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, 2016, 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;
82 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
85 INIT_LIST_HEAD(&kill);
87 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
88 tgt_desc = LTD_TGT(ltd, idx);
91 if (!tgt_desc->ltd_reap)
94 list_add(&tgt_desc->ltd_kill, &kill);
95 LTD_TGT(ltd, idx) = NULL;
96 /*FIXME: only support ost pool for now */
97 if (ltd == &lod->lod_ost_descs) {
98 lod_ost_pool_remove(&lod->lod_pool_info, idx);
99 if (tgt_desc->ltd_active)
100 lod->lod_desc.ld_active_tgt_count--;
103 cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
104 ltd->ltd_death_row--;
106 mutex_unlock(<d->ltd_mutex);
107 up_read(<d->ltd_rw_sem);
109 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
111 list_del(&tgt_desc->ltd_kill);
112 if (ltd == &lod->lod_ost_descs) {
113 /* remove from QoS structures */
114 rc = qos_del_tgt(lod, tgt_desc);
116 CERROR("%s: qos_del_tgt(%s) failed:"
118 lod2obd(lod)->obd_name,
119 obd_uuid2str(&tgt_desc->ltd_uuid),
122 rc = obd_disconnect(tgt_desc->ltd_exp);
124 CERROR("%s: failed to disconnect %s: rc = %d\n",
125 lod2obd(lod)->obd_name,
126 obd_uuid2str(&tgt_desc->ltd_uuid), rc);
127 OBD_FREE_PTR(tgt_desc);
130 mutex_unlock(<d->ltd_mutex);
131 up_read(<d->ltd_rw_sem);
136 * Expand size of target table.
138 * When the target table is full, we have to extend the table. To do so,
139 * we allocate new memory with some reserve, move data from the old table
140 * to the new one and release memory consumed by the old table.
141 * Notice we take ltd_rw_sem exclusively to ensure atomic switch.
143 * \param[in] ltd target table
144 * \param[in] newsize new size of the table
146 * \retval 0 on success
147 * \retval -ENOMEM if reallocation failed
149 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
151 struct cfs_bitmap *new_bitmap, *old_bitmap = NULL;
155 /* grab write reference on the lod. Relocating the array requires
156 * exclusive access */
158 down_write(<d->ltd_rw_sem);
159 if (newsize <= ltd->ltd_tgts_size)
160 /* someone else has already resize the array */
163 /* allocate new bitmap */
164 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
166 GOTO(out, rc = -ENOMEM);
168 if (ltd->ltd_tgts_size > 0) {
169 /* the bitmap already exists, we need
170 * to copy data from old one */
171 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
172 old_bitmap = ltd->ltd_tgt_bitmap;
175 ltd->ltd_tgts_size = newsize;
176 ltd->ltd_tgt_bitmap = new_bitmap;
179 CFS_FREE_BITMAP(old_bitmap);
181 CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
185 up_write(<d->ltd_rw_sem);
190 * Connect LOD to a new OSP and add it to the target table.
192 * Connect to the OSP device passed, initialize all the internal
193 * structures related to the device and add it to the target table.
195 * \param[in] env execution environment for this thread
196 * \param[in] lod LOD device to be connected to the new OSP
197 * \param[in] osp name of OSP device name to be added
198 * \param[in] index index of the new target
199 * \param[in] gen target's generation number
200 * \param[in] tgt_index OSP's group
201 * \param[in] type type of device (mdc or osc)
202 * \param[in] active state of OSP: 0 - inactive, 1 - active
204 * \retval 0 if added successfully
205 * \retval negative error number on failure
207 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
208 char *osp, unsigned index, unsigned gen, int tgt_index,
209 char *type, int active)
211 struct obd_connect_data *data = NULL;
212 struct obd_export *exp = NULL;
213 struct obd_device *obd;
214 struct lu_device *lu_dev;
215 struct dt_device *dt_dev;
217 struct lod_tgt_desc *tgt_desc;
218 struct lod_tgt_descs *ltd;
219 struct lustre_cfg *lcfg;
220 struct obd_uuid obd_uuid;
225 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
228 CERROR("request to add OBD %s with invalid generation: %d\n",
233 obd_str2uuid(&obd_uuid, osp);
235 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
236 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
238 CERROR("can't find %s device\n", osp);
242 LASSERT(obd->obd_lu_dev != NULL);
243 LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
245 lu_dev = obd->obd_lu_dev;
246 dt_dev = lu2dt_dev(lu_dev);
250 GOTO(out_cleanup, rc = -ENOMEM);
252 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
253 data->ocd_version = LUSTRE_VERSION_CODE;
254 data->ocd_index = index;
256 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
258 data->ocd_connect_flags |= OBD_CONNECT_AT |
261 #ifdef HAVE_LRU_RESIZE_SUPPORT
262 OBD_CONNECT_LRU_RESIZE |
265 OBD_CONNECT_REQPORTAL |
266 OBD_CONNECT_SKIP_ORPHAN |
268 OBD_CONNECT_LVB_TYPE |
269 OBD_CONNECT_VERSION |
270 OBD_CONNECT_PINGLESS |
272 OBD_CONNECT_BULK_MBITS;
274 data->ocd_group = tgt_index;
275 ltd = &lod->lod_ost_descs;
277 struct obd_import *imp = obd->u.cli.cl_import;
280 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
281 data->ocd_connect_flags |= OBD_CONNECT_ACL |
283 OBD_CONNECT_MDS_MDS |
288 OBD_CONNECT_BULK_MBITS;
289 spin_lock(&imp->imp_lock);
290 imp->imp_server_timeout = 1;
291 spin_unlock(&imp->imp_lock);
292 imp->imp_client->cli_request_portal = OUT_PORTAL;
293 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
295 ltd = &lod->lod_mdt_descs;
298 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
301 CERROR("%s: cannot connect to next dev %s (%d)\n",
302 obd->obd_name, osp, rc);
303 GOTO(out_cleanup, rc);
306 /* Allocate ost descriptor and fill it */
307 OBD_ALLOC_PTR(tgt_desc);
309 GOTO(out_conn, rc = -ENOMEM);
311 tgt_desc->ltd_tgt = dt_dev;
312 tgt_desc->ltd_exp = exp;
313 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
314 tgt_desc->ltd_gen = gen;
315 tgt_desc->ltd_index = index;
316 tgt_desc->ltd_active = active;
319 if (index >= ltd->ltd_tgts_size) {
320 /* we have to increase the size of the lod_osts array */
323 newsize = max(ltd->ltd_tgts_size, (__u32)2);
324 while (newsize < index + 1)
325 newsize = newsize << 1;
327 /* lod_bitmap_resize() needs lod_rw_sem
328 * which we hold with th reference */
329 lod_putref(lod, ltd);
331 rc = ltd_bitmap_resize(ltd, newsize);
338 mutex_lock(<d->ltd_mutex);
340 if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
341 CERROR("%s: device %d is registered already\n", obd->obd_name,
343 GOTO(out_mutex, rc = -EEXIST);
346 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
347 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
348 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
349 CERROR("can't allocate index to add %s\n",
351 GOTO(out_mutex, rc = -ENOMEM);
356 /* pool and qos are not supported for MDS stack yet */
357 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
360 CERROR("%s: can't set up pool, failed with %d\n",
365 rc = qos_add_tgt(lod, tgt_desc);
367 CERROR("%s: qos_add_tgt failed with %d\n",
372 /* The new OST is now a full citizen */
373 if (index >= lod->lod_desc.ld_tgt_count)
374 lod->lod_desc.ld_tgt_count = index + 1;
376 lod->lod_desc.ld_active_tgt_count++;
379 LTD_TGT(ltd, index) = tgt_desc;
380 cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
382 mutex_unlock(<d->ltd_mutex);
383 lod_putref(lod, ltd);
385 if (lod->lod_recovery_completed)
386 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
388 if (!for_ost && lod->lod_initialized) {
389 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
391 CERROR("%s: cannot start llog on %s:rc = %d\n",
392 lod2obd(lod)->obd_name, osp, rc);
397 rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
399 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
400 "index = %u, rc = %d\n", osp, type, index, rc);
401 GOTO(out_fini_llog, rc);
405 lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
406 tgt_desc->ltd_recovery_thread);
409 mutex_lock(<d->ltd_mutex);
411 if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
412 struct ptlrpc_thread *thread;
414 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
415 OBD_FREE_PTR(thread);
418 cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index);
419 LTD_TGT(ltd, index) = NULL;
421 lod_ost_pool_remove(&lod->lod_pool_info, index);
424 mutex_unlock(<d->ltd_mutex);
425 lod_putref(lod, ltd);
428 OBD_FREE_PTR(tgt_desc);
432 /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
433 * objects from the MDT stack. See LU-7184. */
434 lcfg = &lod_env_info(env)->lti_lustre_cfg;
435 memset(lcfg, 0, sizeof(*lcfg));
436 lcfg->lcfg_version = LUSTRE_CFG_VERSION;
437 lcfg->lcfg_command = LCFG_CLEANUP;
438 lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
444 * Schedule target removal from the target table.
446 * Mark the device as dead. The device is not removed here because it may
447 * still be in use. The device will be removed in lod_putref() when the
448 * last reference is released.
450 * \param[in] env execution environment for this thread
451 * \param[in] lod LOD device the target table belongs to
452 * \param[in] ltd target table
453 * \param[in] idx index of the target
454 * \param[in] for_ost type of the target: 0 - MDT, 1 - OST
456 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
457 struct lod_tgt_descs *ltd, unsigned idx,
460 LASSERT(LTD_TGT(ltd, idx));
462 lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
465 if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
466 struct ptlrpc_thread *thread;
468 thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
469 OBD_FREE_PTR(thread);
472 if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
473 LTD_TGT(ltd, idx)->ltd_reap = 1;
474 ltd->ltd_death_row++;
479 * Schedule removal of all the targets from the given target table.
481 * See more details in the description for __lod_del_device()
483 * \param[in] env execution environment for this thread
484 * \param[in] lod LOD device the target table belongs to
485 * \param[in] ltd target table
486 * \param[in] for_ost type of the target: MDT or OST
490 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
491 struct lod_tgt_descs *ltd, bool for_ost)
495 if (ltd->ltd_tgts_size <= 0)
498 mutex_lock(<d->ltd_mutex);
499 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
500 __lod_del_device(env, lod, ltd, idx, for_ost);
501 mutex_unlock(<d->ltd_mutex);
502 lod_putref(lod, ltd);
503 CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
504 for (idx = 0; idx < TGT_PTRS; idx++) {
505 if (ltd->ltd_tgt_idx[idx])
506 OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
508 ltd->ltd_tgts_size = 0;
513 * Remove device by name.
515 * Remove a device identified by \a osp from the target table. Given
516 * the device can be in use, the real deletion happens in lod_putref().
518 * \param[in] env execution environment for this thread
519 * \param[in] lod LOD device to be connected to the new OSP
520 * \param[in] ltd target table
521 * \param[in] osp name of OSP device to be removed
522 * \param[in] idx index of the target
523 * \param[in] gen generation number, not used currently
524 * \param[in] for_ost type of the target: 0 - MDT, 1 - OST
526 * \retval 0 if the device was scheduled for removal
527 * \retval -EINVAL if no device was found
529 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
530 struct lod_tgt_descs *ltd, char *osp, unsigned idx,
531 unsigned gen, bool for_ost)
533 struct obd_device *obd;
535 struct obd_uuid uuid;
538 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
540 obd_str2uuid(&uuid, osp);
542 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
543 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
545 CERROR("can't find %s device\n", osp);
550 CERROR("%s: request to remove OBD %s with invalid generation %d"
551 "\n", obd->obd_name, osp, gen);
555 obd_str2uuid(&uuid, osp);
558 mutex_lock(<d->ltd_mutex);
559 /* check that the index is allocated in the bitmap */
560 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
561 !LTD_TGT(ltd, idx)) {
562 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
563 GOTO(out, rc = -EINVAL);
566 /* check that the UUID matches */
567 if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
568 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
569 obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
571 GOTO(out, rc = -EINVAL);
574 __lod_del_device(env, lod, ltd, idx, for_ost);
577 mutex_unlock(<d->ltd_mutex);
578 lod_putref(lod, ltd);
583 * Resize per-thread storage to hold specified size.
585 * A helper function to resize per-thread temporary storage. This storage
586 * is used to process LOV/LVM EAs and may be quite large. We do not want to
587 * allocate/release it every time, so instead we put it into the env and
588 * reallocate on demand. The memory is released when the correspondent thread
591 * \param[in] info LOD-specific storage in the environment
592 * \param[in] size new size to grow the buffer to
594 * \retval 0 on success, -ENOMEM if reallocation failed
596 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
598 __u32 round = size_roundup_power2(size);
601 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
602 if (info->lti_ea_store) {
603 LASSERT(info->lti_ea_store_size);
604 LASSERT(info->lti_ea_store_size < round);
605 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
606 info->lti_ea_store_size, round);
607 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
608 info->lti_ea_store = NULL;
609 info->lti_ea_store_size = 0;
612 OBD_ALLOC_LARGE(info->lti_ea_store, round);
613 if (info->lti_ea_store == NULL)
615 info->lti_ea_store_size = round;
620 static void lod_free_comp_buffer(struct lod_layout_component *entries,
621 __u16 count, __u32 bufsize)
623 struct lod_layout_component *entry;
626 for (i = 0; i < count; i++) {
628 if (entry->llc_pool != NULL)
629 lod_set_pool(&entry->llc_pool, NULL);
630 LASSERT(entry->llc_stripe == NULL);
631 LASSERT(entry->llc_stripes_allocated == 0);
635 OBD_FREE_LARGE(entries, bufsize);
638 void lod_free_def_comp_entries(struct lod_default_striping *lds)
640 lod_free_comp_buffer(lds->lds_def_comp_entries,
641 lds->lds_def_comp_size_cnt,
643 sizeof(*lds->lds_def_comp_entries) *
644 lds->lds_def_comp_size_cnt));
645 lds->lds_def_comp_entries = NULL;
646 lds->lds_def_comp_cnt = 0;
647 lds->lds_def_striping_is_composite = 0;
648 lds->lds_def_comp_size_cnt = 0;
652 * Resize per-thread storage to hold default striping component entries
654 * A helper function to resize per-thread temporary storage. This storage
655 * is used to hold default LOV/LVM EAs and may be quite large. We do not want
656 * to allocate/release it every time, so instead we put it into the env and
657 * reallocate it on demand. The memory is released when the correspondent
658 * thread is finished.
660 * \param[in,out] lds default striping
661 * \param[in] count new component count to grow the buffer to
663 * \retval 0 on success, -ENOMEM if reallocation failed
665 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
667 struct lod_layout_component *entries;
668 __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
670 __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
671 lds->lds_def_comp_size_cnt);
676 OBD_ALLOC_LARGE(entries, new);
680 if (lds->lds_def_comp_entries != NULL) {
681 CDEBUG(D_INFO, "default striping component size %d is not "
682 "enough, need %d\n", old, new);
683 lod_free_def_comp_entries(lds);
686 lds->lds_def_comp_entries = entries;
687 lds->lds_def_comp_size_cnt = count;
692 void lod_free_comp_entries(struct lod_object *lo)
694 lod_free_comp_buffer(lo->ldo_comp_entries,
696 sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
697 lo->ldo_comp_entries = NULL;
698 lo->ldo_comp_cnt = 0;
699 lo->ldo_is_composite = 0;
702 int lod_alloc_comp_entries(struct lod_object *lo, int cnt)
705 LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
707 OBD_ALLOC_LARGE(lo->ldo_comp_entries,
708 sizeof(*lo->ldo_comp_entries) * cnt);
709 if (lo->ldo_comp_entries == NULL)
711 lo->ldo_comp_cnt = cnt;
716 * Generate on-disk lov_mds_md structure for each layout component based on
717 * the information in lod_object->ldo_comp_entries[i].
719 * \param[in] env execution environment for this thread
720 * \param[in] lo LOD object
721 * \param[in] comp_idx index of ldo_comp_entries
722 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
723 * \param[in|out] lmm_size buffer size/lmm size
724 * \param[in] is_dir generate lov ea for dir or file? For dir case,
725 * the stripe info is from the default stripe
726 * template, which is collected in lod_ah_init(),
727 * either from parent object or root object; for
728 * file case, it's from the @lo object
730 * \retval 0 if on disk structure is created successfully
731 * \retval negative error number on failure
733 static int lod_gen_component_ea(const struct lu_env *env,
734 struct lod_object *lo, int comp_idx,
735 struct lov_mds_md *lmm, int *lmm_size,
738 struct lod_thread_info *info = lod_env_info(env);
739 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
740 struct lod_device *lod;
741 struct lov_ost_data_v1 *objs;
742 struct lod_layout_component *lod_comp;
750 &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
752 lod_comp = &lo->ldo_comp_entries[comp_idx];
754 magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
755 if (lod_comp->llc_pattern == 0) /* default striping */
756 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
758 lmm->lmm_magic = cpu_to_le32(magic);
759 lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
760 fid_to_lmm_oi(fid, &lmm->lmm_oi);
761 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
762 lmm->lmm_oi.oi.oi_id++;
763 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
765 lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
766 lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripenr);
767 /* for dir, lmm_layout_gen stores default stripe offset. */
768 lmm->lmm_layout_gen = is_dir ?
769 cpu_to_le16(lod_comp->llc_stripe_offset) :
770 cpu_to_le16(lod_comp->llc_layout_gen);
772 if (magic == LOV_MAGIC_V1) {
773 objs = &lmm->lmm_objects[0];
775 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
776 size_t cplen = strlcpy(v3->lmm_pool_name,
778 sizeof(v3->lmm_pool_name));
779 if (cplen >= sizeof(v3->lmm_pool_name))
781 objs = &v3->lmm_objects[0];
784 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
787 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
788 for (i = 0; i < lod_comp->llc_stripenr; i++) {
789 struct dt_object *object;
791 int type = LU_SEQ_RANGE_OST;
793 object = lod_comp->llc_stripe[i];
794 LASSERT(object != NULL);
795 info->lti_fid = *lu_object_fid(&object->do_lu);
797 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
799 if (cfs_fail_val == 0)
800 cfs_fail_val = info->lti_fid.f_oid;
802 info->lti_fid.f_oid = cfs_fail_val;
805 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
808 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
809 objs[i].l_ost_gen = cpu_to_le32(0);
810 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
813 rc = lod_fld_lookup(env, lod, &info->lti_fid,
816 CERROR("%s: Can not locate "DFID": rc = %d\n",
817 lod2obd(lod)->obd_name, PFID(&info->lti_fid),
821 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
824 if (lmm_size != NULL)
825 *lmm_size = lov_mds_md_size(is_dir ?
826 0 : lod_comp->llc_stripenr, magic);
831 * Generate component ID for new created component.
833 * \param[in] lo LOD object
834 * \param[in] comp_idx index of ldo_comp_entries
836 * \retval component ID on success
837 * \retval LCME_ID_INVAL on failure
839 static __u32 lod_gen_component_id(struct lod_object *lo, int comp_idx)
841 struct lod_layout_component *lod_comp;
842 __u32 id, start, end;
845 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
847 lod_obj_inc_layout_gen(lo);
848 id = lo->ldo_layout_gen;
849 if (likely(id <= LCME_ID_MAX))
852 /* Layout generation wraps, need to check collisions. */
853 start = id & LCME_ID_MASK;
854 end = (__u32)LCME_ID_MAX + 1;
856 for (id = start; id < end; id++) {
857 for (i = 0; i < lo->ldo_comp_cnt; i++) {
858 lod_comp = &lo->ldo_comp_entries[i];
859 if (id == lod_comp->llc_id)
862 /* Found the ununsed ID */
863 if (i == lo->ldo_comp_cnt)
866 if (end == (__u32)LCME_ID_MAX + 1) {
868 end = lo->ldo_layout_gen & LCME_ID_MASK;
872 return LCME_ID_INVAL;
876 * Generate on-disk lov_mds_md structure based on the information in
877 * the lod_object->ldo_comp_entries.
879 * \param[in] env execution environment for this thread
880 * \param[in] lo LOD object
881 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
882 * \param[in|out] lmm_size buffer size/lmm size
883 * \param[in] is_dir generate lov ea for dir or file? For dir case,
884 * the stripe info is from the default stripe
885 * template, which is collected in lod_ah_init(),
886 * either from parent object or root object; for
887 * file case, it's from the @lo object
889 * \retval 0 if on disk structure is created successfully
890 * \retval negative error number on failure
892 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
893 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
895 struct lov_comp_md_entry_v1 *lcme;
896 struct lov_comp_md_v1 *lcm;
897 struct lod_layout_component *comp_entries;
900 int i, rc = 0, offset;
904 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
905 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
907 lo->ldo_def_striping->lds_def_striping_is_composite;
909 comp_cnt = lo->ldo_comp_cnt;
910 comp_entries = lo->ldo_comp_entries;
911 is_composite = lo->ldo_is_composite;
914 LASSERT(lmm_size != NULL);
915 LASSERT(comp_cnt != 0 && comp_entries != NULL);
918 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
922 lcm = (struct lov_comp_md_v1 *)lmm;
923 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
924 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
926 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
927 LASSERT(offset % sizeof(__u64) == 0);
929 for (i = 0; i < comp_cnt; i++) {
930 struct lod_layout_component *lod_comp;
931 struct lov_mds_md *sub_md;
934 lod_comp = &comp_entries[i];
935 lcme = &lcm->lcm_entries[i];
937 if (lod_comp->llc_id == LCME_ID_INVAL && !is_dir) {
938 lod_comp->llc_id = lod_gen_component_id(lo, i);
939 if (lod_comp->llc_id == LCME_ID_INVAL)
940 GOTO(out, rc = -ERANGE);
942 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
943 /* component must has been inistantiated */
944 LASSERT(ergo(!is_dir, lod_comp->llc_flags & LCME_FL_INIT));
945 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
946 lcme->lcme_extent.e_start =
947 cpu_to_le64(lod_comp->llc_extent.e_start);
948 lcme->lcme_extent.e_end =
949 cpu_to_le64(lod_comp->llc_extent.e_end);
950 lcme->lcme_offset = cpu_to_le32(offset);
952 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
953 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
956 lcme->lcme_size = cpu_to_le32(size);
958 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
959 "offset:%d lmm_size:%d\n", offset, *lmm_size);
961 lcm->lcm_size = cpu_to_le32(offset);
962 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
964 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
975 * Fill lti_ea_store buffer in the environment with a value for the given
976 * EA. The buffer is reallocated if the value doesn't fit.
978 * \param[in,out] env execution environment for this thread
979 * .lti_ea_store buffer is filled with EA's value
980 * \param[in] lo LOD object
981 * \param[in] name name of the EA
983 * \retval 0 if EA is fetched successfully
984 * \retval negative error number on failure
986 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
989 struct lod_thread_info *info = lod_env_info(env);
990 struct dt_object *next = dt_object_child(&lo->ldo_obj);
996 if (unlikely(info->lti_ea_store == NULL)) {
997 /* just to enter in allocation block below */
1001 info->lti_buf.lb_buf = info->lti_ea_store;
1002 info->lti_buf.lb_len = info->lti_ea_store_size;
1003 rc = dt_xattr_get(env, next, &info->lti_buf, name);
1006 /* if object is not striped or inaccessible */
1007 if (rc == -ENODATA || rc == -ENOENT)
1010 if (rc == -ERANGE) {
1011 /* EA doesn't fit, reallocate new buffer */
1012 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
1013 if (rc == -ENODATA || rc == -ENOENT)
1019 rc = lod_ea_store_resize(info, rc);
1029 * Verify the target index is present in the current configuration.
1031 * \param[in] md LOD device where the target table is stored
1032 * \param[in] idx target's index
1034 * \retval 0 if the index is present
1035 * \retval -EINVAL if not
1037 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1039 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1040 !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
1041 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1042 md->lod_ost_descs.ltd_tgts_size);
1046 if (unlikely(OST_TGT(md, idx) == NULL)) {
1047 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1048 lod2obd(md)->obd_name, idx);
1052 if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
1053 CERROR("%s: invalid lod device, for idx: %d\n",
1054 lod2obd(md)->obd_name , idx);
1062 * Instantiate objects for stripes.
1064 * Allocate and initialize LU-objects representing the stripes. The number
1065 * of the stripes (ldo_stripenr) must be initialized already. The caller
1066 * must ensure nobody else is calling the function on the object at the same
1067 * time. FLDB service must be running to be able to map a FID to the targets
1068 * and find appropriate device representing that target.
1070 * \param[in] env execution environment for this thread
1071 * \param[in,out] lo LOD object
1072 * \param[in] objs an array of IDs to creates the objects from
1073 * \param[in] comp_idx index of ldo_comp_entries
1075 * \retval 0 if the objects are instantiated successfully
1076 * \retval negative error number on failure
1078 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1079 struct lov_ost_data_v1 *objs, int comp_idx)
1081 struct lod_layout_component *lod_comp;
1082 struct lod_thread_info *info = lod_env_info(env);
1083 struct lod_device *md;
1084 struct lu_object *o, *n;
1085 struct lu_device *nd;
1086 struct dt_object **stripe;
1092 LASSERT(lo != NULL);
1093 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1095 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1096 lod_comp = &lo->ldo_comp_entries[comp_idx];
1098 LASSERT(lod_comp->llc_stripe == NULL);
1099 LASSERT(lod_comp->llc_stripenr > 0);
1100 LASSERT(lod_comp->llc_stripe_size > 0);
1102 stripe_len = lod_comp->llc_stripenr;
1103 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1107 for (i = 0; i < lod_comp->llc_stripenr; i++) {
1108 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1111 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1112 idx = le32_to_cpu(objs[i].l_ost_idx);
1113 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1116 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1117 PFID(&info->lti_fid));
1118 lod_getref(&md->lod_ost_descs);
1120 rc = validate_lod_and_idx(md, idx);
1121 if (unlikely(rc != 0)) {
1122 lod_putref(md, &md->lod_ost_descs);
1126 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
1127 lod_putref(md, &md->lod_ost_descs);
1129 /* In the function below, .hs_keycmp resolves to
1130 * u_obj_hop_keycmp() */
1131 /* coverity[overrun-buffer-val] */
1132 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1134 GOTO(out, rc = PTR_ERR(o));
1136 n = lu_object_locate(o->lo_header, nd->ld_type);
1139 stripe[i] = container_of(n, struct dt_object, do_lu);
1144 for (i = 0; i < stripe_len; i++)
1145 if (stripe[i] != NULL)
1146 dt_object_put(env, stripe[i]);
1148 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1149 lod_comp->llc_stripenr = 0;
1151 lod_comp->llc_stripe = stripe;
1152 lod_comp->llc_stripes_allocated = stripe_len;
1159 * Instantiate objects for striping.
1161 * Parse striping information in \a buf and instantiate the objects
1162 * representing the stripes.
1164 * \param[in] env execution environment for this thread
1165 * \param[in] lo LOD object
1166 * \param[in] buf buffer storing LOV EA to parse
1168 * \retval 0 if parsing and objects creation succeed
1169 * \retval negative error number on failure
1171 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1172 const struct lu_buf *buf)
1174 struct lov_mds_md_v1 *lmm;
1175 struct lov_comp_md_v1 *comp_v1 = NULL;
1176 struct lov_ost_data_v1 *objs;
1177 __u32 magic, pattern;
1183 LASSERT(buf->lb_buf);
1184 LASSERT(buf->lb_len);
1186 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1187 magic = le32_to_cpu(lmm->lmm_magic);
1189 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1190 magic != LOV_MAGIC_COMP_V1)
1191 GOTO(out, rc = -EINVAL);
1193 lod_free_comp_entries(lo);
1195 if (magic == LOV_MAGIC_COMP_V1) {
1196 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1197 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1199 GOTO(out, rc = -EINVAL);
1200 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1201 lo->ldo_is_composite = 1;
1204 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1205 lo->ldo_is_composite = 0;
1208 rc = lod_alloc_comp_entries(lo, comp_cnt);
1212 for (i = 0; i < comp_cnt; i++) {
1213 struct lod_layout_component *lod_comp;
1214 struct lu_extent *ext;
1217 lod_comp = &lo->ldo_comp_entries[i];
1218 if (lo->ldo_is_composite) {
1219 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1220 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1221 magic = le32_to_cpu(lmm->lmm_magic);
1223 ext = &comp_v1->lcm_entries[i].lcme_extent;
1224 lod_comp->llc_extent.e_start =
1225 le64_to_cpu(ext->e_start);
1226 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1227 lod_comp->llc_flags =
1228 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1230 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1231 if (lod_comp->llc_id == LCME_ID_INVAL)
1232 GOTO(out, rc = -EINVAL);
1234 lod_comp->llc_flags = LCME_FL_INIT;
1237 pattern = le32_to_cpu(lmm->lmm_pattern);
1238 if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
1239 GOTO(out, rc = -EINVAL);
1241 lod_comp->llc_pattern = pattern;
1242 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1243 lod_comp->llc_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
1244 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1246 if (magic == LOV_MAGIC_V3) {
1247 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1248 objs = &v3->lmm_objects[0];
1249 /* no need to set pool, which is used in create only */
1251 objs = &lmm->lmm_objects[0];
1254 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)) {
1255 rc = lod_initialize_objects(env, lo, objs, i);
1262 lod_object_free_striping(env, lo);
1267 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1268 * is already cached.
1270 * \param[in] lo LOD object
1272 * \retval True if the striping is cached, otherwise
1275 static bool lod_striping_loaded(struct lod_object *lo)
1277 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1278 lo->ldo_comp_cached)
1281 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1282 if (lo->ldo_stripe != NULL)
1285 /* Never load LMV stripe for slaves of striped dir */
1286 if (lo->ldo_dir_slave_stripe)
1294 * Initialize the object representing the stripes.
1296 * Unless the stripes are initialized already, fetch LOV (for regular
1297 * objects) or LMV (for directory objects) EA and call lod_parse_striping()
1298 * to instantiate the objects representing the stripes. Caller should
1299 * hold the dt_write_lock(next).
1301 * \param[in] env execution environment for this thread
1302 * \param[in,out] lo LOD object
1304 * \retval 0 if parsing and object creation succeed
1305 * \retval negative error number on failure
1307 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
1309 struct lod_thread_info *info = lod_env_info(env);
1310 struct lu_buf *buf = &info->lti_buf;
1311 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1315 if (!dt_object_exists(next))
1318 if (lod_striping_loaded(lo))
1321 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1322 rc = lod_get_lov_ea(env, lo);
1326 * there is LOV EA (striping information) in this object
1327 * let's parse it and create in-core objects for the stripes
1329 buf->lb_buf = info->lti_ea_store;
1330 buf->lb_len = info->lti_ea_store_size;
1331 rc = lod_parse_striping(env, lo, buf);
1333 lo->ldo_comp_cached = 1;
1334 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1335 rc = lod_get_lmv_ea(env, lo);
1336 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1))
1337 GOTO(out, rc = rc > 0 ? -EINVAL : rc);
1339 buf->lb_buf = info->lti_ea_store;
1340 buf->lb_len = info->lti_ea_store_size;
1341 if (rc == sizeof(struct lmv_mds_md_v1)) {
1342 rc = lod_load_lmv_shards(env, lo, buf, true);
1343 if (buf->lb_buf != info->lti_ea_store) {
1344 OBD_FREE_LARGE(info->lti_ea_store,
1345 info->lti_ea_store_size);
1346 info->lti_ea_store = buf->lb_buf;
1347 info->lti_ea_store_size = buf->lb_len;
1355 * there is LMV EA (striping information) in this object
1356 * let's parse it and create in-core objects for the stripes
1358 rc = lod_parse_dir_striping(env, lo, buf);
1365 * A generic function to initialize the stripe objects.
1367 * A protected version of lod_load_striping_locked() - load the striping
1368 * information from storage, parse that and instantiate LU objects to
1369 * represent the stripes. The LOD object \a lo supplies a pointer to the
1370 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1371 * return an array of references to the newly instantiated objects.
1373 * \param[in] env execution environment for this thread
1374 * \param[in,out] lo LOD object, where striping is stored and
1375 * which gets an array of references
1377 * \retval 0 if parsing and object creation succeed
1378 * \retval negative error number on failure
1380 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
1382 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1385 if (!dt_object_exists(next))
1388 /* Check without locking first */
1389 if (lod_striping_loaded(lo))
1392 /* currently this code is supposed to be called from declaration
1393 * phase only, thus the object is not expected to be locked by caller */
1394 dt_write_lock(env, next, 0);
1395 rc = lod_load_striping_locked(env, lo);
1396 dt_write_unlock(env, next);
1401 * Verify lov_user_md_v1/v3 striping.
1403 * Check the validity of all fields including the magic, stripe size,
1404 * stripe count, stripe offset and that the pool is present. Also check
1405 * that each target index points to an existing target. The additional
1406 * \a is_from_disk turns additional checks. In some cases zero fields
1407 * are allowed (like pattern=0).
1409 * \param[in] d LOD device
1410 * \param[in] buf buffer with LOV EA to verify
1411 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1412 * 1 - from disk, do not allow
1414 * \retval 0 if the striping is valid
1415 * \retval -EINVAL if striping is invalid
1417 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1420 struct lov_user_md_v1 *lum;
1421 struct lov_user_md_v3 *lum3;
1422 struct pool_desc *pool = NULL;
1426 __u16 stripe_offset;
1433 if (buf->lb_len < sizeof(*lum)) {
1434 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1436 GOTO(out, rc = -EINVAL);
1439 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEF;
1440 if (magic != LOV_USER_MAGIC_V1 &&
1441 magic != LOV_USER_MAGIC_V3 &&
1442 magic != LOV_USER_MAGIC_SPECIFIC) {
1443 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1444 le32_to_cpu(lum->lmm_magic));
1445 GOTO(out, rc = -EINVAL);
1448 /* the user uses "0" for default stripe pattern normally. */
1449 if (!is_from_disk && lum->lmm_pattern == 0)
1450 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1452 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1453 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1454 le32_to_cpu(lum->lmm_pattern));
1455 GOTO(out, rc = -EINVAL);
1458 /* a released lum comes from creating orphan on hsm release,
1459 * doesn't make sense to verify it. */
1460 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1463 /* 64kB is the largest common page size we see (ia64), and matches the
1465 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1466 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1467 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1468 stripe_size, LOV_MIN_STRIPE_SIZE);
1469 GOTO(out, rc = -EINVAL);
1472 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1473 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1474 /* if offset is not within valid range [0, osts_size) */
1475 if (stripe_offset >= d->lod_osts_size) {
1476 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1477 stripe_offset, d->lod_osts_size);
1478 GOTO(out, rc = -EINVAL);
1481 /* if lmm_stripe_offset is *not* in bitmap */
1482 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1483 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1485 GOTO(out, rc = -EINVAL);
1489 if (magic == LOV_USER_MAGIC_V1)
1490 lum_size = offsetof(struct lov_user_md_v1,
1492 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1493 lum_size = offsetof(struct lov_user_md_v3,
1496 GOTO(out, rc = -EINVAL);
1498 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1499 if (buf->lb_len < lum_size) {
1500 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1501 "magic %#x and stripe_count %u\n",
1502 buf->lb_len, lum_size, magic, stripe_count);
1503 GOTO(out, rc = -EINVAL);
1506 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1510 /* In the function below, .hs_keycmp resolves to
1511 * pool_hashkey_keycmp() */
1512 /* coverity[overrun-buffer-val] */
1513 pool = lod_find_pool(d, lum3->lmm_pool_name);
1517 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1518 rc = lod_check_index_in_pool(stripe_offset, pool);
1520 GOTO(out, rc = -EINVAL);
1523 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1524 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1525 stripe_count, pool_tgt_count(pool));
1526 GOTO(out, rc = -EINVAL);
1531 lod_pool_putref(pool);
1537 * Verify LOV striping.
1539 * \param[in] d LOD device
1540 * \param[in] buf buffer with LOV EA to verify
1541 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1542 * 1 - from disk, do not allow
1543 * \param[in] start extent start for composite layout
1545 * \retval 0 if the striping is valid
1546 * \retval -EINVAL if striping is invalid
1548 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
1549 bool is_from_disk, __u64 start)
1551 struct lov_user_md_v1 *lum;
1552 struct lov_comp_md_v1 *comp_v1;
1559 if (buf->lb_len < sizeof(*lum)) {
1560 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1565 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEF;
1566 if (magic != LOV_USER_MAGIC_V1 &&
1567 magic != LOV_USER_MAGIC_V3 &&
1568 magic != LOV_USER_MAGIC_SPECIFIC &&
1569 magic != LOV_USER_MAGIC_COMP_V1) {
1570 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1571 le32_to_cpu(lum->lmm_magic));
1575 if (magic == LOV_USER_MAGIC_COMP_V1) {
1576 struct lov_comp_md_entry_v1 *ent;
1577 struct lu_extent *ext;
1578 struct lov_desc *desc = &d->lod_desc;
1580 __u32 stripe_size = 0;
1581 __u64 prev_end = start;
1583 comp_v1 = buf->lb_buf;
1584 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1585 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1586 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1590 if (le32_to_cpu(comp_v1->lcm_entry_count) == 0) {
1591 CDEBUG(D_LAYOUT, "entry count is zero\n");
1595 for (i = 0; i < le32_to_cpu(comp_v1->lcm_entry_count); i++) {
1596 ent = &comp_v1->lcm_entries[i];
1597 ext = &ent->lcme_extent;
1600 (le32_to_cpu(ent->lcme_id) == 0 ||
1601 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX)) {
1602 CDEBUG(D_LAYOUT, "invalid id %u\n",
1603 le32_to_cpu(ent->lcme_id));
1607 if (le64_to_cpu(ext->e_start) >=
1608 le64_to_cpu(ext->e_end)) {
1609 CDEBUG(D_LAYOUT, "invalid extent "
1611 le64_to_cpu(ext->e_start),
1612 le64_to_cpu(ext->e_end));
1616 /* first component must start with 0, and the next
1617 * must be adjacent with the previous one */
1618 if (le64_to_cpu(ext->e_start) != prev_end) {
1619 CDEBUG(D_LAYOUT, "invalid start "
1620 "actual:%llu, expect:%llu\n",
1621 le64_to_cpu(ext->e_start), prev_end);
1624 prev_end = le64_to_cpu(ext->e_end);
1626 tmp.lb_buf = (char *)comp_v1 +
1627 le32_to_cpu(ent->lcme_offset);
1628 tmp.lb_len = le32_to_cpu(ent->lcme_size);
1629 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
1635 /* extent end must be aligned with the stripe_size */
1636 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1637 if (stripe_size == 0)
1638 stripe_size = desc->ld_default_stripe_size;
1639 if (stripe_size == 0 ||
1640 (prev_end != LUSTRE_EOF &&
1641 (prev_end & (stripe_size - 1)))) {
1642 CDEBUG(D_LAYOUT, "stripe size isn't aligned. "
1643 " stripe_sz: %u, [%llu, %llu)\n",
1644 stripe_size, ext->e_start, prev_end);
1649 rc = lod_verify_v1v3(d, buf, is_from_disk);
1655 void lod_fix_desc_stripe_size(__u64 *val)
1657 if (*val < LOV_MIN_STRIPE_SIZE) {
1659 LCONSOLE_INFO("Increasing default stripe size to "
1660 "minimum value %u\n",
1661 LOV_DESC_STRIPE_SIZE_DEFAULT);
1662 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1663 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1664 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1665 LCONSOLE_WARN("Changing default stripe size to %llu (a "
1666 "multiple of %u)\n",
1667 *val, LOV_MIN_STRIPE_SIZE);
1671 void lod_fix_desc_stripe_count(__u32 *val)
1677 void lod_fix_desc_pattern(__u32 *val)
1679 /* from lov_setstripe */
1680 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1681 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1686 void lod_fix_desc_qos_maxage(__u32 *val)
1688 /* fix qos_maxage */
1690 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1694 * Used to fix insane default striping.
1696 * \param[in] desc striping description
1698 void lod_fix_desc(struct lov_desc *desc)
1700 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1701 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1702 lod_fix_desc_pattern(&desc->ld_pattern);
1703 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1707 * Initialize the structures used to store pools and default striping.
1709 * \param[in] lod LOD device
1710 * \param[in] lcfg configuration structure storing default striping.
1712 * \retval 0 if initialization succeeds
1713 * \retval negative error number on failure
1715 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1717 struct obd_device *obd;
1718 struct lov_desc *desc;
1722 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1723 LASSERT(obd != NULL);
1724 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1726 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1727 CERROR("LOD setup requires a descriptor\n");
1731 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1733 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1734 CERROR("descriptor size wrong: %d > %d\n",
1735 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1739 if (desc->ld_magic != LOV_DESC_MAGIC) {
1740 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1741 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1742 obd->obd_name, desc);
1743 lustre_swab_lov_desc(desc);
1745 CERROR("%s: Bad lov desc magic: %#x\n",
1746 obd->obd_name, desc->ld_magic);
1753 desc->ld_active_tgt_count = 0;
1754 lod->lod_desc = *desc;
1756 lod->lod_sp_me = LUSTRE_SP_CLI;
1758 /* Set up allocation policy (QoS and RR) */
1759 INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1760 init_rwsem(&lod->lod_qos.lq_rw_sem);
1761 lod->lod_qos.lq_dirty = 1;
1762 lod->lod_qos.lq_rr.lqr_dirty = 1;
1763 lod->lod_qos.lq_reset = 1;
1764 /* Default priority is toward free space balance */
1765 lod->lod_qos.lq_prio_free = 232;
1766 /* Default threshold for rr (roughly 17%) */
1767 lod->lod_qos.lq_threshold_rr = 43;
1769 /* Set up OST pool environment */
1770 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1771 HASH_POOLS_MAX_BITS,
1772 HASH_POOLS_BKT_BITS, 0,
1775 &pool_hash_operations,
1777 if (lod->lod_pools_hash_body == NULL)
1780 INIT_LIST_HEAD(&lod->lod_pool_list);
1781 lod->lod_pool_count = 0;
1782 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1785 lod_qos_rr_init(&lod->lod_qos.lq_rr);
1786 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1788 GOTO(out_pool_info, rc);
1793 lod_ost_pool_free(&lod->lod_pool_info);
1795 cfs_hash_putref(lod->lod_pools_hash_body);
1801 * Release the structures describing the pools.
1803 * \param[in] lod LOD device from which we release the structures
1807 int lod_pools_fini(struct lod_device *lod)
1809 struct obd_device *obd = lod2obd(lod);
1810 struct pool_desc *pool, *tmp;
1813 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
1814 /* free pool structs */
1815 CDEBUG(D_INFO, "delete pool %p\n", pool);
1816 /* In the function below, .hs_keycmp resolves to
1817 * pool_hashkey_keycmp() */
1818 /* coverity[overrun-buffer-val] */
1819 lod_pool_del(obd, pool->pool_name);
1822 cfs_hash_putref(lod->lod_pools_hash_body);
1823 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1824 lod_ost_pool_free(&lod->lod_pool_info);