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 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
749 stripe_count = lod_comp_entry_stripe_count(lo, comp_idx, is_dir);
750 if (stripe_count == 0 && !is_dir &&
751 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
752 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
753 /* Try again if all active targets are disconnected.
754 * It is possible when MDS does failover. */
755 if (!lod->lod_ost_active_count &&
761 if (!is_dir && lo->ldo_is_composite)
762 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
764 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
767 /* generate ost_idx of this component stripe */
768 for (i = 0; i < stripe_count; i++) {
769 struct dt_object *object;
770 __u32 ost_idx = (__u32)-1UL;
771 int type = LU_SEQ_RANGE_OST;
773 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
774 object = lod_comp->llc_stripe[i];
775 /* instantiated component */
776 info->lti_fid = *lu_object_fid(&object->do_lu);
778 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
780 if (cfs_fail_val == 0)
781 cfs_fail_val = info->lti_fid.f_oid;
783 info->lti_fid.f_oid = cfs_fail_val;
786 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
789 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
790 objs[i].l_ost_gen = cpu_to_le32(0);
791 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
794 rc = lod_fld_lookup(env, lod, &info->lti_fid,
797 CERROR("%s: Can not locate "DFID": rc = %d\n",
798 lod2obd(lod)->obd_name,
799 PFID(&info->lti_fid), rc);
802 } else if (lod_comp->llc_ostlist.op_array &&
803 lod_comp->llc_ostlist.op_count) {
804 /* user specified ost list */
805 ost_idx = lod_comp->llc_ostlist.op_array[i];
808 * with un-instantiated or with no specified ost list
809 * component, its l_ost_idx does not matter.
811 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
814 if (lmm_size != NULL)
815 *lmm_size = lov_mds_md_size(stripe_count, magic);
820 * Generate on-disk lov_mds_md structure based on the information in
821 * the lod_object->ldo_comp_entries.
823 * \param[in] env execution environment for this thread
824 * \param[in] lo LOD object
825 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
826 * \param[in|out] lmm_size buffer size/lmm size
827 * \param[in] is_dir generate lov ea for dir or file? For dir case,
828 * the stripe info is from the default stripe
829 * template, which is collected in lod_ah_init(),
830 * either from parent object or root object; for
831 * file case, it's from the @lo object
833 * \retval 0 if on disk structure is created successfully
834 * \retval negative error number on failure
836 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
837 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
839 struct lov_comp_md_entry_v1 *lcme;
840 struct lov_comp_md_v1 *lcm;
841 struct lod_layout_component *comp_entries;
842 __u16 comp_cnt, mirror_cnt;
843 bool is_composite, is_foreign = false;
844 int i, rc = 0, offset;
848 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
849 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
850 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
852 lo->ldo_def_striping->lds_def_striping_is_composite;
854 comp_cnt = lo->ldo_comp_cnt;
855 mirror_cnt = lo->ldo_mirror_count;
856 comp_entries = lo->ldo_comp_entries;
857 is_composite = lo->ldo_is_composite;
858 is_foreign = lo->ldo_is_foreign;
861 LASSERT(lmm_size != NULL);
864 struct lov_foreign_md *lfm;
866 lfm = (struct lov_foreign_md *)lmm;
867 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
868 /* need to store little-endian */
869 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
870 __swab32s(&lfm->lfm_magic);
871 __swab32s(&lfm->lfm_length);
872 __swab32s(&lfm->lfm_type);
873 __swab32s(&lfm->lfm_flags);
875 *lmm_size = lo->ldo_foreign_lov_size;
879 LASSERT(comp_cnt != 0 && comp_entries != NULL);
882 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
886 lcm = (struct lov_comp_md_v1 *)lmm;
887 memset(lcm, 0, sizeof(*lcm));
889 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
890 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
891 lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
892 lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
894 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
895 LASSERT(offset % sizeof(__u64) == 0);
897 for (i = 0; i < comp_cnt; i++) {
898 struct lod_layout_component *lod_comp;
899 struct lov_mds_md *sub_md;
902 lod_comp = &comp_entries[i];
903 lcme = &lcm->lcm_entries[i];
905 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
906 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
908 /* component could be un-inistantiated */
909 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
910 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
911 lcme->lcme_timestamp =
912 cpu_to_le64(lod_comp->llc_timestamp);
913 if (lod_comp->llc_flags & LCME_FL_EXTENSION && !is_dir)
914 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
916 lcme->lcme_extent.e_start =
917 cpu_to_le64(lod_comp->llc_extent.e_start);
918 lcme->lcme_extent.e_end =
919 cpu_to_le64(lod_comp->llc_extent.e_end);
920 lcme->lcme_offset = cpu_to_le32(offset);
922 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
923 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
926 lcme->lcme_size = cpu_to_le32(size);
928 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
929 "offset:%d lmm_size:%d\n", offset, *lmm_size);
931 lcm->lcm_size = cpu_to_le32(offset);
932 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
934 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
945 * Fill lti_ea_store buffer in the environment with a value for the given
946 * EA. The buffer is reallocated if the value doesn't fit.
948 * \param[in,out] env execution environment for this thread
949 * .lti_ea_store buffer is filled with EA's value
950 * \param[in] lo LOD object
951 * \param[in] name name of the EA
953 * \retval > 0 if EA is fetched successfully
954 * \retval 0 if EA is empty
955 * \retval negative error number on failure
957 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
960 struct lod_thread_info *info = lod_env_info(env);
961 struct dt_object *next = dt_object_child(&lo->ldo_obj);
967 if (unlikely(info->lti_ea_store == NULL)) {
968 /* just to enter in allocation block below */
972 info->lti_buf.lb_buf = info->lti_ea_store;
973 info->lti_buf.lb_len = info->lti_ea_store_size;
974 rc = dt_xattr_get(env, next, &info->lti_buf, name);
977 /* if object is not striped or inaccessible */
978 if (rc == -ENODATA || rc == -ENOENT)
982 /* EA doesn't fit, reallocate new buffer */
983 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
984 if (rc == -ENODATA || rc == -ENOENT)
990 rc = lod_ea_store_resize(info, rc);
1000 * Verify the target index is present in the current configuration.
1002 * \param[in] md LOD device where the target table is stored
1003 * \param[in] idx target's index
1005 * \retval 0 if the index is present
1006 * \retval -EINVAL if not
1008 int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1010 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1011 !test_bit(idx, md->lod_ost_bitmap))) {
1012 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1013 md->lod_ost_descs.ltd_tgts_size);
1017 if (unlikely(OST_TGT(md, idx) == NULL)) {
1018 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1019 lod2obd(md)->obd_name, idx);
1023 if (unlikely(OST_TGT(md, idx)->ltd_tgt == NULL)) {
1024 CERROR("%s: invalid lod device, for idx: %d\n",
1025 lod2obd(md)->obd_name , idx);
1033 * Instantiate objects for stripes.
1035 * Allocate and initialize LU-objects representing the stripes. The number
1036 * of the stripes (ldo_stripe_count) must be initialized already. The caller
1037 * must ensure nobody else is calling the function on the object at the same
1038 * time. FLDB service must be running to be able to map a FID to the targets
1039 * and find appropriate device representing that target.
1041 * \param[in] env execution environment for this thread
1042 * \param[in,out] lo LOD object
1043 * \param[in] objs an array of IDs to creates the objects from
1044 * \param[in] comp_idx index of ldo_comp_entries
1046 * \retval 0 if the objects are instantiated successfully
1047 * \retval negative error number on failure
1049 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1050 struct lov_ost_data_v1 *objs, int comp_idx)
1052 struct lod_layout_component *lod_comp;
1053 struct lod_thread_info *info = lod_env_info(env);
1054 struct lod_device *md;
1055 struct lu_object *o, *n;
1056 struct lu_device *nd;
1057 struct dt_object **stripe = NULL;
1058 __u32 *ost_indices = NULL;
1064 LASSERT(lo != NULL);
1065 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1067 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1068 lod_comp = &lo->ldo_comp_entries[comp_idx];
1070 LASSERT(lod_comp->llc_stripe == NULL);
1071 LASSERT(lod_comp->llc_stripe_count > 0);
1072 LASSERT(lod_comp->llc_stripe_size > 0);
1074 stripe_len = lod_comp->llc_stripe_count;
1075 OBD_ALLOC_PTR_ARRAY(stripe, stripe_len);
1078 OBD_ALLOC_PTR_ARRAY(ost_indices, stripe_len);
1080 GOTO(out, rc = -ENOMEM);
1082 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1083 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1086 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1087 idx = le32_to_cpu(objs[i].l_ost_idx);
1088 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1091 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1092 PFID(&info->lti_fid));
1093 lod_getref(&md->lod_ost_descs);
1095 rc = validate_lod_and_idx(md, idx);
1096 if (unlikely(rc != 0)) {
1097 lod_putref(md, &md->lod_ost_descs);
1101 nd = &OST_TGT(md, idx)->ltd_tgt->dd_lu_dev;
1102 lod_putref(md, &md->lod_ost_descs);
1104 /* In the function below, .hs_keycmp resolves to
1105 * u_obj_hop_keycmp() */
1106 /* coverity[overrun-buffer-val] */
1107 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1109 GOTO(out, rc = PTR_ERR(o));
1111 n = lu_object_locate(o->lo_header, nd->ld_type);
1114 stripe[i] = container_of(n, struct dt_object, do_lu);
1115 ost_indices[i] = idx;
1120 for (i = 0; i < stripe_len; i++)
1121 if (stripe[i] != NULL)
1122 dt_object_put(env, stripe[i]);
1124 OBD_FREE_PTR_ARRAY(stripe, stripe_len);
1125 lod_comp->llc_stripe_count = 0;
1127 OBD_FREE_PTR_ARRAY(ost_indices, stripe_len);
1129 lod_comp->llc_stripe = stripe;
1130 lod_comp->llc_ost_indices = ost_indices;
1131 lod_comp->llc_stripes_allocated = stripe_len;
1138 * Instantiate objects for striping.
1140 * Parse striping information in \a buf and instantiate the objects
1141 * representing the stripes.
1143 * \param[in] env execution environment for this thread
1144 * \param[in] lo LOD object
1145 * \param[in] buf buffer storing LOV EA to parse
1146 * \param[in] lvf verify flags when parsing the layout
1148 * \retval 0 if parsing and objects creation succeed
1149 * \retval negative error number on failure
1151 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1152 const struct lu_buf *buf, enum layout_verify_flags lvf)
1154 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1155 struct lov_mds_md_v1 *lmm;
1156 struct lov_comp_md_v1 *comp_v1 = NULL;
1157 struct lov_foreign_md *foreign = NULL;
1158 struct lov_ost_data_v1 *objs;
1159 __u32 magic, pattern;
1160 __u16 mirror_cnt = 0;
1163 __u16 mirror_id = MIRROR_ID_NEG;
1165 int stale_mirrors = 0;
1169 LASSERT(buf->lb_buf);
1170 LASSERT(buf->lb_len);
1171 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1173 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1174 magic = le32_to_cpu(lmm->lmm_magic);
1176 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1177 magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN &&
1178 magic != LOV_MAGIC_SEL)
1179 GOTO(out, rc = -EINVAL);
1181 lod_striping_free_nolock(env, lo);
1183 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1184 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1185 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1187 GOTO(out, rc = -EINVAL);
1188 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1189 lo->ldo_is_composite = 1;
1190 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1192 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1195 lo->ldo_flr_state = LCM_FL_NONE;
1196 } else if (magic == LOV_MAGIC_FOREIGN) {
1199 foreign = (struct lov_foreign_md *)buf->lb_buf;
1200 length = offsetof(typeof(*foreign), lfm_value);
1201 if (buf->lb_len < length ||
1202 buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1204 "buf len %zu too small for lov_foreign_md\n",
1206 GOTO(out, rc = -EINVAL);
1209 /* just cache foreign LOV EA raw */
1210 rc = lod_alloc_foreign_lov(lo, length);
1213 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1217 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1218 lo->ldo_is_composite = 0;
1221 rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1225 for (i = 0; i < comp_cnt; i++) {
1226 struct lod_layout_component *lod_comp;
1227 struct lu_extent *ext;
1230 lod_comp = &lo->ldo_comp_entries[i];
1231 if (lo->ldo_is_composite) {
1232 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1233 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1235 ext = &comp_v1->lcm_entries[i].lcme_extent;
1236 lod_comp->llc_extent.e_start =
1237 le64_to_cpu(ext->e_start);
1238 if (lod_comp->llc_extent.e_start &
1239 (LOV_MIN_STRIPE_SIZE - 1)) {
1241 "extent start %llu is not a multiple of min size %u\n",
1242 lod_comp->llc_extent.e_start,
1243 LOV_MIN_STRIPE_SIZE);
1244 GOTO(out, rc = -EINVAL);
1247 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1248 if (lod_comp->llc_extent.e_end != LUSTRE_EOF &&
1249 lod_comp->llc_extent.e_end &
1250 (LOV_MIN_STRIPE_SIZE - 1)) {
1252 "extent end %llu is not a multiple of min size %u\n",
1253 lod_comp->llc_extent.e_end,
1254 LOV_MIN_STRIPE_SIZE);
1255 GOTO(out, rc = -EINVAL);
1258 lod_comp->llc_flags =
1259 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1261 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1262 lod_comp->llc_timestamp = le64_to_cpu(
1263 comp_v1->lcm_entries[i].lcme_timestamp);
1265 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1266 if (lod_comp->llc_id == LCME_ID_INVAL)
1267 GOTO(out, rc = -EINVAL);
1269 if (lvf & LVF_ALL_STALE) {
1270 if (mirror_id_of(lod_comp->llc_id) ==
1272 /* remaining comps in the mirror */
1273 stale |= lod_comp->llc_flags &
1277 * new mirror, check last mirror's
1284 mirror_id_of(lod_comp->llc_id);
1286 /* the first comp of the new mirror */
1287 stale = lod_comp->llc_flags &
1292 if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
1293 comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
1294 CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
1295 lod2obd(d)->obd_name,
1296 lod_comp->llc_flags, lod_comp->llc_id, i,
1297 PFID(lod_object_fid(lo)),
1298 le32_to_cpu(comp_v1->lcm_magic));
1301 lod_comp_set_init(lod_comp);
1304 pattern = le32_to_cpu(lmm->lmm_pattern);
1305 if (!lov_pattern_supported(lov_pattern(pattern)))
1306 GOTO(out, rc = -EINVAL);
1308 lod_comp->llc_pattern = pattern;
1309 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1310 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1311 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1313 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1314 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1316 lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1317 objs = &v3->lmm_objects[0];
1319 lod_set_pool(&lod_comp->llc_pool, NULL);
1320 objs = &lmm->lmm_objects[0];
1324 * If uninstantiated template component has valid l_ost_idx,
1325 * then user has specified ost list for this component.
1327 if (!lod_comp_inited(lod_comp)) {
1330 if (objs[0].l_ost_idx != (__u32)-1UL) {
1333 stripe_count = lod_comp_entry_stripe_count(
1335 if (stripe_count == 0 &&
1336 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1337 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1338 GOTO(out, rc = -E2BIG);
1340 * load the user specified ost list, when this
1341 * component is instantiated later, it will be
1342 * used in lod_alloc_ost_list().
1344 lod_comp->llc_ostlist.op_count = stripe_count;
1345 lod_comp->llc_ostlist.op_size =
1346 stripe_count * sizeof(__u32);
1347 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1348 lod_comp->llc_ostlist.op_size);
1349 if (!lod_comp->llc_ostlist.op_array)
1350 GOTO(out, rc = -ENOMEM);
1352 for (j = 0; j < stripe_count; j++)
1353 lod_comp->llc_ostlist.op_array[j] =
1354 le32_to_cpu(objs[j].l_ost_idx);
1357 * this component OST objects starts from the
1358 * first ost_idx, lod_alloc_ost_list() will
1361 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1364 * for uninstantiated component,
1365 * lmm_layout_gen stores default stripe offset.
1367 lod_comp->llc_stripe_offset =
1368 lmm->lmm_layout_gen;
1372 /* skip un-instantiated component object initialization */
1373 if (!lod_comp_inited(lod_comp))
1376 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1377 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1378 rc = lod_initialize_objects(env, lo, objs, i);
1384 if (lo->ldo_is_composite && (lvf & LVF_ALL_STALE)) {
1385 /* check the last mirror stale-ness */
1389 if (mirror_cnt == stale_mirrors) {
1391 CERROR("%s: can not set all stale mirrors for "
1393 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
1399 rc = lod_fill_mirrors(lo);
1405 lod_striping_free_nolock(env, lo);
1410 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1411 * is already cached.
1413 * \param[in] lo LOD object
1415 * \retval True if the striping is cached, otherwise
1418 static bool lod_striping_loaded(struct lod_object *lo)
1420 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1421 lo->ldo_comp_cached)
1424 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1425 if (lo->ldo_dir_stripe_loaded)
1428 /* Never load LMV stripe for slaves of striped dir */
1429 if (lo->ldo_dir_slave_stripe)
1437 * A generic function to initialize the stripe objects.
1439 * A protected version of lod_striping_load_locked() - load the striping
1440 * information from storage, parse that and instantiate LU objects to
1441 * represent the stripes. The LOD object \a lo supplies a pointer to the
1442 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1443 * return an array of references to the newly instantiated objects.
1445 * \param[in] env execution environment for this thread
1446 * \param[in,out] lo LOD object, where striping is stored and
1447 * which gets an array of references
1449 * \retval 0 if parsing and object creation succeed
1450 * \retval negative error number on failure
1452 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1454 struct lod_thread_info *info = lod_env_info(env);
1455 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1456 struct lu_buf *buf = &info->lti_buf;
1461 if (!dt_object_exists(next))
1464 if (lod_striping_loaded(lo))
1467 mutex_lock(&lo->ldo_layout_mutex);
1468 if (lod_striping_loaded(lo))
1469 GOTO(unlock, rc = 0);
1471 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1472 rc = lod_get_lov_ea(env, lo);
1477 * there is LOV EA (striping information) in this object
1478 * let's parse it and create in-core objects for the stripes
1480 buf->lb_buf = info->lti_ea_store;
1481 buf->lb_len = info->lti_ea_store_size;
1482 rc = lod_parse_striping(env, lo, buf, 0);
1484 lo->ldo_comp_cached = 1;
1485 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1486 rc = lod_get_lmv_ea(env, lo);
1487 if (rc > sizeof(struct lmv_foreign_md)) {
1488 struct lmv_foreign_md *lfm = info->lti_ea_store;
1490 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
1491 lo->ldo_foreign_lmv = info->lti_ea_store;
1492 lo->ldo_foreign_lmv_size =
1493 info->lti_ea_store_size;
1494 info->lti_ea_store = NULL;
1495 info->lti_ea_store_size = 0;
1497 lo->ldo_dir_stripe_loaded = 1;
1498 lo->ldo_dir_is_foreign = 1;
1499 GOTO(unlock, rc = 0);
1503 if (rc < (int)sizeof(struct lmv_mds_md_v1)) {
1504 /* Let's set stripe_loaded to avoid further
1505 * stripe loading especially for non-stripe directory,
1506 * which can hurt performance. (See LU-9840)
1509 lo->ldo_dir_stripe_loaded = 1;
1510 GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1512 buf->lb_buf = info->lti_ea_store;
1513 buf->lb_len = info->lti_ea_store_size;
1514 if (rc == sizeof(struct lmv_mds_md_v1)) {
1515 rc = lod_load_lmv_shards(env, lo, buf, true);
1516 if (buf->lb_buf != info->lti_ea_store) {
1517 OBD_FREE_LARGE(info->lti_ea_store,
1518 info->lti_ea_store_size);
1519 info->lti_ea_store = buf->lb_buf;
1520 info->lti_ea_store_size = buf->lb_len;
1528 * there is LMV EA (striping information) in this object
1529 * let's parse it and create in-core objects for the stripes
1531 rc = lod_parse_dir_striping(env, lo, buf);
1533 lo->ldo_dir_stripe_loaded = 1;
1537 mutex_unlock(&lo->ldo_layout_mutex);
1542 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1543 const struct lu_buf *buf, enum layout_verify_flags lvf)
1549 mutex_lock(&lo->ldo_layout_mutex);
1550 rc = lod_parse_striping(env, lo, buf, lvf);
1551 mutex_unlock(&lo->ldo_layout_mutex);
1557 * Verify lov_user_md_v1/v3 striping.
1559 * Check the validity of all fields including the magic, stripe size,
1560 * stripe count, stripe offset and that the pool is present. Also check
1561 * that each target index points to an existing target. The additional
1562 * \a is_from_disk turns additional checks. In some cases zero fields
1563 * are allowed (like pattern=0).
1565 * \param[in] d LOD device
1566 * \param[in] buf buffer with LOV EA to verify
1567 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1568 * 1 - from disk, do not allow
1570 * \retval 0 if the striping is valid
1571 * \retval -EINVAL if striping is invalid
1573 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1576 struct lov_user_md_v1 *lum;
1577 struct lov_user_md_v3 *lum3;
1578 struct pool_desc *pool = NULL;
1582 __u16 stripe_offset;
1589 if (buf->lb_len < sizeof(*lum)) {
1590 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1592 GOTO(out, rc = -EINVAL);
1595 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1596 if (magic != LOV_USER_MAGIC_V1 &&
1597 magic != LOV_USER_MAGIC_V3 &&
1598 magic != LOV_USER_MAGIC_SPECIFIC) {
1599 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1600 le32_to_cpu(lum->lmm_magic));
1601 GOTO(out, rc = -EINVAL);
1604 /* the user uses "0" for default stripe pattern normally. */
1605 if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1606 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1608 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1609 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1610 le32_to_cpu(lum->lmm_pattern));
1611 GOTO(out, rc = -EINVAL);
1614 /* a released lum comes from creating orphan on hsm release,
1615 * doesn't make sense to verify it. */
1616 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1619 /* 64kB is the largest common page size we see (ia64), and matches the
1621 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1622 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1623 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1624 stripe_size, LOV_MIN_STRIPE_SIZE);
1625 GOTO(out, rc = -EINVAL);
1628 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1629 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1630 lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1631 /* if offset is not within valid range [0, osts_size) */
1632 if (stripe_offset >= d->lod_ost_descs.ltd_tgts_size) {
1633 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1634 stripe_offset, d->lod_ost_descs.ltd_tgts_size);
1635 GOTO(out, rc = -EINVAL);
1638 /* if lmm_stripe_offset is *not* in bitmap */
1639 if (!test_bit(stripe_offset, d->lod_ost_bitmap)) {
1640 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1642 GOTO(out, rc = -EINVAL);
1646 if (magic == LOV_USER_MAGIC_V1)
1647 lum_size = offsetof(struct lov_user_md_v1,
1649 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1650 lum_size = offsetof(struct lov_user_md_v3,
1653 GOTO(out, rc = -EINVAL);
1655 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1656 if (buf->lb_len < lum_size) {
1657 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1658 "magic %#x and stripe_count %u\n",
1659 buf->lb_len, lum_size, magic, stripe_count);
1660 GOTO(out, rc = -EINVAL);
1663 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1667 /* In the function below, .hs_keycmp resolves to
1668 * pool_hashkey_keycmp() */
1669 /* coverity[overrun-buffer-val] */
1670 pool = lod_find_pool(d, lum3->lmm_pool_name);
1674 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1675 rc = lod_check_index_in_pool(stripe_offset, pool);
1677 GOTO(out, rc = -EINVAL);
1680 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1681 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1682 stripe_count, pool_tgt_count(pool));
1683 GOTO(out, rc = -EINVAL);
1688 lod_pool_putref(pool);
1694 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1696 LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1697 LOV_USER_MAGIC_COMP_V1 ||
1698 (le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1699 LOV_USER_MAGIC_SEL, "Wrong magic %x\n",
1700 le32_to_cpu(comp->lcm_magic));
1701 LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1702 "bad index %d, max = %d\n",
1703 i, le16_to_cpu(comp->lcm_entry_count));
1705 return &comp->lcm_entries[i];
1708 #define for_each_comp_entry_v1(comp, entry) \
1709 for (entry = comp_entry_v1(comp, 0); \
1710 entry <= comp_entry_v1(comp, \
1711 le16_to_cpu(comp->lcm_entry_count) - 1); \
1714 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
1715 struct lov_comp_md_entry_v1 *dom_ent)
1717 struct lov_comp_md_entry_v1 *ent;
1719 __u32 dom_off, dom_size, comp_size, off;
1721 unsigned int size, shift;
1723 entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1724 LASSERT(entries > 0);
1725 comp_v1->lcm_entry_count = cpu_to_le16(entries);
1727 comp_size = le32_to_cpu(comp_v1->lcm_size);
1728 dom_off = le32_to_cpu(dom_ent->lcme_offset);
1729 dom_size = le32_to_cpu(dom_ent->lcme_size);
1731 /* all entries offsets are shifted by entry size at least */
1732 shift = sizeof(*dom_ent);
1733 for_each_comp_entry_v1(comp_v1, ent) {
1734 off = le32_to_cpu(ent->lcme_offset);
1735 if (off == dom_off) {
1736 /* Entry deletion creates two holes in layout data:
1737 * - hole in entries array
1738 * - hole in layout data at dom_off with dom_size
1740 * First memmove is one entry shift from next entry
1741 * start with size up to dom_off in blob
1744 src = (void *)(ent + 1);
1745 size = (unsigned long)((void *)comp_v1 + dom_off - src);
1746 memmove(dst, src, size);
1747 /* take 'off' from just moved entry */
1748 off = le32_to_cpu(ent->lcme_offset);
1749 /* second memmove is blob tail after 'off' up to
1752 dst = (void *)comp_v1 + dom_off - sizeof(*ent);
1753 src = (void *)comp_v1 + off;
1754 size = (unsigned long)(comp_size - off);
1755 memmove(dst, src, size);
1756 /* all entries offsets after DoM entry are shifted by
1757 * dom_size additionally
1761 ent->lcme_offset = cpu_to_le32(off - shift);
1763 comp_v1->lcm_size = cpu_to_le32(comp_size - shift);
1765 /* notify a caller to re-check entry */
1769 void lod_dom_stripesize_recalc(struct lod_device *d)
1771 __u64 threshold_mb = d->lod_dom_threshold_free_mb;
1772 __u32 max_size = d->lod_dom_stripesize_max_kb;
1773 __u32 def_size = d->lod_dom_stripesize_cur_kb;
1775 /* use maximum allowed value if free space is above threshold */
1776 if (d->lod_lsfs_free_mb >= threshold_mb) {
1777 def_size = max_size;
1778 } else if (!d->lod_lsfs_free_mb || max_size <= LOD_DOM_MIN_SIZE_KB) {
1781 /* recalc threshold like it would be with def_size as max */
1782 threshold_mb = mult_frac(threshold_mb, def_size, max_size);
1783 if (d->lod_lsfs_free_mb < threshold_mb)
1784 def_size = rounddown(def_size / 2, LOD_DOM_MIN_SIZE_KB);
1785 else if (d->lod_lsfs_free_mb > threshold_mb * 2)
1786 def_size = max_t(unsigned int, def_size * 2,
1787 LOD_DOM_MIN_SIZE_KB);
1790 if (d->lod_dom_stripesize_cur_kb != def_size) {
1791 CDEBUG(D_LAYOUT, "Change default DOM stripe size %d->%d\n",
1792 d->lod_dom_stripesize_cur_kb, def_size);
1793 d->lod_dom_stripesize_cur_kb = def_size;
1797 static __u32 lod_dom_stripesize_limit(const struct lu_env *env,
1798 struct lod_device *d)
1802 /* set bfree as fraction of total space */
1803 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_SPOOF)) {
1804 spin_lock(&d->lod_lsfs_lock);
1805 d->lod_lsfs_free_mb = mult_frac(d->lod_lsfs_total_mb,
1806 min_t(int, cfs_fail_val, 100), 100);
1807 GOTO(recalc, rc = 0);
1810 if (d->lod_lsfs_age < ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE) {
1811 struct obd_statfs sfs;
1813 spin_lock(&d->lod_lsfs_lock);
1814 if (d->lod_lsfs_age > ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE)
1815 GOTO(unlock, rc = 0);
1817 d->lod_lsfs_age = ktime_get_seconds();
1818 spin_unlock(&d->lod_lsfs_lock);
1819 rc = dt_statfs(env, d->lod_child, &sfs);
1822 "%s: failed to get OSD statfs: rc = %d\n",
1823 lod2obd(d)->obd_name, rc);
1826 /* udpate local OSD cached statfs data */
1827 spin_lock(&d->lod_lsfs_lock);
1828 d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
1829 d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
1831 lod_dom_stripesize_recalc(d);
1833 spin_unlock(&d->lod_lsfs_lock);
1836 return d->lod_dom_stripesize_cur_kb << 10;
1839 int lod_dom_stripesize_choose(const struct lu_env *env, struct lod_device *d,
1840 struct lov_comp_md_v1 *comp_v1,
1841 struct lov_comp_md_entry_v1 *dom_ent,
1844 struct lov_comp_md_entry_v1 *ent;
1845 struct lu_extent *dom_ext, *ext;
1846 struct lov_user_md_v1 *lum;
1847 __u32 max_stripe_size;
1850 bool dom_next_entry = false;
1852 dom_ext = &dom_ent->lcme_extent;
1853 dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1854 max_stripe_size = lod_dom_stripesize_limit(env, d);
1856 /* Check stripe size againts current per-MDT limit */
1857 if (stripe_size <= max_stripe_size)
1860 lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1861 CDEBUG(D_LAYOUT, "overwrite DoM component size %u with MDT limit %u\n",
1862 stripe_size, max_stripe_size);
1863 lum->lmm_stripe_size = cpu_to_le32(max_stripe_size);
1865 /* In common case the DoM stripe is first entry in a mirror and
1866 * can be deleted only if it is not single entry in layout or
1867 * mirror, otherwise error should be returned.
1869 for_each_comp_entry_v1(comp_v1, ent) {
1873 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1877 ext = &ent->lcme_extent;
1878 if (ext->e_start != dom_ext->e_end)
1881 /* Found next component after the DoM one with the same
1882 * mirror_id and adjust its start with DoM component end.
1884 * NOTE: we are considering here that there can be only one
1885 * DoM component in a file, all replicas are located on OSTs
1886 * always and don't need adjustment since use own layouts.
1888 ext->e_start = cpu_to_le64(max_stripe_size);
1889 dom_next_entry = true;
1893 if (max_stripe_size == 0) {
1894 /* DoM component size is zero due to server setting, remove
1895 * it from the layout but only if next component exists in
1896 * the same mirror. That must be checked prior calling the
1897 * lod_erase_dom_stripe().
1899 if (!dom_next_entry)
1902 rc = lod_erase_dom_stripe(comp_v1, dom_ent);
1904 /* Update DoM extent end finally */
1905 dom_ext->e_end = cpu_to_le64(max_stripe_size);
1912 * Verify LOV striping.
1914 * \param[in] d LOD device
1915 * \param[in] buf buffer with LOV EA to verify
1916 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1917 * 1 - from disk, do not allow
1918 * \param[in] start extent start for composite layout
1920 * \retval 0 if the striping is valid
1921 * \retval -EINVAL if striping is invalid
1923 int lod_verify_striping(const struct lu_env *env, struct lod_device *d,
1924 struct lod_object *lo, const struct lu_buf *buf,
1927 struct lov_user_md_v1 *lum;
1928 struct lov_comp_md_v1 *comp_v1;
1929 struct lov_comp_md_entry_v1 *ent;
1930 struct lu_extent *ext;
1933 __u32 stripe_size = 0;
1934 __u16 prev_mid = -1, mirror_id = -1;
1940 if (buf->lb_len < sizeof(lum->lmm_magic)) {
1941 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
1947 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1948 /* treat foreign LOV EA/object case first
1949 * XXX is it expected to try setting again a foreign?
1950 * XXX should we care about different current vs new layouts ?
1952 if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
1953 struct lov_foreign_md *lfm = buf->lb_buf;
1955 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
1957 "buf len %zu < min lov_foreign_md size (%zu)\n",
1958 buf->lb_len, offsetof(typeof(*lfm),
1963 if (foreign_size_le(lfm) > buf->lb_len) {
1965 "buf len %zu < this lov_foreign_md size (%zu)\n",
1966 buf->lb_len, foreign_size_le(lfm));
1969 /* Don't do anything with foreign layouts */
1973 /* normal LOV/layout cases */
1975 if (buf->lb_len < sizeof(*lum)) {
1976 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1982 case LOV_USER_MAGIC_FOREIGN:
1984 case LOV_USER_MAGIC_V1:
1985 case LOV_USER_MAGIC_V3:
1986 case LOV_USER_MAGIC_SPECIFIC:
1987 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1988 case LOV_USER_MAGIC_COMP_V1:
1989 case LOV_USER_MAGIC_SEL:
1992 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1993 le32_to_cpu(lum->lmm_magic));
1997 /* magic == LOV_USER_MAGIC_COMP_V1 */
1998 comp_v1 = buf->lb_buf;
1999 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
2000 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
2001 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
2007 if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
2008 CDEBUG(D_LAYOUT, "entry count is zero\n");
2012 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
2013 lo->ldo_comp_cnt > 0) {
2014 /* could be called from lustre.lov.add */
2015 __u32 cnt = lo->ldo_comp_cnt;
2017 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
2018 prev_end = ext->e_end;
2023 for_each_comp_entry_v1(comp_v1, ent) {
2024 ext = &ent->lcme_extent;
2026 if (le64_to_cpu(ext->e_start) > le64_to_cpu(ext->e_end) ||
2027 le64_to_cpu(ext->e_start) & (LOV_MIN_STRIPE_SIZE - 1) ||
2028 (le64_to_cpu(ext->e_end) != LUSTRE_EOF &&
2029 le64_to_cpu(ext->e_end) & (LOV_MIN_STRIPE_SIZE - 1))) {
2030 CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
2031 le64_to_cpu(ext->e_start),
2032 le64_to_cpu(ext->e_end));
2037 /* lcme_id contains valid value */
2038 if (le32_to_cpu(ent->lcme_id) == 0 ||
2039 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
2040 CDEBUG(D_LAYOUT, "invalid id %u\n",
2041 le32_to_cpu(ent->lcme_id));
2045 if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
2046 mirror_id = mirror_id_of(
2047 le32_to_cpu(ent->lcme_id));
2049 /* first component must start with 0 */
2050 if (mirror_id != prev_mid &&
2051 le64_to_cpu(ext->e_start) != 0) {
2053 "invalid start:%llu, expect:0\n",
2054 le64_to_cpu(ext->e_start));
2058 prev_mid = mirror_id;
2062 if (le64_to_cpu(ext->e_start) == 0) {
2067 /* the next must be adjacent with the previous one */
2068 if (le64_to_cpu(ext->e_start) != prev_end) {
2070 "invalid start actual:%llu, expect:%llu\n",
2071 le64_to_cpu(ext->e_start), prev_end);
2075 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
2076 tmp.lb_len = le32_to_cpu(ent->lcme_size);
2078 /* Check DoM entry is always the first one */
2080 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
2082 /* DoM component must be the first in a mirror */
2083 if (le64_to_cpu(ext->e_start) > 0) {
2084 CDEBUG(D_LAYOUT, "invalid DoM component "
2085 "with %llu extent start\n",
2086 le64_to_cpu(ext->e_start));
2089 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2090 /* There is just one stripe on MDT and it must
2091 * cover whole component size. */
2092 if (stripe_size != le64_to_cpu(ext->e_end)) {
2093 CDEBUG(D_LAYOUT, "invalid DoM layout "
2094 "stripe size %u != %llu "
2095 "(component size)\n",
2096 stripe_size, prev_end);
2099 /* Check and adjust stripe size by per-MDT limit */
2100 rc = lod_dom_stripesize_choose(env, d, comp_v1, ent,
2102 /* DoM entry was removed, re-check layout from start */
2103 if (rc == -ERESTART)
2108 if (le16_to_cpu(lum->lmm_stripe_count) == 1)
2109 lum->lmm_stripe_count = 0;
2110 /* Any stripe count is forbidden on DoM component */
2111 if (lum->lmm_stripe_count > 0) {
2113 "invalid DoM layout stripe count %u, must be 0\n",
2114 le16_to_cpu(lum->lmm_stripe_count));
2118 /* Any pool is forbidden on DoM component */
2119 if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
2120 struct lov_user_md_v3 *v3 = (void *)lum;
2122 if (v3->lmm_pool_name[0] != '\0') {
2124 "DoM component cannot have pool assigned\n");
2130 prev_end = le64_to_cpu(ext->e_end);
2132 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
2136 if (prev_end == LUSTRE_EOF || ext->e_start == prev_end)
2139 /* extent end must be aligned with the stripe_size */
2140 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2141 if (stripe_size && prev_end % stripe_size) {
2142 CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
2143 "stripe_sz: %u, [%llu, %llu)\n",
2144 stripe_size, ext->e_start, prev_end);
2149 /* make sure that the mirror_count is telling the truth */
2150 if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
2157 * set the default stripe size, if unset.
2159 * \param[in,out] val number of bytes per OST stripe
2161 * The minimum stripe size is 64KB to ensure that a single stripe is an
2162 * even multiple of a client PAGE_SIZE (IA64, PPC, etc). Otherwise, it
2163 * is difficult to split dirty pages across OSCs during writes.
2165 void lod_fix_desc_stripe_size(__u64 *val)
2167 if (*val < LOV_MIN_STRIPE_SIZE) {
2169 LCONSOLE_INFO("Increasing default stripe size to "
2170 "minimum value %u\n",
2171 LOV_DESC_STRIPE_SIZE_DEFAULT);
2172 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
2173 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
2174 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
2175 LCONSOLE_WARN("Changing default stripe size to %llu (a "
2176 "multiple of %u)\n",
2177 *val, LOV_MIN_STRIPE_SIZE);
2182 * set the filesystem default number of stripes, if unset.
2184 * \param[in,out] val number of stripes
2186 * A value of "0" means "use the system-wide default stripe count", which
2187 * has either been inherited by now, or falls back to 1 stripe per file.
2188 * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
2189 * and is a valid value, so is left unchanged here.
2191 void lod_fix_desc_stripe_count(__u32 *val)
2198 * set the filesystem default layout pattern
2200 * \param[in,out] val LOV_PATTERN_* layout
2202 * A value of "0" means "use the system-wide default layout type", which
2203 * has either been inherited by now, or falls back to plain RAID0 striping.
2205 void lod_fix_desc_pattern(__u32 *val)
2207 /* from lov_setstripe */
2208 if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
2209 LCONSOLE_WARN("lod: Unknown stripe pattern: %#x\n", *val);
2214 void lod_fix_lmv_desc_pattern(__u32 *val)
2216 if ((*val) && !lmv_is_known_hash_type(*val)) {
2217 LCONSOLE_WARN("lod: Unknown md stripe pattern: %#x\n", *val);
2222 void lod_fix_desc_qos_maxage(__u32 *val)
2224 /* fix qos_maxage */
2226 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2230 * Used to fix insane default striping.
2232 * \param[in] desc striping description
2234 void lod_fix_desc(struct lov_desc *desc)
2236 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2237 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2238 lod_fix_desc_pattern(&desc->ld_pattern);
2239 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2242 static void lod_fix_lmv_desc(struct lmv_desc *desc)
2244 desc->ld_active_tgt_count = 0;
2245 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2246 lod_fix_lmv_desc_pattern(&desc->ld_pattern);
2247 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2251 * Initialize the structures used to store pools and default striping.
2253 * \param[in] lod LOD device
2254 * \param[in] lcfg configuration structure storing default striping.
2256 * \retval 0 if initialization succeeds
2257 * \retval negative error number on failure
2259 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2261 struct obd_device *obd;
2262 struct lov_desc *desc;
2266 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2267 LASSERT(obd != NULL);
2268 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2270 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2271 CERROR("LOD setup requires a descriptor\n");
2275 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2277 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2278 CERROR("descriptor size wrong: %d > %d\n",
2279 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2283 if (desc->ld_magic != LOV_DESC_MAGIC) {
2284 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2285 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2286 obd->obd_name, desc);
2287 lustre_swab_lov_desc(desc);
2289 CERROR("%s: Bad lov desc magic: %#x\n",
2290 obd->obd_name, desc->ld_magic);
2297 desc->ld_active_tgt_count = 0;
2298 lod->lod_ost_descs.ltd_lov_desc = *desc;
2300 /* NB: config doesn't contain lmv_desc, alter it via sysfs. */
2301 lod_fix_lmv_desc(&lod->lod_mdt_descs.ltd_lmv_desc);
2303 lod->lod_sp_me = LUSTRE_SP_CLI;
2305 /* Set up OST pool environment */
2306 lod->lod_pool_count = 0;
2307 rc = lod_pool_hash_init(&lod->lod_pools_hash_body);
2311 INIT_LIST_HEAD(&lod->lod_pool_list);
2312 lod->lod_pool_count = 0;
2313 rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_tgt_pool, 0);
2317 rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool, 0);
2319 GOTO(out_mdt_pool, rc);
2321 rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_tgt_pool, 0);
2323 GOTO(out_mdt_rr_pool, rc);
2325 rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
2327 GOTO(out_ost_pool, rc);
2332 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2334 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2336 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
2338 lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2344 * Release the structures describing the pools.
2346 * \param[in] lod LOD device from which we release the structures
2350 int lod_pools_fini(struct lod_device *lod)
2352 struct obd_device *obd = lod2obd(lod);
2353 struct pool_desc *pool, *tmp;
2356 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2357 /* free pool structs */
2358 CDEBUG(D_INFO, "delete pool %p\n", pool);
2359 /* In the function below, .hs_keycmp resolves to
2360 * pool_hashkey_keycmp() */
2361 /* coverity[overrun-buffer-val] */
2362 lod_pool_del(obd, pool->pool_name);
2365 lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2366 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool);
2367 lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2368 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2369 lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);