lod_lsfs_free_mb || max_size <= LOD_DOM_MIN_SIZE_KB) {
1925 /* recalc threshold like it would be with def_size as max */
1926 threshold_mb = mult_frac(threshold_mb, def_size, max_size);
1927 if (d->lod_lsfs_free_mb < threshold_mb)
1928 def_size = rounddown(def_size / 2, LOD_DOM_MIN_SIZE_KB);
1929 else if (d->lod_lsfs_free_mb > threshold_mb * 2)
1930 def_size = max_t(unsigned int, def_size * 2,
1931 LOD_DOM_MIN_SIZE_KB);
1934 if (d->lod_dom_stripesize_cur_kb != def_size) {
1935 CDEBUG(D_LAYOUT, "Change default DOM stripe size %d->%d\n",
1936 d->lod_dom_stripesize_cur_kb, def_size);
1937 d->lod_dom_stripesize_cur_kb = def_size;
1941 static __u32 lod_dom_stripesize_limit(const struct lu_env *env,
1942 struct lod_device *d)
1946 /* set bfree as fraction of total space */
1947 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_STATFS_SPOOF)) {
1948 spin_lock(&d->lod_lsfs_lock);
1949 d->lod_lsfs_free_mb = mult_frac(d->lod_lsfs_total_mb,
1950 min_t(int, cfs_fail_val, 100), 100);
1951 GOTO(recalc, rc = 0);
1954 if (d->lod_lsfs_age < ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE) {
1955 struct obd_statfs sfs;
1957 spin_lock(&d->lod_lsfs_lock);
1958 if (d->lod_lsfs_age > ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE)
1959 GOTO(unlock, rc = 0);
1961 d->lod_lsfs_age = ktime_get_seconds();
1962 spin_unlock(&d->lod_lsfs_lock);
1963 rc = dt_statfs(env, d->lod_child, &sfs);
1966 "%s: failed to get OSD statfs: rc = %d\n",
1967 lod2obd(d)->obd_name, rc);
1970 /* udpate local OSD cached statfs data */
1971 spin_lock(&d->lod_lsfs_lock);
1972 d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
1973 d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
1975 lod_dom_stripesize_recalc(d);
1977 spin_unlock(&d->lod_lsfs_lock);
1980 return d->lod_dom_stripesize_cur_kb << 10;
1983 int lod_dom_stripesize_choose(const struct lu_env *env, struct lod_device *d,
1984 struct lov_comp_md_v1 *comp_v1,
1985 struct lov_comp_md_entry_v1 *dom_ent,
1988 struct lov_comp_md_entry_v1 *ent;
1989 struct lu_extent *dom_ext, *ext;
1990 struct lov_user_md_v1 *lum;
1991 __u32 max_stripe_size;
1994 bool dom_next_entry = false;
1996 dom_ext = &dom_ent->lcme_extent;
1997 dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1998 max_stripe_size = lod_dom_stripesize_limit(env, d);
2000 /* Check stripe size againts current per-MDT limit */
2001 if (stripe_size <= max_stripe_size)
2004 lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
2005 CDEBUG(D_LAYOUT, "overwrite DoM component size %u with MDT limit %u\n",
2006 stripe_size, max_stripe_size);
2007 lum->lmm_stripe_size = cpu_to_le32(max_stripe_size);
2009 /* In common case the DoM stripe is first entry in a mirror and
2010 * can be deleted only if it is not single entry in layout or
2011 * mirror, otherwise error should be returned.
2013 for_each_comp_entry_v1(comp_v1, ent) {
2017 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
2021 ext = &ent->lcme_extent;
2022 if (ext->e_start != dom_ext->e_end)
2025 /* Found next component after the DoM one with the same
2026 * mirror_id and adjust its start with DoM component end.
2028 * NOTE: we are considering here that there can be only one
2029 * DoM component in a file, all replicas are located on OSTs
2030 * always and don't need adjustment since use own layouts.
2032 ext->e_start = cpu_to_le64(max_stripe_size);
2033 dom_next_entry = true;
2037 if (max_stripe_size == 0) {
2038 /* DoM component size is zero due to server setting, remove
2039 * it from the layout but only if next component exists in
2040 * the same mirror. That must be checked prior calling the
2041 * lod_erase_dom_stripe().
2043 if (!dom_next_entry)
2046 rc = lod_erase_dom_stripe(comp_v1, dom_ent);
2048 /* Update DoM extent end finally */
2049 dom_ext->e_end = cpu_to_le64(max_stripe_size);
2056 * Verify LOV striping.
2058 * \param[in] d LOD device
2059 * \param[in] buf buffer with LOV EA to verify
2060 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
2061 * 1 - from disk, do not allow
2062 * \param[in] start extent start for composite layout
2064 * \retval 0 if the striping is valid
2065 * \retval -EINVAL if striping is invalid
2067 int lod_verify_striping(const struct lu_env *env, struct lod_device *d,
2068 struct lod_object *lo, const struct lu_buf *buf,
2071 struct lov_user_md_v1 *lum;
2072 struct lov_comp_md_v1 *comp_v1;
2073 struct lov_comp_md_entry_v1 *ent;
2074 struct lu_extent *ext;
2077 __u32 stripe_size = 0;
2078 __u16 prev_mid = -1, mirror_id = -1;
2084 if (buf->lb_len < sizeof(lum->lmm_magic)) {
2085 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
2091 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
2092 /* treat foreign LOV EA/object case first
2093 * XXX is it expected to try setting again a foreign?
2094 * XXX should we care about different current vs new layouts ?
2096 if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
2097 struct lov_foreign_md *lfm = buf->lb_buf;
2099 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
2101 "buf len %zu < min lov_foreign_md size (%zu)\n",
2102 buf->lb_len, offsetof(typeof(*lfm),
2107 if (lov_foreign_size_le(lfm) > buf->lb_len) {
2109 "buf len %zu < this lov_foreign_md size (%zu)\n",
2110 buf->lb_len, lov_foreign_size_le(lfm));
2113 /* Don't do anything with foreign layouts */
2117 /* normal LOV/layout cases */
2119 if (buf->lb_len < sizeof(*lum)) {
2120 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
2126 case LOV_USER_MAGIC_FOREIGN:
2128 case LOV_USER_MAGIC_V1:
2129 case LOV_USER_MAGIC_V3:
2130 case LOV_USER_MAGIC_SPECIFIC:
2131 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) &
2133 /* DoM must use composite layout */
2134 CDEBUG(D_LAYOUT, "DoM without composite layout\n");
2137 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
2138 case LOV_USER_MAGIC_COMP_V1:
2139 case LOV_USER_MAGIC_SEL:
2142 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
2143 le32_to_cpu(lum->lmm_magic));
2147 /* magic == LOV_USER_MAGIC_COMP_V1 */
2148 comp_v1 = buf->lb_buf;
2149 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
2150 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
2151 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
2157 if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
2158 CDEBUG(D_LAYOUT, "entry count is zero\n");
2162 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
2163 lo->ldo_comp_cnt > 0) {
2164 /* could be called from lustre.lov.add */
2165 __u32 cnt = lo->ldo_comp_cnt;
2167 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
2168 prev_end = ext->e_end;
2173 for_each_comp_entry_v1(comp_v1, ent) {
2174 ext = &ent->lcme_extent;
2176 if (le64_to_cpu(ext->e_start) > le64_to_cpu(ext->e_end) ||
2177 le64_to_cpu(ext->e_start) & (LOV_MIN_STRIPE_SIZE - 1) ||
2178 (le64_to_cpu(ext->e_end) != LUSTRE_EOF &&
2179 le64_to_cpu(ext->e_end) & (LOV_MIN_STRIPE_SIZE - 1))) {
2180 CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
2181 le64_to_cpu(ext->e_start),
2182 le64_to_cpu(ext->e_end));
2187 /* lcme_id contains valid value */
2188 if (le32_to_cpu(ent->lcme_id) == 0 ||
2189 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
2190 CDEBUG(D_LAYOUT, "invalid id %u\n",
2191 le32_to_cpu(ent->lcme_id));
2195 if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
2196 mirror_id = mirror_id_of(
2197 le32_to_cpu(ent->lcme_id));
2199 /* first component must start with 0 */
2200 if (mirror_id != prev_mid &&
2201 le64_to_cpu(ext->e_start) != 0) {
2203 "invalid start:%llu, expect:0\n",
2204 le64_to_cpu(ext->e_start));
2208 prev_mid = mirror_id;
2212 if (le64_to_cpu(ext->e_start) == 0) {
2217 /* the next must be adjacent with the previous one */
2218 if (le64_to_cpu(ext->e_start) != prev_end) {
2220 "invalid start actual:%llu, expect:%llu\n",
2221 le64_to_cpu(ext->e_start), prev_end);
2225 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
2226 tmp.lb_len = le32_to_cpu(ent->lcme_size);
2229 if (le32_to_cpu(lum->lmm_magic) == LOV_MAGIC_FOREIGN) {
2230 struct lov_foreign_md *lfm;
2231 struct lov_hsm_md *lhm;
2236 * Currently when the foreign layout is used as a basic
2237 * layout component, it only supports HSM foreign types:
2238 * LU_FOREIGN_TYPE_{POSIX, S3, PCCRW, PCCRO}.
2240 lfm = (struct lov_foreign_md *)lum;
2241 ftype = le32_to_cpu(lfm->lfm_type);
2242 if (!lov_hsm_type_supported(ftype)) {
2244 "Foreign type %#x is not HSM\n", ftype);
2248 /* Current HSM component must cover [0, EOF]. */
2249 if (le64_to_cpu(ext->e_start) > 0) {
2250 CDEBUG(D_LAYOUT, "Invalid HSM component with %llu extent start\n",
2251 le64_to_cpu(ext->e_start));
2254 if (le64_to_cpu(ext->e_end) != LUSTRE_EOF) {
2255 CDEBUG(D_LAYOUT, "Invalid HSM component with %llu extent end\n",
2256 le64_to_cpu(ext->e_end));
2260 lhm = (struct lov_hsm_md *)lfm;
2261 if (le32_to_cpu(lhm->lhm_length) !=
2262 sizeof(struct lov_hsm_base)) {
2264 "Invalid HSM component size %u != %u\n",
2265 le32_to_cpu(ent->lcme_size), hsmsize);
2269 hsmsize = lov_foreign_size_le(lhm);
2270 if (le32_to_cpu(ent->lcme_size) < hsmsize) {
2272 "Invalid HSM component size %u != %u\n",
2273 le32_to_cpu(ent->lcme_size), hsmsize);
2276 if (le32_to_cpu(lhm->lhm_flags) & ~HSM_FLAGS_MASK ||
2277 !(le32_to_cpu(lhm->lhm_flags) & HSM_FLAGS_MASK)) {
2279 "Invalid HSM component flags %#x\n",
2280 le32_to_cpu(lhm->lhm_flags));
2286 /* Check DoM entry is always the first one */
2287 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) &
2289 /* DoM component must be the first in a mirror */
2290 if (le64_to_cpu(ext->e_start) > 0) {
2291 CDEBUG(D_LAYOUT, "invalid DoM component "
2292 "with %llu extent start\n",
2293 le64_to_cpu(ext->e_start));
2296 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2297 /* There is just one stripe on MDT and it must
2298 * cover whole component size. */
2299 if (stripe_size != le64_to_cpu(ext->e_end)) {
2300 CDEBUG(D_LAYOUT, "invalid DoM layout "
2301 "stripe size %u != %llu "
2302 "(component size)\n",
2303 stripe_size, prev_end);
2306 /* Check and adjust stripe size by per-MDT limit */
2307 rc = lod_dom_stripesize_choose(env, d, comp_v1, ent,
2309 /* DoM entry was removed, re-check layout from start */
2310 if (rc == -ERESTART)
2315 if (le16_to_cpu(lum->lmm_stripe_count) == 1)
2316 lum->lmm_stripe_count = 0;
2317 /* Any stripe count is forbidden on DoM component */
2318 if (lum->lmm_stripe_count > 0) {
2320 "invalid DoM layout stripe count %u, must be 0\n",
2321 le16_to_cpu(lum->lmm_stripe_count));
2325 /* Any pool is forbidden on DoM component */
2326 if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
2327 struct lov_user_md_v3 *v3 = (void *)lum;
2329 if (v3->lmm_pool_name[0] != '\0') {
2331 "DoM component cannot have pool assigned\n");
2337 prev_end = le64_to_cpu(ext->e_end);
2339 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
2343 if (prev_end == LUSTRE_EOF || ext->e_start == prev_end)
2346 /* extent end must be aligned with the stripe_size */
2347 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2348 if (stripe_size && prev_end % stripe_size) {
2349 CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
2350 "stripe_sz: %u, [%llu, %llu)\n",
2351 stripe_size, ext->e_start, prev_end);
2356 /* make sure that the mirror_count is telling the truth */
2357 if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
2364 * set the default stripe size, if unset.
2366 * \param[in,out] val number of bytes per OST stripe
2368 * The minimum stripe size is 64KB to ensure that a single stripe is an
2369 * even multiple of a client PAGE_SIZE (IA64, PPC, etc). Otherwise, it
2370 * is difficult to split dirty pages across OSCs during writes.
2372 void lod_fix_desc_stripe_size(__u64 *val)
2374 if (*val < LOV_MIN_STRIPE_SIZE) {
2376 LCONSOLE_INFO("Increasing default stripe size to "
2377 "minimum value %u\n",
2378 LOV_DESC_STRIPE_SIZE_DEFAULT);
2379 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
2380 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
2381 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
2382 LCONSOLE_WARN("Changing default stripe size to %llu (a "
2383 "multiple of %u)\n",
2384 *val, LOV_MIN_STRIPE_SIZE);
2389 * set the filesystem default number of stripes, if unset.
2391 * \param[in,out] val number of stripes
2393 * A value of "0" means "use the system-wide default stripe count", which
2394 * has either been inherited by now, or falls back to 1 stripe per file.
2395 * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
2396 * and is a valid value, so is left unchanged here.
2398 void lod_fix_desc_stripe_count(__u32 *val)
2405 * set the filesystem default layout pattern
2407 * \param[in,out] val LOV_PATTERN_* layout
2409 * A value of "0" means "use the system-wide default layout type", which
2410 * has either been inherited by now, or falls back to plain RAID0 striping.
2412 void lod_fix_desc_pattern(__u32 *val)
2414 /* from lov_setstripe */
2415 if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
2416 LCONSOLE_WARN("lod: Unknown stripe pattern: %#x\n", *val);
2421 void lod_fix_lmv_desc_pattern(__u32 *val)
2423 if ((*val) && !lmv_is_known_hash_type(*val)) {
2424 LCONSOLE_WARN("lod: Unknown md stripe pattern: %#x\n", *val);
2429 void lod_fix_desc_qos_maxage(__u32 *val)
2431 /* fix qos_maxage */
2433 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2437 * Used to fix insane default striping.
2439 * \param[in] desc striping description
2441 void lod_fix_desc(struct lov_desc *desc)
2443 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2444 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2445 lod_fix_desc_pattern(&desc->ld_pattern);
2446 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2449 static void lod_fix_lmv_desc(struct lmv_desc *desc)
2451 desc->ld_active_tgt_count = 0;
2452 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2453 lod_fix_lmv_desc_pattern(&desc->ld_pattern);
2454 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2458 * Initialize the structures used to store pools and default striping.
2460 * \param[in] lod LOD device
2461 * \param[in] lcfg configuration structure storing default striping.
2463 * \retval 0 if initialization succeeds
2464 * \retval negative error number on failure
2466 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2468 struct obd_device *obd;
2469 struct lov_desc *desc;
2473 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2474 LASSERT(obd != NULL);
2475 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2477 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2478 CERROR("LOD setup requires a descriptor\n");
2482 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2484 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2485 CERROR("descriptor size wrong: %d > %d\n",
2486 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2490 if (desc->ld_magic != LOV_DESC_MAGIC) {
2491 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2492 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2493 obd->obd_name, desc);
2494 lustre_swab_lov_desc(desc);
2496 CERROR("%s: Bad lov desc magic: %#x\n",
2497 obd->obd_name, desc->ld_magic);
2504 desc->ld_active_tgt_count = 0;
2505 lod->lod_ost_descs.ltd_lov_desc = *desc;
2507 /* NB: config doesn't contain lmv_desc, alter it via sysfs. */
2508 lod_fix_lmv_desc(&lod->lod_mdt_descs.ltd_lmv_desc);
2510 lod->lod_sp_me = LUSTRE_SP_CLI;
2512 /* Set up OST pool environment */
2513 lod->lod_pool_count = 0;
2514 rc = lod_pool_hash_init(&lod->lod_pools_hash_body);
2518 INIT_LIST_HEAD(&lod->lod_pool_list);
2519 lod->lod_pool_count = 0;
2520 rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_tgt_pool, 0);
2524 rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool, 0);
2526 GOTO(out_mdt_pool, rc);
2528 rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_tgt_pool, 0);
2530 GOTO(out_mdt_rr_pool, rc);
2532 rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
2534 GOTO(out_ost_pool, rc);
2539 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2541 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2543 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
2545 lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2551 * Release the structures describing the pools.
2553 * \param[in] lod LOD device from which we release the structures
2557 int lod_pools_fini(struct lod_device *lod)
2559 struct obd_device *obd = lod2obd(lod);
2560 struct pool_desc *pool, *tmp;
2563 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2564 /* free pool structs */
2565 CDEBUG(D_INFO, "delete pool %p\n", pool);
2566 /* In the function below, .hs_keycmp resolves to
2567 * pool_hashkey_keycmp() */
2568 lod_pool_del(obd, pool->pool_name);
2571 lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2572 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool);
2573 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2574 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2575 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);