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;
79 struct list_head kill;
81 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
84 INIT_LIST_HEAD(&kill);
86 ltd_foreach_tgt_safe(ltd, tgt_desc, tmp) {
88 if (!tgt_desc->ltd_reap)
91 list_add(&tgt_desc->ltd_kill, &kill);
92 /*FIXME: only support ost pool for now */
93 if (ltd == &lod->lod_ost_descs)
94 lod_ost_pool_remove(<d->ltd_tgt_pool,
96 ltd_del_tgt(ltd, tgt_desc);
99 mutex_unlock(<d->ltd_mutex);
100 up_read(<d->ltd_rw_sem);
102 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
105 list_del(&tgt_desc->ltd_kill);
106 rc = obd_disconnect(tgt_desc->ltd_exp);
108 CERROR("%s: failed to disconnect %s: rc = %d\n",
109 lod2obd(lod)->obd_name,
110 obd_uuid2str(&tgt_desc->ltd_uuid), rc);
111 OBD_FREE_PTR(tgt_desc);
114 mutex_unlock(<d->ltd_mutex);
115 up_read(<d->ltd_rw_sem);
120 * Connect LOD to a new OSP and add it to the target table.
122 * Connect to the OSP device passed, initialize all the internal
123 * structures related to the device and add it to the target table.
125 * \param[in] env execution environment for this thread
126 * \param[in] lod LOD device to be connected to the new OSP
127 * \param[in] osp name of OSP device name to be added
128 * \param[in] index index of the new target
129 * \param[in] gen target's generation number
130 * \param[in] tgt_index OSP's group
131 * \param[in] type type of device (mdc or osc)
132 * \param[in] active state of OSP: 0 - inactive, 1 - active
134 * \retval 0 if added successfully
135 * \retval negative error number on failure
137 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
138 char *osp, unsigned index, unsigned gen, int tgt_index,
139 char *type, int active)
141 struct obd_connect_data *data = NULL;
142 struct obd_export *exp = NULL;
143 struct obd_device *obd;
144 struct lu_device *lu_dev;
145 struct dt_device *dt_dev;
147 struct lod_tgt_desc *tgt_desc;
148 struct lod_tgt_descs *ltd;
149 struct lustre_cfg *lcfg;
150 struct obd_uuid obd_uuid;
152 bool connected = false;
155 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
158 CERROR("request to add OBD %s with invalid generation: %d\n",
163 obd_str2uuid(&obd_uuid, osp);
165 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
166 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
168 CERROR("can't find %s device\n", osp);
172 LASSERT(obd->obd_lu_dev != NULL);
173 LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
175 lu_dev = obd->obd_lu_dev;
176 dt_dev = lu2dt_dev(lu_dev);
180 GOTO(out_cleanup, rc = -ENOMEM);
182 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
183 data->ocd_version = LUSTRE_VERSION_CODE;
184 data->ocd_index = index;
186 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
188 data->ocd_connect_flags |= OBD_CONNECT_AT |
191 #ifdef HAVE_LRU_RESIZE_SUPPORT
192 OBD_CONNECT_LRU_RESIZE |
195 OBD_CONNECT_REQPORTAL |
196 OBD_CONNECT_SKIP_ORPHAN |
198 OBD_CONNECT_LVB_TYPE |
199 OBD_CONNECT_VERSION |
200 OBD_CONNECT_PINGLESS |
202 OBD_CONNECT_BULK_MBITS;
204 data->ocd_group = tgt_index;
205 ltd = &lod->lod_ost_descs;
207 struct obd_import *imp = obd->u.cli.cl_import;
210 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
211 data->ocd_connect_flags |= OBD_CONNECT_ACL |
213 OBD_CONNECT_MDS_MDS |
218 OBD_CONNECT_BULK_MBITS;
219 spin_lock(&imp->imp_lock);
220 imp->imp_server_timeout = 1;
221 spin_unlock(&imp->imp_lock);
222 imp->imp_client->cli_request_portal = OUT_PORTAL;
223 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
225 ltd = &lod->lod_mdt_descs;
228 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
231 CERROR("%s: cannot connect to next dev %s (%d)\n",
232 obd->obd_name, osp, rc);
233 GOTO(out_cleanup, rc);
237 /* Allocate ost descriptor and fill it */
238 OBD_ALLOC_PTR(tgt_desc);
240 GOTO(out_cleanup, rc = -ENOMEM);
242 tgt_desc->ltd_tgt = dt_dev;
243 tgt_desc->ltd_exp = exp;
244 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
245 tgt_desc->ltd_gen = gen;
246 tgt_desc->ltd_index = index;
247 tgt_desc->ltd_active = active;
249 down_write(<d->ltd_rw_sem);
250 mutex_lock(<d->ltd_mutex);
251 rc = ltd_add_tgt(ltd, tgt_desc);
255 rc = lu_qos_add_tgt(<d->ltd_qos, tgt_desc);
257 GOTO(out_del_tgt, rc);
260 /* pool is not supported for MDS stack yet */
261 rc = lod_ost_pool_add(<d->ltd_tgt_pool, index,
264 CERROR("%s: can't set up pool, failed with %d\n",
266 GOTO(out_del_tgt, rc);
270 mutex_unlock(<d->ltd_mutex);
271 up_write(<d->ltd_rw_sem);
273 if (lod->lod_recovery_completed)
274 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
276 if (!for_ost && lod->lod_initialized) {
277 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
279 CERROR("%s: cannot start llog on %s:rc = %d\n",
280 lod2obd(lod)->obd_name, osp, rc);
285 rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
287 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
288 "index = %u, rc = %d\n", osp, type, index, rc);
289 GOTO(out_fini_llog, rc);
293 lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
294 tgt_desc->ltd_recovery_thread);
296 down_write(<d->ltd_rw_sem);
297 mutex_lock(<d->ltd_mutex);
298 if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
299 struct ptlrpc_thread *thread;
301 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
302 OBD_FREE_PTR(thread);
304 lod_ost_pool_remove(<d->ltd_tgt_pool, index);
306 ltd_del_tgt(ltd, tgt_desc);
308 mutex_unlock(<d->ltd_mutex);
309 up_write(<d->ltd_rw_sem);
310 OBD_FREE_PTR(tgt_desc);
312 /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
313 * objects from the MDT stack. See LU-7184. */
314 lcfg = &lod_env_info(env)->lti_lustre_cfg;
315 memset(lcfg, 0, sizeof(*lcfg));
316 lcfg->lcfg_version = LUSTRE_CFG_VERSION;
317 lcfg->lcfg_command = LCFG_CLEANUP;
318 lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
327 * Schedule target removal from the target table.
329 * Mark the device as dead. The device is not removed here because it may
330 * still be in use. The device will be removed in lod_putref() when the
331 * last reference is released.
333 * \param[in] env execution environment for this thread
334 * \param[in] lod LOD device the target table belongs to
335 * \param[in] ltd target table
336 * \param[in] tgt target
338 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
339 struct lod_tgt_descs *ltd, struct lu_tgt_desc *tgt)
341 lfsck_del_target(env, lod->lod_child, tgt->ltd_tgt, tgt->ltd_index,
344 if (ltd->ltd_is_mdt && tgt->ltd_recovery_thread)
345 OBD_FREE_PTR(tgt->ltd_recovery_thread);
347 if (!tgt->ltd_reap) {
349 ltd->ltd_death_row++;
354 * Schedule removal of all the targets from the given target table.
356 * See more details in the description for __lod_del_device()
358 * \param[in] env execution environment for this thread
359 * \param[in] lod LOD device the target table belongs to
360 * \param[in] ltd target table
364 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
365 struct lod_tgt_descs *ltd)
367 struct lu_tgt_desc *tgt;
369 if (ltd->ltd_tgts_size <= 0)
373 mutex_lock(<d->ltd_mutex);
374 ltd_foreach_tgt(ltd, tgt)
375 __lod_del_device(env, lod, ltd, tgt);
376 mutex_unlock(<d->ltd_mutex);
377 lod_putref(lod, ltd);
379 lu_tgt_descs_fini(ltd);
385 * Remove device by name.
387 * Remove a device identified by \a osp from the target table. Given
388 * the device can be in use, the real deletion happens in lod_putref().
390 * \param[in] env execution environment for this thread
391 * \param[in] lod LOD device to be connected to the new OSP
392 * \param[in] ltd target table
393 * \param[in] osp name of OSP device to be removed
394 * \param[in] idx index of the target
395 * \param[in] gen generation number, not used currently
397 * \retval 0 if the device was scheduled for removal
398 * \retval -EINVAL if no device was found
400 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
401 struct lod_tgt_descs *ltd, char *osp, unsigned int idx,
404 struct obd_device *obd;
405 struct lu_tgt_desc *tgt;
406 struct obd_uuid uuid;
411 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
413 obd_str2uuid(&uuid, osp);
415 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
416 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
418 CERROR("can't find %s device\n", osp);
423 CERROR("%s: request to remove OBD %s with invalid generation %d"
424 "\n", obd->obd_name, osp, gen);
428 obd_str2uuid(&uuid, osp);
431 mutex_lock(<d->ltd_mutex);
432 tgt = LTD_TGT(ltd, idx);
433 /* check that the index is allocated in the bitmap */
434 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) || !tgt) {
435 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
436 GOTO(out, rc = -EINVAL);
439 /* check that the UUID matches */
440 if (!obd_uuid_equals(&uuid, &tgt->ltd_uuid)) {
441 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
442 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), idx, osp);
443 GOTO(out, rc = -EINVAL);
446 __lod_del_device(env, lod, ltd, tgt);
449 mutex_unlock(<d->ltd_mutex);
450 lod_putref(lod, ltd);
455 * Resize per-thread storage to hold specified size.
457 * A helper function to resize per-thread temporary storage. This storage
458 * is used to process LOV/LVM EAs and may be quite large. We do not want to
459 * allocate/release it every time, so instead we put it into the env and
460 * reallocate on demand. The memory is released when the correspondent thread
463 * \param[in] info LOD-specific storage in the environment
464 * \param[in] size new size to grow the buffer to
466 * \retval 0 on success, -ENOMEM if reallocation failed
468 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
470 __u32 round = size_roundup_power2(size);
472 if (info->lti_ea_store) {
473 LASSERT(info->lti_ea_store_size);
474 LASSERT(info->lti_ea_store_size < round);
475 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
476 info->lti_ea_store_size, round);
477 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
478 info->lti_ea_store = NULL;
479 info->lti_ea_store_size = 0;
482 OBD_ALLOC_LARGE(info->lti_ea_store, round);
483 if (info->lti_ea_store == NULL)
485 info->lti_ea_store_size = round;
490 static void lod_free_comp_buffer(struct lod_layout_component *entries,
491 __u16 count, __u32 bufsize)
493 struct lod_layout_component *entry;
496 for (i = 0; i < count; i++) {
498 if (entry->llc_pool != NULL)
499 lod_set_pool(&entry->llc_pool, NULL);
500 if (entry->llc_ostlist.op_array)
501 OBD_FREE(entry->llc_ostlist.op_array,
502 entry->llc_ostlist.op_size);
503 LASSERT(entry->llc_stripe == NULL);
504 LASSERT(entry->llc_stripes_allocated == 0);
508 OBD_FREE_LARGE(entries, bufsize);
511 void lod_free_def_comp_entries(struct lod_default_striping *lds)
513 lod_free_comp_buffer(lds->lds_def_comp_entries,
514 lds->lds_def_comp_size_cnt,
516 sizeof(*lds->lds_def_comp_entries) *
517 lds->lds_def_comp_size_cnt));
518 lds->lds_def_comp_entries = NULL;
519 lds->lds_def_comp_cnt = 0;
520 lds->lds_def_striping_is_composite = 0;
521 lds->lds_def_comp_size_cnt = 0;
525 * Resize per-thread storage to hold default striping component entries
527 * A helper function to resize per-thread temporary storage. This storage
528 * is used to hold default LOV/LVM EAs and may be quite large. We do not want
529 * to allocate/release it every time, so instead we put it into the env and
530 * reallocate it on demand. The memory is released when the correspondent
531 * thread is finished.
533 * \param[in,out] lds default striping
534 * \param[in] count new component count to grow the buffer to
536 * \retval 0 on success, -ENOMEM if reallocation failed
538 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
540 struct lod_layout_component *entries;
541 __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
543 __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
544 lds->lds_def_comp_size_cnt);
549 OBD_ALLOC_LARGE(entries, new);
553 if (lds->lds_def_comp_entries != NULL) {
554 CDEBUG(D_INFO, "default striping component size %d is not "
555 "enough, need %d\n", old, new);
556 lod_free_def_comp_entries(lds);
559 lds->lds_def_comp_entries = entries;
560 lds->lds_def_comp_size_cnt = count;
565 void lod_free_comp_entries(struct lod_object *lo)
567 if (lo->ldo_mirrors) {
568 OBD_FREE(lo->ldo_mirrors,
569 sizeof(*lo->ldo_mirrors) * lo->ldo_mirror_count);
570 lo->ldo_mirrors = NULL;
571 lo->ldo_mirror_count = 0;
573 lod_free_comp_buffer(lo->ldo_comp_entries,
575 sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
576 lo->ldo_comp_entries = NULL;
577 lo->ldo_comp_cnt = 0;
578 lo->ldo_is_composite = 0;
581 int lod_alloc_comp_entries(struct lod_object *lo,
582 int mirror_count, int comp_count)
584 LASSERT(comp_count != 0);
585 LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
587 if (mirror_count > 0) {
588 OBD_ALLOC(lo->ldo_mirrors,
589 sizeof(*lo->ldo_mirrors) * mirror_count);
590 if (!lo->ldo_mirrors)
593 lo->ldo_mirror_count = mirror_count;
596 OBD_ALLOC_LARGE(lo->ldo_comp_entries,
597 sizeof(*lo->ldo_comp_entries) * comp_count);
598 if (lo->ldo_comp_entries == NULL) {
599 OBD_FREE(lo->ldo_mirrors,
600 sizeof(*lo->ldo_mirrors) * mirror_count);
601 lo->ldo_mirror_count = 0;
605 lo->ldo_comp_cnt = comp_count;
609 int lod_fill_mirrors(struct lod_object *lo)
611 struct lod_layout_component *lod_comp;
613 __u16 mirror_id = 0xffff;
617 LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
619 if (!lo->ldo_is_composite)
622 lod_comp = &lo->ldo_comp_entries[0];
623 for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
624 int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
625 int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
627 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
628 lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
629 lo->ldo_mirrors[mirror_idx].lme_primary |= preferred;
630 lo->ldo_mirrors[mirror_idx].lme_end = i;
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_primary = preferred;
644 lo->ldo_mirrors[mirror_idx].lme_start = i;
645 lo->ldo_mirrors[mirror_idx].lme_end = i;
647 if (mirror_idx != lo->ldo_mirror_count - 1)
654 * Generate on-disk lov_mds_md structure for each layout component based on
655 * the information in lod_object->ldo_comp_entries[i].
657 * \param[in] env execution environment for this thread
658 * \param[in] lo LOD object
659 * \param[in] comp_idx index of ldo_comp_entries
660 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
661 * \param[in|out] lmm_size buffer size/lmm size
662 * \param[in] is_dir generate lov ea for dir or file? For dir case,
663 * the stripe info is from the default stripe
664 * template, which is collected in lod_ah_init(),
665 * either from parent object or root object; for
666 * file case, it's from the @lo object
668 * \retval 0 if on disk structure is created successfully
669 * \retval negative error number on failure
671 static int lod_gen_component_ea(const struct lu_env *env,
672 struct lod_object *lo, int comp_idx,
673 struct lov_mds_md *lmm, int *lmm_size,
676 struct lod_thread_info *info = lod_env_info(env);
677 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
678 struct lod_device *lod;
679 struct lov_ost_data_v1 *objs;
680 struct lod_layout_component *lod_comp;
689 &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
691 lod_comp = &lo->ldo_comp_entries[comp_idx];
693 magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
694 if (lod_comp->llc_pattern == 0) /* default striping */
695 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
697 lmm->lmm_magic = cpu_to_le32(magic);
698 lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
699 fid_to_lmm_oi(fid, &lmm->lmm_oi);
700 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
701 lmm->lmm_oi.oi.oi_id++;
702 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
704 lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
705 lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
707 * for dir and uninstantiated component, lmm_layout_gen stores
708 * default stripe offset.
710 lmm->lmm_layout_gen =
711 (is_dir || !lod_comp_inited(lod_comp)) ?
712 cpu_to_le16(lod_comp->llc_stripe_offset) :
713 cpu_to_le16(lod_comp->llc_layout_gen);
715 if (magic == LOV_MAGIC_V1) {
716 objs = &lmm->lmm_objects[0];
718 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
719 size_t cplen = strlcpy(v3->lmm_pool_name,
721 sizeof(v3->lmm_pool_name));
722 if (cplen >= sizeof(v3->lmm_pool_name))
724 objs = &v3->lmm_objects[0];
726 stripe_count = lod_comp_entry_stripe_count(lo, lod_comp, is_dir);
727 if (stripe_count == 0 && !is_dir &&
728 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
729 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
732 if (!is_dir && lo->ldo_is_composite)
733 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
735 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
738 /* generate ost_idx of this component stripe */
739 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
740 for (i = 0; i < stripe_count; i++) {
741 struct dt_object *object;
742 __u32 ost_idx = (__u32)-1UL;
743 int type = LU_SEQ_RANGE_OST;
745 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
746 object = lod_comp->llc_stripe[i];
747 /* instantiated component */
748 info->lti_fid = *lu_object_fid(&object->do_lu);
750 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
752 if (cfs_fail_val == 0)
753 cfs_fail_val = info->lti_fid.f_oid;
755 info->lti_fid.f_oid = cfs_fail_val;
758 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
761 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
762 objs[i].l_ost_gen = cpu_to_le32(0);
763 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
766 rc = lod_fld_lookup(env, lod, &info->lti_fid,
769 CERROR("%s: Can not locate "DFID": rc = %d\n",
770 lod2obd(lod)->obd_name,
771 PFID(&info->lti_fid), rc);
774 } else if (lod_comp->llc_ostlist.op_array &&
775 lod_comp->llc_ostlist.op_count) {
776 /* user specified ost list */
777 ost_idx = lod_comp->llc_ostlist.op_array[i];
780 * with un-instantiated or with no specified ost list
781 * component, its l_ost_idx does not matter.
783 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
786 if (lmm_size != NULL)
787 *lmm_size = lov_mds_md_size(stripe_count, magic);
792 * Generate on-disk lov_mds_md structure based on the information in
793 * the lod_object->ldo_comp_entries.
795 * \param[in] env execution environment for this thread
796 * \param[in] lo LOD object
797 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
798 * \param[in|out] lmm_size buffer size/lmm size
799 * \param[in] is_dir generate lov ea for dir or file? For dir case,
800 * the stripe info is from the default stripe
801 * template, which is collected in lod_ah_init(),
802 * either from parent object or root object; for
803 * file case, it's from the @lo object
805 * \retval 0 if on disk structure is created successfully
806 * \retval negative error number on failure
808 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
809 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
811 struct lov_comp_md_entry_v1 *lcme;
812 struct lov_comp_md_v1 *lcm;
813 struct lod_layout_component *comp_entries;
814 __u16 comp_cnt, mirror_cnt;
815 bool is_composite, is_foreign = false;
816 int i, rc = 0, offset;
820 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
821 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
822 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
824 lo->ldo_def_striping->lds_def_striping_is_composite;
826 comp_cnt = lo->ldo_comp_cnt;
827 mirror_cnt = lo->ldo_mirror_count;
828 comp_entries = lo->ldo_comp_entries;
829 is_composite = lo->ldo_is_composite;
830 is_foreign = lo->ldo_is_foreign;
833 LASSERT(lmm_size != NULL);
836 struct lov_foreign_md *lfm;
838 lfm = (struct lov_foreign_md *)lmm;
839 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
840 /* need to store little-endian */
841 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
842 __swab32s(&lfm->lfm_magic);
843 __swab32s(&lfm->lfm_length);
844 __swab32s(&lfm->lfm_type);
845 __swab32s(&lfm->lfm_flags);
847 *lmm_size = lo->ldo_foreign_lov_size;
851 LASSERT(comp_cnt != 0 && comp_entries != NULL);
854 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
858 lcm = (struct lov_comp_md_v1 *)lmm;
859 memset(lcm, 0, sizeof(*lcm));
861 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
862 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
863 lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
864 lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
866 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
867 LASSERT(offset % sizeof(__u64) == 0);
869 for (i = 0; i < comp_cnt; i++) {
870 struct lod_layout_component *lod_comp;
871 struct lov_mds_md *sub_md;
874 lod_comp = &comp_entries[i];
875 lcme = &lcm->lcm_entries[i];
877 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
878 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
880 /* component could be un-inistantiated */
881 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
882 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
883 lcme->lcme_timestamp =
884 cpu_to_le64(lod_comp->llc_timestamp);
885 if (lod_comp->llc_flags & LCME_FL_EXTENSION && !is_dir)
886 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
888 lcme->lcme_extent.e_start =
889 cpu_to_le64(lod_comp->llc_extent.e_start);
890 lcme->lcme_extent.e_end =
891 cpu_to_le64(lod_comp->llc_extent.e_end);
892 lcme->lcme_offset = cpu_to_le32(offset);
894 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
895 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
898 lcme->lcme_size = cpu_to_le32(size);
900 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
901 "offset:%d lmm_size:%d\n", offset, *lmm_size);
903 lcm->lcm_size = cpu_to_le32(offset);
904 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
906 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
917 * Fill lti_ea_store buffer in the environment with a value for the given
918 * EA. The buffer is reallocated if the value doesn't fit.
920 * \param[in,out] env execution environment for this thread
921 * .lti_ea_store buffer is filled with EA's value
922 * \param[in] lo LOD object
923 * \param[in] name name of the EA
925 * \retval > 0 if EA is fetched successfully
926 * \retval 0 if EA is empty
927 * \retval negative error number on failure
929 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
932 struct lod_thread_info *info = lod_env_info(env);
933 struct dt_object *next = dt_object_child(&lo->ldo_obj);
939 if (unlikely(info->lti_ea_store == NULL)) {
940 /* just to enter in allocation block below */
944 info->lti_buf.lb_buf = info->lti_ea_store;
945 info->lti_buf.lb_len = info->lti_ea_store_size;
946 rc = dt_xattr_get(env, next, &info->lti_buf, name);
949 /* if object is not striped or inaccessible */
950 if (rc == -ENODATA || rc == -ENOENT)
954 /* EA doesn't fit, reallocate new buffer */
955 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
956 if (rc == -ENODATA || rc == -ENOENT)
962 rc = lod_ea_store_resize(info, rc);
972 * Verify the target index is present in the current configuration.
974 * \param[in] md LOD device where the target table is stored
975 * \param[in] idx target's index
977 * \retval 0 if the index is present
978 * \retval -EINVAL if not
980 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
982 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
983 !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
984 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
985 md->lod_ost_descs.ltd_tgts_size);
989 if (unlikely(OST_TGT(md, idx) == NULL)) {
990 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
991 lod2obd(md)->obd_name, idx);
995 if (unlikely(OST_TGT(md, idx)->ltd_tgt == NULL)) {
996 CERROR("%s: invalid lod device, for idx: %d\n",
997 lod2obd(md)->obd_name , idx);
1005 * Instantiate objects for stripes.
1007 * Allocate and initialize LU-objects representing the stripes. The number
1008 * of the stripes (ldo_stripe_count) must be initialized already. The caller
1009 * must ensure nobody else is calling the function on the object at the same
1010 * time. FLDB service must be running to be able to map a FID to the targets
1011 * and find appropriate device representing that target.
1013 * \param[in] env execution environment for this thread
1014 * \param[in,out] lo LOD object
1015 * \param[in] objs an array of IDs to creates the objects from
1016 * \param[in] comp_idx index of ldo_comp_entries
1018 * \retval 0 if the objects are instantiated successfully
1019 * \retval negative error number on failure
1021 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1022 struct lov_ost_data_v1 *objs, int comp_idx)
1024 struct lod_layout_component *lod_comp;
1025 struct lod_thread_info *info = lod_env_info(env);
1026 struct lod_device *md;
1027 struct lu_object *o, *n;
1028 struct lu_device *nd;
1029 struct dt_object **stripe = NULL;
1030 __u32 *ost_indices = NULL;
1036 LASSERT(lo != NULL);
1037 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1039 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1040 lod_comp = &lo->ldo_comp_entries[comp_idx];
1042 LASSERT(lod_comp->llc_stripe == NULL);
1043 LASSERT(lod_comp->llc_stripe_count > 0);
1044 LASSERT(lod_comp->llc_stripe_size > 0);
1046 stripe_len = lod_comp->llc_stripe_count;
1047 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1050 OBD_ALLOC(ost_indices, sizeof(*ost_indices) * stripe_len);
1052 GOTO(out, rc = -ENOMEM);
1054 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1055 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1058 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1059 idx = le32_to_cpu(objs[i].l_ost_idx);
1060 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1063 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1064 PFID(&info->lti_fid));
1065 lod_getref(&md->lod_ost_descs);
1067 rc = validate_lod_and_idx(md, idx);
1068 if (unlikely(rc != 0)) {
1069 lod_putref(md, &md->lod_ost_descs);
1073 nd = &OST_TGT(md, idx)->ltd_tgt->dd_lu_dev;
1074 lod_putref(md, &md->lod_ost_descs);
1076 /* In the function below, .hs_keycmp resolves to
1077 * u_obj_hop_keycmp() */
1078 /* coverity[overrun-buffer-val] */
1079 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1081 GOTO(out, rc = PTR_ERR(o));
1083 n = lu_object_locate(o->lo_header, nd->ld_type);
1086 stripe[i] = container_of(n, struct dt_object, do_lu);
1087 ost_indices[i] = idx;
1092 for (i = 0; i < stripe_len; i++)
1093 if (stripe[i] != NULL)
1094 dt_object_put(env, stripe[i]);
1096 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1097 lod_comp->llc_stripe_count = 0;
1099 OBD_FREE(ost_indices,
1100 sizeof(*ost_indices) * stripe_len);
1102 lod_comp->llc_stripe = stripe;
1103 lod_comp->llc_ost_indices = ost_indices;
1104 lod_comp->llc_stripes_allocated = stripe_len;
1111 * Instantiate objects for striping.
1113 * Parse striping information in \a buf and instantiate the objects
1114 * representing the stripes.
1116 * \param[in] env execution environment for this thread
1117 * \param[in] lo LOD object
1118 * \param[in] buf buffer storing LOV EA to parse
1120 * \retval 0 if parsing and objects creation succeed
1121 * \retval negative error number on failure
1123 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1124 const struct lu_buf *buf)
1126 struct lov_mds_md_v1 *lmm;
1127 struct lov_comp_md_v1 *comp_v1 = NULL;
1128 struct lov_foreign_md *foreign = NULL;
1129 struct lov_ost_data_v1 *objs;
1130 __u32 magic, pattern;
1131 __u16 mirror_cnt = 0;
1137 LASSERT(buf->lb_buf);
1138 LASSERT(buf->lb_len);
1139 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1141 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1142 magic = le32_to_cpu(lmm->lmm_magic);
1144 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1145 magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN &&
1146 magic != LOV_MAGIC_SEL)
1147 GOTO(out, rc = -EINVAL);
1149 if (lo->ldo_is_foreign)
1150 lod_free_foreign_lov(lo);
1152 lod_free_comp_entries(lo);
1154 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1155 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1156 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1158 GOTO(out, rc = -EINVAL);
1159 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1160 lo->ldo_is_composite = 1;
1161 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1163 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1164 } else if (magic == LOV_MAGIC_FOREIGN) {
1167 foreign = (struct lov_foreign_md *)buf->lb_buf;
1168 length = offsetof(typeof(*foreign), lfm_value);
1169 if (buf->lb_len < length ||
1170 buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1172 "buf len %zu too small for lov_foreign_md\n",
1174 GOTO(out, rc = -EINVAL);
1177 /* just cache foreign LOV EA raw */
1178 rc = lod_alloc_foreign_lov(lo, length);
1181 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1185 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1186 lo->ldo_is_composite = 0;
1189 rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1193 for (i = 0; i < comp_cnt; i++) {
1194 struct lod_layout_component *lod_comp;
1195 struct lu_extent *ext;
1198 lod_comp = &lo->ldo_comp_entries[i];
1199 if (lo->ldo_is_composite) {
1200 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1201 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1203 ext = &comp_v1->lcm_entries[i].lcme_extent;
1204 lod_comp->llc_extent.e_start =
1205 le64_to_cpu(ext->e_start);
1206 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1207 lod_comp->llc_flags =
1208 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1209 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1210 lod_comp->llc_timestamp = le64_to_cpu(
1211 comp_v1->lcm_entries[i].lcme_timestamp);
1213 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1214 if (lod_comp->llc_id == LCME_ID_INVAL)
1215 GOTO(out, rc = -EINVAL);
1217 if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
1218 comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
1219 struct lod_device *d =
1220 lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1222 CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
1223 lod2obd(d)->obd_name,
1224 lod_comp->llc_flags, lod_comp->llc_id, i,
1225 PFID(lod_object_fid(lo)),
1226 le32_to_cpu(comp_v1->lcm_magic));
1229 lod_comp_set_init(lod_comp);
1232 pattern = le32_to_cpu(lmm->lmm_pattern);
1233 if (!lov_pattern_supported(lov_pattern(pattern)))
1234 GOTO(out, rc = -EINVAL);
1236 lod_comp->llc_pattern = pattern;
1237 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1238 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1239 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1241 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1242 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1244 lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1245 objs = &v3->lmm_objects[0];
1247 lod_set_pool(&lod_comp->llc_pool, NULL);
1248 objs = &lmm->lmm_objects[0];
1252 * If uninstantiated template component has valid l_ost_idx,
1253 * then user has specified ost list for this component.
1255 if (!lod_comp_inited(lod_comp)) {
1258 if (objs[0].l_ost_idx != (__u32)-1UL) {
1261 stripe_count = lod_comp_entry_stripe_count(
1262 lo, lod_comp, false);
1263 if (stripe_count == 0 &&
1264 !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1265 !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1266 GOTO(out, rc = -E2BIG);
1268 * load the user specified ost list, when this
1269 * component is instantiated later, it will be
1270 * used in lod_alloc_ost_list().
1272 lod_comp->llc_ostlist.op_count = stripe_count;
1273 lod_comp->llc_ostlist.op_size =
1274 stripe_count * sizeof(__u32);
1275 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1276 lod_comp->llc_ostlist.op_size);
1277 if (!lod_comp->llc_ostlist.op_array)
1278 GOTO(out, rc = -ENOMEM);
1280 for (j = 0; j < stripe_count; j++)
1281 lod_comp->llc_ostlist.op_array[j] =
1282 le32_to_cpu(objs[j].l_ost_idx);
1285 * this component OST objects starts from the
1286 * first ost_idx, lod_alloc_ost_list() will
1289 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1292 * for uninstantiated component,
1293 * lmm_layout_gen stores default stripe offset.
1295 lod_comp->llc_stripe_offset =
1296 lmm->lmm_layout_gen;
1300 /* skip un-instantiated component object initialization */
1301 if (!lod_comp_inited(lod_comp))
1304 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1305 !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1306 rc = lod_initialize_objects(env, lo, objs, i);
1312 rc = lod_fill_mirrors(lo);
1318 lod_striping_free_nolock(env, lo);
1323 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1324 * is already cached.
1326 * \param[in] lo LOD object
1328 * \retval True if the striping is cached, otherwise
1331 static bool lod_striping_loaded(struct lod_object *lo)
1333 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1334 lo->ldo_comp_cached)
1337 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1338 if (lo->ldo_dir_stripe_loaded)
1341 /* Never load LMV stripe for slaves of striped dir */
1342 if (lo->ldo_dir_slave_stripe)
1350 * A generic function to initialize the stripe objects.
1352 * A protected version of lod_striping_load_locked() - load the striping
1353 * information from storage, parse that and instantiate LU objects to
1354 * represent the stripes. The LOD object \a lo supplies a pointer to the
1355 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1356 * return an array of references to the newly instantiated objects.
1358 * \param[in] env execution environment for this thread
1359 * \param[in,out] lo LOD object, where striping is stored and
1360 * which gets an array of references
1362 * \retval 0 if parsing and object creation succeed
1363 * \retval negative error number on failure
1365 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1367 struct lod_thread_info *info = lod_env_info(env);
1368 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1369 struct lu_buf *buf = &info->lti_buf;
1374 if (!dt_object_exists(next))
1377 if (lod_striping_loaded(lo))
1380 mutex_lock(&lo->ldo_layout_mutex);
1381 if (lod_striping_loaded(lo))
1382 GOTO(unlock, rc = 0);
1384 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1385 rc = lod_get_lov_ea(env, lo);
1390 * there is LOV EA (striping information) in this object
1391 * let's parse it and create in-core objects for the stripes
1393 buf->lb_buf = info->lti_ea_store;
1394 buf->lb_len = info->lti_ea_store_size;
1395 rc = lod_parse_striping(env, lo, buf);
1397 lo->ldo_comp_cached = 1;
1398 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1399 rc = lod_get_lmv_ea(env, lo);
1400 if (rc > sizeof(struct lmv_foreign_md)) {
1401 struct lmv_foreign_md *lfm = info->lti_ea_store;
1403 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
1404 lo->ldo_foreign_lmv = info->lti_ea_store;
1405 lo->ldo_foreign_lmv_size =
1406 info->lti_ea_store_size;
1407 info->lti_ea_store = NULL;
1408 info->lti_ea_store_size = 0;
1410 lo->ldo_dir_stripe_loaded = 1;
1411 lo->ldo_dir_is_foreign = 1;
1412 GOTO(unlock, rc = 0);
1416 if (rc < (int)sizeof(struct lmv_mds_md_v1)) {
1417 /* Let's set stripe_loaded to avoid further
1418 * stripe loading especially for non-stripe directory,
1419 * which can hurt performance. (See LU-9840)
1422 lo->ldo_dir_stripe_loaded = 1;
1423 GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1425 buf->lb_buf = info->lti_ea_store;
1426 buf->lb_len = info->lti_ea_store_size;
1427 if (rc == sizeof(struct lmv_mds_md_v1)) {
1428 rc = lod_load_lmv_shards(env, lo, buf, true);
1429 if (buf->lb_buf != info->lti_ea_store) {
1430 OBD_FREE_LARGE(info->lti_ea_store,
1431 info->lti_ea_store_size);
1432 info->lti_ea_store = buf->lb_buf;
1433 info->lti_ea_store_size = buf->lb_len;
1441 * there is LMV EA (striping information) in this object
1442 * let's parse it and create in-core objects for the stripes
1444 rc = lod_parse_dir_striping(env, lo, buf);
1446 lo->ldo_dir_stripe_loaded = 1;
1450 mutex_unlock(&lo->ldo_layout_mutex);
1455 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1456 const struct lu_buf *buf)
1462 mutex_lock(&lo->ldo_layout_mutex);
1463 lod_striping_free_nolock(env, lo);
1464 rc = lod_parse_striping(env, lo, buf);
1465 mutex_unlock(&lo->ldo_layout_mutex);
1471 * Verify lov_user_md_v1/v3 striping.
1473 * Check the validity of all fields including the magic, stripe size,
1474 * stripe count, stripe offset and that the pool is present. Also check
1475 * that each target index points to an existing target. The additional
1476 * \a is_from_disk turns additional checks. In some cases zero fields
1477 * are allowed (like pattern=0).
1479 * \param[in] d LOD device
1480 * \param[in] buf buffer with LOV EA to verify
1481 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1482 * 1 - from disk, do not allow
1484 * \retval 0 if the striping is valid
1485 * \retval -EINVAL if striping is invalid
1487 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1490 struct lov_user_md_v1 *lum;
1491 struct lov_user_md_v3 *lum3;
1492 struct pool_desc *pool = NULL;
1496 __u16 stripe_offset;
1503 if (buf->lb_len < sizeof(*lum)) {
1504 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1506 GOTO(out, rc = -EINVAL);
1509 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1510 if (magic != LOV_USER_MAGIC_V1 &&
1511 magic != LOV_USER_MAGIC_V3 &&
1512 magic != LOV_USER_MAGIC_SPECIFIC) {
1513 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1514 le32_to_cpu(lum->lmm_magic));
1515 GOTO(out, rc = -EINVAL);
1518 /* the user uses "0" for default stripe pattern normally. */
1519 if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1520 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1522 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1523 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1524 le32_to_cpu(lum->lmm_pattern));
1525 GOTO(out, rc = -EINVAL);
1528 /* a released lum comes from creating orphan on hsm release,
1529 * doesn't make sense to verify it. */
1530 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1533 /* 64kB is the largest common page size we see (ia64), and matches the
1535 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1536 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1537 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1538 stripe_size, LOV_MIN_STRIPE_SIZE);
1539 GOTO(out, rc = -EINVAL);
1542 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1543 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1544 lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1545 /* if offset is not within valid range [0, osts_size) */
1546 if (stripe_offset >= d->lod_ost_descs.ltd_tgts_size) {
1547 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1548 stripe_offset, d->lod_ost_descs.ltd_tgts_size);
1549 GOTO(out, rc = -EINVAL);
1552 /* if lmm_stripe_offset is *not* in bitmap */
1553 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1554 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1556 GOTO(out, rc = -EINVAL);
1560 if (magic == LOV_USER_MAGIC_V1)
1561 lum_size = offsetof(struct lov_user_md_v1,
1563 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1564 lum_size = offsetof(struct lov_user_md_v3,
1567 GOTO(out, rc = -EINVAL);
1569 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1570 if (buf->lb_len < lum_size) {
1571 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1572 "magic %#x and stripe_count %u\n",
1573 buf->lb_len, lum_size, magic, stripe_count);
1574 GOTO(out, rc = -EINVAL);
1577 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1581 /* In the function below, .hs_keycmp resolves to
1582 * pool_hashkey_keycmp() */
1583 /* coverity[overrun-buffer-val] */
1584 pool = lod_find_pool(d, lum3->lmm_pool_name);
1588 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1589 rc = lod_check_index_in_pool(stripe_offset, pool);
1591 GOTO(out, rc = -EINVAL);
1594 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1595 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1596 stripe_count, pool_tgt_count(pool));
1597 GOTO(out, rc = -EINVAL);
1602 lod_pool_putref(pool);
1608 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1610 LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1611 LOV_USER_MAGIC_COMP_V1, "Wrong magic %x\n",
1612 le32_to_cpu(comp->lcm_magic));
1613 LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1614 "bad index %d, max = %d\n",
1615 i, le16_to_cpu(comp->lcm_entry_count));
1617 return &comp->lcm_entries[i];
1620 #define for_each_comp_entry_v1(comp, entry) \
1621 for (entry = comp_entry_v1(comp, 0); \
1622 entry <= comp_entry_v1(comp, \
1623 le16_to_cpu(comp->lcm_entry_count) - 1); \
1626 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
1627 struct lov_comp_md_entry_v1 *dom_ent)
1629 struct lov_comp_md_entry_v1 *ent;
1631 __u32 dom_off, dom_size, comp_size;
1632 void *blob_src, *blob_dst;
1633 unsigned int blob_size, blob_shift;
1635 entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1636 /* if file has only DoM stripe return just error */
1640 comp_size = le32_to_cpu(comp_v1->lcm_size);
1641 dom_off = le32_to_cpu(dom_ent->lcme_offset);
1642 dom_size = le32_to_cpu(dom_ent->lcme_size);
1644 /* shift entries array first */
1645 comp_v1->lcm_entry_count = cpu_to_le16(entries);
1646 memmove(dom_ent, dom_ent + 1,
1647 entries * sizeof(struct lov_comp_md_entry_v1));
1649 /* now move blob of layouts */
1650 blob_dst = (void *)comp_v1 + dom_off - sizeof(*dom_ent);
1651 blob_src = (void *)comp_v1 + dom_off + dom_size;
1652 blob_size = (unsigned long)((void *)comp_v1 + comp_size - blob_src);
1653 blob_shift = sizeof(*dom_ent) + dom_size;
1655 memmove(blob_dst, blob_src, blob_size);
1657 for_each_comp_entry_v1(comp_v1, ent) {
1660 off = le32_to_cpu(ent->lcme_offset);
1661 ent->lcme_offset = cpu_to_le32(off - blob_shift);
1664 comp_v1->lcm_size = cpu_to_le32(comp_size - blob_shift);
1666 /* notify a caller to re-check entry */
1670 int lod_fix_dom_stripe(struct lod_device *d, struct lov_comp_md_v1 *comp_v1,
1671 struct lov_comp_md_entry_v1 *dom_ent)
1673 struct lov_comp_md_entry_v1 *ent;
1674 struct lu_extent *dom_ext, *ext;
1675 struct lov_user_md_v1 *lum;
1680 dom_ext = &dom_ent->lcme_extent;
1681 dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1682 stripe_size = d->lod_dom_max_stripesize;
1684 lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1685 CDEBUG(D_LAYOUT, "DoM component size %u was bigger than MDT limit %u, "
1686 "new size is %u\n", le32_to_cpu(lum->lmm_stripe_size),
1687 d->lod_dom_max_stripesize, stripe_size);
1688 lum->lmm_stripe_size = cpu_to_le32(stripe_size);
1690 for_each_comp_entry_v1(comp_v1, ent) {
1694 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1698 ext = &ent->lcme_extent;
1699 if (ext->e_start != dom_ext->e_end)
1702 /* Found next component after the DoM one with the same
1703 * mirror_id and adjust its start with DoM component end.
1705 * NOTE: we are considering here that there can be only one
1706 * DoM component in a file, all replicas are located on OSTs
1707 * always and don't need adjustment since use own layouts.
1709 ext->e_start = cpu_to_le64(stripe_size);
1713 if (stripe_size == 0) {
1714 /* DoM component size is zero due to server setting,
1715 * remove it from the layout */
1716 rc = lod_erase_dom_stripe(comp_v1, dom_ent);
1718 /* Update DoM extent end finally */
1719 dom_ext->e_end = cpu_to_le64(stripe_size);
1726 * Verify LOV striping.
1728 * \param[in] d LOD device
1729 * \param[in] buf buffer with LOV EA to verify
1730 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1731 * 1 - from disk, do not allow
1732 * \param[in] start extent start for composite layout
1734 * \retval 0 if the striping is valid
1735 * \retval -EINVAL if striping is invalid
1737 int lod_verify_striping(struct lod_device *d, struct lod_object *lo,
1738 const struct lu_buf *buf, bool is_from_disk)
1740 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
1741 struct lov_user_md_v1 *lum;
1742 struct lov_comp_md_v1 *comp_v1;
1743 struct lov_comp_md_entry_v1 *ent;
1744 struct lu_extent *ext;
1747 __u32 stripe_size = 0;
1748 __u16 prev_mid = -1, mirror_id = -1;
1754 if (buf->lb_len < sizeof(lum->lmm_magic)) {
1755 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
1761 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1762 /* treat foreign LOV EA/object case first
1763 * XXX is it expected to try setting again a foreign?
1764 * XXX should we care about different current vs new layouts ?
1766 if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
1767 struct lov_foreign_md *lfm = buf->lb_buf;
1769 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
1771 "buf len %zu < min lov_foreign_md size (%zu)\n",
1772 buf->lb_len, offsetof(typeof(*lfm),
1777 if (foreign_size_le(lfm) > buf->lb_len) {
1779 "buf len %zu < this lov_foreign_md size (%zu)\n",
1780 buf->lb_len, foreign_size_le(lfm));
1783 /* Don't do anything with foreign layouts */
1787 /* normal LOV/layout cases */
1789 if (buf->lb_len < sizeof(*lum)) {
1790 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1795 if (magic != LOV_USER_MAGIC_V1 &&
1796 magic != LOV_USER_MAGIC_V3 &&
1797 magic != LOV_USER_MAGIC_SPECIFIC &&
1798 magic != LOV_USER_MAGIC_COMP_V1) {
1799 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1800 le32_to_cpu(lum->lmm_magic));
1804 if (magic != LOV_USER_MAGIC_COMP_V1)
1805 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1807 /* magic == LOV_USER_MAGIC_COMP_V1 */
1808 comp_v1 = buf->lb_buf;
1809 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1810 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1811 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1817 if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1818 CDEBUG(D_LAYOUT, "entry count is zero\n");
1822 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1823 lo->ldo_comp_cnt > 0) {
1824 /* could be called from lustre.lov.add */
1825 __u32 cnt = lo->ldo_comp_cnt;
1827 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
1828 prev_end = ext->e_end;
1833 for_each_comp_entry_v1(comp_v1, ent) {
1834 ext = &ent->lcme_extent;
1836 if (le64_to_cpu(ext->e_start) > le64_to_cpu(ext->e_end)) {
1837 CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
1838 le64_to_cpu(ext->e_start),
1839 le64_to_cpu(ext->e_end));
1844 /* lcme_id contains valid value */
1845 if (le32_to_cpu(ent->lcme_id) == 0 ||
1846 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
1847 CDEBUG(D_LAYOUT, "invalid id %u\n",
1848 le32_to_cpu(ent->lcme_id));
1852 if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
1853 mirror_id = mirror_id_of(
1854 le32_to_cpu(ent->lcme_id));
1856 /* first component must start with 0 */
1857 if (mirror_id != prev_mid &&
1858 le64_to_cpu(ext->e_start) != 0) {
1860 "invalid start:%llu, expect:0\n",
1861 le64_to_cpu(ext->e_start));
1865 prev_mid = mirror_id;
1869 if (le64_to_cpu(ext->e_start) == 0) {
1874 /* the next must be adjacent with the previous one */
1875 if (le64_to_cpu(ext->e_start) != prev_end) {
1877 "invalid start actual:%llu, expect:%llu\n",
1878 le64_to_cpu(ext->e_start), prev_end);
1882 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
1883 tmp.lb_len = le32_to_cpu(ent->lcme_size);
1885 /* Check DoM entry is always the first one */
1887 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
1889 /* DoM component must be the first in a mirror */
1890 if (le64_to_cpu(ext->e_start) > 0) {
1891 CDEBUG(D_LAYOUT, "invalid DoM component "
1892 "with %llu extent start\n",
1893 le64_to_cpu(ext->e_start));
1896 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1897 /* There is just one stripe on MDT and it must
1898 * cover whole component size. */
1899 if (stripe_size != le64_to_cpu(ext->e_end)) {
1900 CDEBUG(D_LAYOUT, "invalid DoM layout "
1901 "stripe size %u != %llu "
1902 "(component size)\n",
1903 stripe_size, prev_end);
1906 /* Check stripe size againts per-MDT limit */
1907 if (stripe_size > d->lod_dom_max_stripesize) {
1908 CDEBUG(D_LAYOUT, "DoM component size "
1909 "%u is bigger than MDT limit %u, check "
1910 "dom_max_stripesize parameter\n",
1911 stripe_size, d->lod_dom_max_stripesize);
1912 rc = lod_fix_dom_stripe(d, comp_v1, ent);
1913 if (rc == -ERESTART) {
1914 /* DoM entry was removed, re-check
1915 * new layout from start */
1923 prev_end = le64_to_cpu(ext->e_end);
1925 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
1929 if (prev_end == LUSTRE_EOF)
1932 /* extent end must be aligned with the stripe_size */
1933 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1934 if (stripe_size == 0)
1935 stripe_size = desc->ld_default_stripe_size;
1936 if (prev_end % stripe_size) {
1937 CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
1938 "stripe_sz: %u, [%llu, %llu)\n",
1939 stripe_size, ext->e_start, prev_end);
1944 /* make sure that the mirror_count is telling the truth */
1945 if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
1952 * set the default stripe size, if unset.
1954 * \param[in,out] val number of bytes per OST stripe
1956 * The minimum stripe size is 64KB to ensure that a single stripe is an
1957 * even multiple of a client PAGE_SIZE (IA64, PPC, etc). Otherwise, it
1958 * is difficult to split dirty pages across OSCs during writes.
1960 void lod_fix_desc_stripe_size(__u64 *val)
1962 if (*val < LOV_MIN_STRIPE_SIZE) {
1964 LCONSOLE_INFO("Increasing default stripe size to "
1965 "minimum value %u\n",
1966 LOV_DESC_STRIPE_SIZE_DEFAULT);
1967 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1968 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1969 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1970 LCONSOLE_WARN("Changing default stripe size to %llu (a "
1971 "multiple of %u)\n",
1972 *val, LOV_MIN_STRIPE_SIZE);
1977 * set the filesystem default number of stripes, if unset.
1979 * \param[in,out] val number of stripes
1981 * A value of "0" means "use the system-wide default stripe count", which
1982 * has either been inherited by now, or falls back to 1 stripe per file.
1983 * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
1984 * and is a valid value, so is left unchanged here.
1986 void lod_fix_desc_stripe_count(__u32 *val)
1993 * set the filesystem default layout pattern
1995 * \param[in,out] val LOV_PATTERN_* layout
1997 * A value of "0" means "use the system-wide default layout type", which
1998 * has either been inherited by now, or falls back to plain RAID0 striping.
2000 void lod_fix_desc_pattern(__u32 *val)
2002 /* from lov_setstripe */
2003 if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
2004 LCONSOLE_WARN("lod: Unknown stripe pattern: %#x\n", *val);
2009 void lod_fix_desc_qos_maxage(__u32 *val)
2011 /* fix qos_maxage */
2013 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2017 * Used to fix insane default striping.
2019 * \param[in] desc striping description
2021 void lod_fix_desc(struct lov_desc *desc)
2023 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2024 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2025 lod_fix_desc_pattern(&desc->ld_pattern);
2026 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2030 * Initialize the structures used to store pools and default striping.
2032 * \param[in] lod LOD device
2033 * \param[in] lcfg configuration structure storing default striping.
2035 * \retval 0 if initialization succeeds
2036 * \retval negative error number on failure
2038 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2040 struct obd_device *obd;
2041 struct lov_desc *desc;
2045 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2046 LASSERT(obd != NULL);
2047 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2049 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2050 CERROR("LOD setup requires a descriptor\n");
2054 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2056 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2057 CERROR("descriptor size wrong: %d > %d\n",
2058 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2062 if (desc->ld_magic != LOV_DESC_MAGIC) {
2063 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2064 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2065 obd->obd_name, desc);
2066 lustre_swab_lov_desc(desc);
2068 CERROR("%s: Bad lov desc magic: %#x\n",
2069 obd->obd_name, desc->ld_magic);
2076 desc->ld_active_tgt_count = 0;
2077 lod->lod_ost_descs.ltd_lov_desc = *desc;
2079 lod->lod_sp_me = LUSTRE_SP_CLI;
2081 /* Set up OST pool environment */
2082 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
2083 HASH_POOLS_MAX_BITS,
2084 HASH_POOLS_BKT_BITS, 0,
2087 &pool_hash_operations,
2089 if (lod->lod_pools_hash_body == NULL)
2092 INIT_LIST_HEAD(&lod->lod_pool_list);
2093 lod->lod_pool_count = 0;
2094 rc = lod_ost_pool_init(&lod->lod_ost_descs.ltd_tgt_pool, 0);
2097 rc = lod_ost_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
2099 GOTO(out_pool_info, rc);
2104 lod_ost_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2106 cfs_hash_putref(lod->lod_pools_hash_body);
2112 * Release the structures describing the pools.
2114 * \param[in] lod LOD device from which we release the structures
2118 int lod_pools_fini(struct lod_device *lod)
2120 struct obd_device *obd = lod2obd(lod);
2121 struct pool_desc *pool, *tmp;
2124 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2125 /* free pool structs */
2126 CDEBUG(D_INFO, "delete pool %p\n", pool);
2127 /* In the function below, .hs_keycmp resolves to
2128 * pool_hashkey_keycmp() */
2129 /* coverity[overrun-buffer-val] */
2130 lod_pool_del(obd, pool->pool_name);
2133 cfs_hash_putref(lod->lod_pools_hash_body);
2134 lod_ost_pool_free(&(lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool));
2135 lod_ost_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);