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 if (entry->llc_ostlist.op_array)
631 OBD_FREE(entry->llc_ostlist.op_array,
632 entry->llc_ostlist.op_size);
633 LASSERT(entry->llc_stripe == NULL);
634 LASSERT(entry->llc_stripes_allocated == 0);
638 OBD_FREE_LARGE(entries, bufsize);
641 void lod_free_def_comp_entries(struct lod_default_striping *lds)
643 lod_free_comp_buffer(lds->lds_def_comp_entries,
644 lds->lds_def_comp_size_cnt,
646 sizeof(*lds->lds_def_comp_entries) *
647 lds->lds_def_comp_size_cnt));
648 lds->lds_def_comp_entries = NULL;
649 lds->lds_def_comp_cnt = 0;
650 lds->lds_def_striping_is_composite = 0;
651 lds->lds_def_comp_size_cnt = 0;
655 * Resize per-thread storage to hold default striping component entries
657 * A helper function to resize per-thread temporary storage. This storage
658 * is used to hold default LOV/LVM EAs and may be quite large. We do not want
659 * to allocate/release it every time, so instead we put it into the env and
660 * reallocate it on demand. The memory is released when the correspondent
661 * thread is finished.
663 * \param[in,out] lds default striping
664 * \param[in] count new component count to grow the buffer to
666 * \retval 0 on success, -ENOMEM if reallocation failed
668 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
670 struct lod_layout_component *entries;
671 __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
673 __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
674 lds->lds_def_comp_size_cnt);
679 OBD_ALLOC_LARGE(entries, new);
683 if (lds->lds_def_comp_entries != NULL) {
684 CDEBUG(D_INFO, "default striping component size %d is not "
685 "enough, need %d\n", old, new);
686 lod_free_def_comp_entries(lds);
689 lds->lds_def_comp_entries = entries;
690 lds->lds_def_comp_size_cnt = count;
695 void lod_free_comp_entries(struct lod_object *lo)
697 lod_free_comp_buffer(lo->ldo_comp_entries,
699 sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
700 lo->ldo_comp_entries = NULL;
701 lo->ldo_comp_cnt = 0;
702 lo->ldo_is_composite = 0;
705 int lod_alloc_comp_entries(struct lod_object *lo, int cnt)
708 LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
710 OBD_ALLOC_LARGE(lo->ldo_comp_entries,
711 sizeof(*lo->ldo_comp_entries) * cnt);
712 if (lo->ldo_comp_entries == NULL)
714 lo->ldo_comp_cnt = cnt;
719 * Generate on-disk lov_mds_md structure for each layout component based on
720 * the information in lod_object->ldo_comp_entries[i].
722 * \param[in] env execution environment for this thread
723 * \param[in] lo LOD object
724 * \param[in] comp_idx index of ldo_comp_entries
725 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
726 * \param[in|out] lmm_size buffer size/lmm size
727 * \param[in] is_dir generate lov ea for dir or file? For dir case,
728 * the stripe info is from the default stripe
729 * template, which is collected in lod_ah_init(),
730 * either from parent object or root object; for
731 * file case, it's from the @lo object
733 * \retval 0 if on disk structure is created successfully
734 * \retval negative error number on failure
736 static int lod_gen_component_ea(const struct lu_env *env,
737 struct lod_object *lo, int comp_idx,
738 struct lov_mds_md *lmm, int *lmm_size,
741 struct lod_thread_info *info = lod_env_info(env);
742 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
743 struct lod_device *lod;
744 struct lov_ost_data_v1 *objs;
745 struct lod_layout_component *lod_comp;
754 &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
756 lod_comp = &lo->ldo_comp_entries[comp_idx];
758 magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
759 if (lod_comp->llc_pattern == 0) /* default striping */
760 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
762 lmm->lmm_magic = cpu_to_le32(magic);
763 lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
764 fid_to_lmm_oi(fid, &lmm->lmm_oi);
765 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
766 lmm->lmm_oi.oi.oi_id++;
767 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
769 lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
770 lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
772 * for dir and uninstantiated component, lmm_layout_gen stores
773 * default stripe offset.
775 lmm->lmm_layout_gen =
776 (is_dir || !lod_comp_inited(lod_comp)) ?
777 cpu_to_le16(lod_comp->llc_stripe_offset) :
778 cpu_to_le16(lod_comp->llc_layout_gen);
780 if (magic == LOV_MAGIC_V1) {
781 objs = &lmm->lmm_objects[0];
783 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
784 size_t cplen = strlcpy(v3->lmm_pool_name,
786 sizeof(v3->lmm_pool_name));
787 if (cplen >= sizeof(v3->lmm_pool_name))
789 objs = &v3->lmm_objects[0];
791 stripe_count = lod_comp_entry_stripe_count(lo, lod_comp, is_dir);
792 if (!is_dir && lo->ldo_is_composite)
793 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
795 if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
798 /* generate ost_idx of this component stripe */
799 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
800 for (i = 0; i < stripe_count; i++) {
801 struct dt_object *object;
802 __u32 ost_idx = (__u32)-1UL;
803 int type = LU_SEQ_RANGE_OST;
805 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
806 object = lod_comp->llc_stripe[i];
807 /* instantiated component */
808 info->lti_fid = *lu_object_fid(&object->do_lu);
810 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
812 if (cfs_fail_val == 0)
813 cfs_fail_val = info->lti_fid.f_oid;
815 info->lti_fid.f_oid = cfs_fail_val;
818 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
821 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
822 objs[i].l_ost_gen = cpu_to_le32(0);
823 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
826 rc = lod_fld_lookup(env, lod, &info->lti_fid,
829 CERROR("%s: Can not locate "DFID": rc = %d\n",
830 lod2obd(lod)->obd_name,
831 PFID(&info->lti_fid), rc);
834 } else if (lod_comp->llc_ostlist.op_array) {
835 /* user specified ost list */
836 ost_idx = lod_comp->llc_ostlist.op_array[i];
839 * with un-instantiated or with no specified ost list
840 * component, its l_ost_idx does not matter.
842 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
845 if (lmm_size != NULL)
846 *lmm_size = lov_mds_md_size(stripe_count, magic);
851 * Generate component ID for new created component.
853 * \param[in] lo LOD object
854 * \param[in] comp_idx index of ldo_comp_entries
856 * \retval component ID on success
857 * \retval LCME_ID_INVAL on failure
859 static __u32 lod_gen_component_id(struct lod_object *lo, int comp_idx)
861 struct lod_layout_component *lod_comp;
862 __u32 id, start, end;
865 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
867 lod_obj_inc_layout_gen(lo);
868 id = lo->ldo_layout_gen;
869 if (likely(id <= LCME_ID_MAX))
872 /* Layout generation wraps, need to check collisions. */
873 start = id & LCME_ID_MASK;
876 for (id = start; id <= end; id++) {
877 for (i = 0; i < lo->ldo_comp_cnt; i++) {
878 lod_comp = &lo->ldo_comp_entries[i];
879 if (id == lod_comp->llc_id)
882 /* Found the ununsed ID */
883 if (i == lo->ldo_comp_cnt)
886 if (end == LCME_ID_MAX) {
888 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
889 (__u32)(LCME_ID_MAX - 1));
893 return LCME_ID_INVAL;
897 * Generate on-disk lov_mds_md structure based on the information in
898 * the lod_object->ldo_comp_entries.
900 * \param[in] env execution environment for this thread
901 * \param[in] lo LOD object
902 * \param[in] lmm buffer to cotain the on-disk lov_mds_md
903 * \param[in|out] lmm_size buffer size/lmm size
904 * \param[in] is_dir generate lov ea for dir or file? For dir case,
905 * the stripe info is from the default stripe
906 * template, which is collected in lod_ah_init(),
907 * either from parent object or root object; for
908 * file case, it's from the @lo object
910 * \retval 0 if on disk structure is created successfully
911 * \retval negative error number on failure
913 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
914 struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
916 struct lov_comp_md_entry_v1 *lcme;
917 struct lov_comp_md_v1 *lcm;
918 struct lod_layout_component *comp_entries;
921 int i, rc = 0, offset;
925 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
926 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
928 lo->ldo_def_striping->lds_def_striping_is_composite;
930 comp_cnt = lo->ldo_comp_cnt;
931 comp_entries = lo->ldo_comp_entries;
932 is_composite = lo->ldo_is_composite;
935 LASSERT(lmm_size != NULL);
936 LASSERT(comp_cnt != 0 && comp_entries != NULL);
939 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
943 lcm = (struct lov_comp_md_v1 *)lmm;
944 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
945 lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
947 offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
948 LASSERT(offset % sizeof(__u64) == 0);
950 for (i = 0; i < comp_cnt; i++) {
951 struct lod_layout_component *lod_comp;
952 struct lov_mds_md *sub_md;
955 lod_comp = &comp_entries[i];
956 lcme = &lcm->lcm_entries[i];
958 if (lod_comp->llc_id == LCME_ID_INVAL && !is_dir) {
959 lod_comp->llc_id = lod_gen_component_id(lo, i);
960 if (lod_comp->llc_id == LCME_ID_INVAL)
961 GOTO(out, rc = -ERANGE);
963 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
965 /* component could be un-inistantiated */
966 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
967 lcme->lcme_extent.e_start =
968 cpu_to_le64(lod_comp->llc_extent.e_start);
969 lcme->lcme_extent.e_end =
970 cpu_to_le64(lod_comp->llc_extent.e_end);
971 lcme->lcme_offset = cpu_to_le32(offset);
973 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
974 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
977 lcme->lcme_size = cpu_to_le32(size);
979 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
980 "offset:%d lmm_size:%d\n", offset, *lmm_size);
982 lcm->lcm_size = cpu_to_le32(offset);
983 lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
985 lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
996 * Fill lti_ea_store buffer in the environment with a value for the given
997 * EA. The buffer is reallocated if the value doesn't fit.
999 * \param[in,out] env execution environment for this thread
1000 * .lti_ea_store buffer is filled with EA's value
1001 * \param[in] lo LOD object
1002 * \param[in] name name of the EA
1004 * \retval > 0 if EA is fetched successfully
1005 * \retval 0 if EA is empty
1006 * \retval negative error number on failure
1008 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
1011 struct lod_thread_info *info = lod_env_info(env);
1012 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1018 if (unlikely(info->lti_ea_store == NULL)) {
1019 /* just to enter in allocation block below */
1023 info->lti_buf.lb_buf = info->lti_ea_store;
1024 info->lti_buf.lb_len = info->lti_ea_store_size;
1025 rc = dt_xattr_get(env, next, &info->lti_buf, name);
1028 /* if object is not striped or inaccessible */
1029 if (rc == -ENODATA || rc == -ENOENT)
1032 if (rc == -ERANGE) {
1033 /* EA doesn't fit, reallocate new buffer */
1034 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
1035 if (rc == -ENODATA || rc == -ENOENT)
1041 rc = lod_ea_store_resize(info, rc);
1051 * Verify the target index is present in the current configuration.
1053 * \param[in] md LOD device where the target table is stored
1054 * \param[in] idx target's index
1056 * \retval 0 if the index is present
1057 * \retval -EINVAL if not
1059 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1061 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1062 !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
1063 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1064 md->lod_ost_descs.ltd_tgts_size);
1068 if (unlikely(OST_TGT(md, idx) == NULL)) {
1069 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1070 lod2obd(md)->obd_name, idx);
1074 if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
1075 CERROR("%s: invalid lod device, for idx: %d\n",
1076 lod2obd(md)->obd_name , idx);
1084 * Instantiate objects for stripes.
1086 * Allocate and initialize LU-objects representing the stripes. The number
1087 * of the stripes (ldo_stripe_count) must be initialized already. The caller
1088 * must ensure nobody else is calling the function on the object at the same
1089 * time. FLDB service must be running to be able to map a FID to the targets
1090 * and find appropriate device representing that target.
1092 * \param[in] env execution environment for this thread
1093 * \param[in,out] lo LOD object
1094 * \param[in] objs an array of IDs to creates the objects from
1095 * \param[in] comp_idx index of ldo_comp_entries
1097 * \retval 0 if the objects are instantiated successfully
1098 * \retval negative error number on failure
1100 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1101 struct lov_ost_data_v1 *objs, int comp_idx)
1103 struct lod_layout_component *lod_comp;
1104 struct lod_thread_info *info = lod_env_info(env);
1105 struct lod_device *md;
1106 struct lu_object *o, *n;
1107 struct lu_device *nd;
1108 struct dt_object **stripe;
1114 LASSERT(lo != NULL);
1115 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1117 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1118 lod_comp = &lo->ldo_comp_entries[comp_idx];
1120 LASSERT(lod_comp->llc_stripe == NULL);
1121 LASSERT(lod_comp->llc_stripe_count > 0);
1122 LASSERT(lod_comp->llc_stripe_size > 0);
1124 stripe_len = lod_comp->llc_stripe_count;
1125 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1129 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1130 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1133 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1134 idx = le32_to_cpu(objs[i].l_ost_idx);
1135 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1138 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1139 PFID(&info->lti_fid));
1140 lod_getref(&md->lod_ost_descs);
1142 rc = validate_lod_and_idx(md, idx);
1143 if (unlikely(rc != 0)) {
1144 lod_putref(md, &md->lod_ost_descs);
1148 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
1149 lod_putref(md, &md->lod_ost_descs);
1151 /* In the function below, .hs_keycmp resolves to
1152 * u_obj_hop_keycmp() */
1153 /* coverity[overrun-buffer-val] */
1154 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1156 GOTO(out, rc = PTR_ERR(o));
1158 n = lu_object_locate(o->lo_header, nd->ld_type);
1161 stripe[i] = container_of(n, struct dt_object, do_lu);
1166 for (i = 0; i < stripe_len; i++)
1167 if (stripe[i] != NULL)
1168 dt_object_put(env, stripe[i]);
1170 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1171 lod_comp->llc_stripe_count = 0;
1173 lod_comp->llc_stripe = stripe;
1174 lod_comp->llc_stripes_allocated = stripe_len;
1181 * Instantiate objects for striping.
1183 * Parse striping information in \a buf and instantiate the objects
1184 * representing the stripes.
1186 * \param[in] env execution environment for this thread
1187 * \param[in] lo LOD object
1188 * \param[in] buf buffer storing LOV EA to parse
1190 * \retval 0 if parsing and objects creation succeed
1191 * \retval negative error number on failure
1193 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1194 const struct lu_buf *buf)
1196 struct lov_mds_md_v1 *lmm;
1197 struct lov_comp_md_v1 *comp_v1 = NULL;
1198 struct lov_ost_data_v1 *objs;
1199 __u32 magic, pattern;
1205 LASSERT(buf->lb_buf);
1206 LASSERT(buf->lb_len);
1208 lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1209 magic = le32_to_cpu(lmm->lmm_magic);
1211 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1212 magic != LOV_MAGIC_COMP_V1)
1213 GOTO(out, rc = -EINVAL);
1215 lod_free_comp_entries(lo);
1217 if (magic == LOV_MAGIC_COMP_V1) {
1218 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1219 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1221 GOTO(out, rc = -EINVAL);
1222 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1223 lo->ldo_is_composite = 1;
1226 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1227 lo->ldo_is_composite = 0;
1230 rc = lod_alloc_comp_entries(lo, comp_cnt);
1234 for (i = 0; i < comp_cnt; i++) {
1235 struct lod_layout_component *lod_comp;
1236 struct lu_extent *ext;
1239 lod_comp = &lo->ldo_comp_entries[i];
1240 if (lo->ldo_is_composite) {
1241 offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1242 lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1243 magic = le32_to_cpu(lmm->lmm_magic);
1245 ext = &comp_v1->lcm_entries[i].lcme_extent;
1246 lod_comp->llc_extent.e_start =
1247 le64_to_cpu(ext->e_start);
1248 lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1249 lod_comp->llc_flags =
1250 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1252 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1253 if (lod_comp->llc_id == LCME_ID_INVAL)
1254 GOTO(out, rc = -EINVAL);
1256 lod_comp_set_init(lod_comp);
1259 pattern = le32_to_cpu(lmm->lmm_pattern);
1260 if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
1261 GOTO(out, rc = -EINVAL);
1263 lod_comp->llc_pattern = pattern;
1264 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1265 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1266 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1268 if (magic == LOV_MAGIC_V3) {
1269 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1270 objs = &v3->lmm_objects[0];
1271 /* no need to set pool, which is used in create only */
1273 objs = &lmm->lmm_objects[0];
1277 * If uninstantiated template component has valid l_ost_idx,
1278 * then user has specified ost list for this component.
1280 if (!lod_comp_inited(lod_comp)) {
1281 if (objs[0].l_ost_idx != (__u32)-1UL) {
1283 * load the user specified ost list, when this
1284 * component is instantiated later, it will be
1285 * used in lod_alloc_ost_list().
1287 lod_comp->llc_ostlist.op_count =
1288 lod_comp->llc_stripe_count;
1289 lod_comp->llc_ostlist.op_size =
1290 lod_comp->llc_stripe_count *
1292 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1293 lod_comp->llc_ostlist.op_size);
1294 if (!lod_comp->llc_ostlist.op_array)
1295 GOTO(out, rc = -ENOMEM);
1297 for (j = 0; j < lod_comp->llc_stripe_count; j++)
1298 lod_comp->llc_ostlist.op_array[j] =
1299 le32_to_cpu(objs[j].l_ost_idx);
1302 * this component OST objects starts from the
1303 * first ost_idx, lod_alloc_ost_list() will
1306 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1309 * for uninstantiated component,
1310 * lmm_layout_gen stores default stripe offset.
1312 lod_comp->llc_stripe_offset =
1313 lmm->lmm_layout_gen;
1317 /* skip un-instantiated component object initialization */
1318 if (!lod_comp_inited(lod_comp))
1321 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)) {
1322 rc = lod_initialize_objects(env, lo, objs, i);
1329 lod_object_free_striping(env, lo);
1334 * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1335 * is already cached.
1337 * \param[in] lo LOD object
1339 * \retval True if the striping is cached, otherwise
1342 static bool lod_striping_loaded(struct lod_object *lo)
1344 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1345 lo->ldo_comp_cached)
1348 if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1349 if (lo->ldo_stripe != NULL || lo->ldo_dir_stripe_loaded)
1352 /* Never load LMV stripe for slaves of striped dir */
1353 if (lo->ldo_dir_slave_stripe)
1361 * Initialize the object representing the stripes.
1363 * Unless the stripes are initialized already, fetch LOV (for regular
1364 * objects) or LMV (for directory objects) EA and call lod_parse_striping()
1365 * to instantiate the objects representing the stripes. Caller should
1366 * hold the dt_write_lock(next).
1368 * \param[in] env execution environment for this thread
1369 * \param[in,out] lo LOD object
1371 * \retval 0 if parsing and object creation succeed
1372 * \retval negative error number on failure
1374 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
1376 struct lod_thread_info *info = lod_env_info(env);
1377 struct lu_buf *buf = &info->lti_buf;
1378 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1382 if (!dt_object_exists(next))
1385 if (lod_striping_loaded(lo))
1388 if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1389 rc = lod_get_lov_ea(env, lo);
1393 * there is LOV EA (striping information) in this object
1394 * let's parse it and create in-core objects for the stripes
1396 buf->lb_buf = info->lti_ea_store;
1397 buf->lb_len = info->lti_ea_store_size;
1398 rc = lod_parse_striping(env, lo, buf);
1400 lo->ldo_comp_cached = 1;
1401 } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1402 rc = lod_get_lmv_ea(env, lo);
1403 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1)) {
1404 /* Let's set stripe_loaded to avoid further
1405 * stripe loading especially for non-stripe directory,
1406 * which can hurt performance. (See LU-9840)
1409 lo->ldo_dir_stripe_loaded = 1;
1410 GOTO(out, rc = rc > 0 ? -EINVAL : rc);
1412 buf->lb_buf = info->lti_ea_store;
1413 buf->lb_len = info->lti_ea_store_size;
1414 if (rc == sizeof(struct lmv_mds_md_v1)) {
1415 rc = lod_load_lmv_shards(env, lo, buf, true);
1416 if (buf->lb_buf != info->lti_ea_store) {
1417 OBD_FREE_LARGE(info->lti_ea_store,
1418 info->lti_ea_store_size);
1419 info->lti_ea_store = buf->lb_buf;
1420 info->lti_ea_store_size = buf->lb_len;
1428 * there is LMV EA (striping information) in this object
1429 * let's parse it and create in-core objects for the stripes
1431 rc = lod_parse_dir_striping(env, lo, buf);
1433 lo->ldo_dir_stripe_loaded = 1;
1440 * A generic function to initialize the stripe objects.
1442 * A protected version of lod_load_striping_locked() - load the striping
1443 * information from storage, parse that and instantiate LU objects to
1444 * represent the stripes. The LOD object \a lo supplies a pointer to the
1445 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1446 * return an array of references to the newly instantiated objects.
1448 * \param[in] env execution environment for this thread
1449 * \param[in,out] lo LOD object, where striping is stored and
1450 * which gets an array of references
1452 * \retval 0 if parsing and object creation succeed
1453 * \retval negative error number on failure
1455 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
1457 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1460 if (!dt_object_exists(next))
1463 /* Check without locking first */
1464 if (lod_striping_loaded(lo))
1467 /* currently this code is supposed to be called from declaration
1468 * phase only, thus the object is not expected to be locked by caller */
1469 dt_write_lock(env, next, 0);
1470 rc = lod_load_striping_locked(env, lo);
1471 dt_write_unlock(env, next);
1476 * Verify lov_user_md_v1/v3 striping.
1478 * Check the validity of all fields including the magic, stripe size,
1479 * stripe count, stripe offset and that the pool is present. Also check
1480 * that each target index points to an existing target. The additional
1481 * \a is_from_disk turns additional checks. In some cases zero fields
1482 * are allowed (like pattern=0).
1484 * \param[in] d LOD device
1485 * \param[in] buf buffer with LOV EA to verify
1486 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1487 * 1 - from disk, do not allow
1489 * \retval 0 if the striping is valid
1490 * \retval -EINVAL if striping is invalid
1492 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1495 struct lov_user_md_v1 *lum;
1496 struct lov_user_md_v3 *lum3;
1497 struct pool_desc *pool = NULL;
1501 __u16 stripe_offset;
1508 if (buf->lb_len < sizeof(*lum)) {
1509 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1511 GOTO(out, rc = -EINVAL);
1514 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEF;
1515 if (magic != LOV_USER_MAGIC_V1 &&
1516 magic != LOV_USER_MAGIC_V3 &&
1517 magic != LOV_USER_MAGIC_SPECIFIC) {
1518 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1519 le32_to_cpu(lum->lmm_magic));
1520 GOTO(out, rc = -EINVAL);
1523 /* the user uses "0" for default stripe pattern normally. */
1524 if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1525 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1527 if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1528 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1529 le32_to_cpu(lum->lmm_pattern));
1530 GOTO(out, rc = -EINVAL);
1533 /* a released lum comes from creating orphan on hsm release,
1534 * doesn't make sense to verify it. */
1535 if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1538 /* 64kB is the largest common page size we see (ia64), and matches the
1540 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1541 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1542 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1543 stripe_size, LOV_MIN_STRIPE_SIZE);
1544 GOTO(out, rc = -EINVAL);
1547 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1548 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1549 /* if offset is not within valid range [0, osts_size) */
1550 if (stripe_offset >= d->lod_osts_size) {
1551 CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1552 stripe_offset, d->lod_osts_size);
1553 GOTO(out, rc = -EINVAL);
1556 /* if lmm_stripe_offset is *not* in bitmap */
1557 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1558 CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1560 GOTO(out, rc = -EINVAL);
1564 if (magic == LOV_USER_MAGIC_V1)
1565 lum_size = offsetof(struct lov_user_md_v1,
1567 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1568 lum_size = offsetof(struct lov_user_md_v3,
1571 GOTO(out, rc = -EINVAL);
1573 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1574 if (buf->lb_len < lum_size) {
1575 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1576 "magic %#x and stripe_count %u\n",
1577 buf->lb_len, lum_size, magic, stripe_count);
1578 GOTO(out, rc = -EINVAL);
1581 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1585 /* In the function below, .hs_keycmp resolves to
1586 * pool_hashkey_keycmp() */
1587 /* coverity[overrun-buffer-val] */
1588 pool = lod_find_pool(d, lum3->lmm_pool_name);
1592 if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1593 rc = lod_check_index_in_pool(stripe_offset, pool);
1595 GOTO(out, rc = -EINVAL);
1598 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1599 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1600 stripe_count, pool_tgt_count(pool));
1601 GOTO(out, rc = -EINVAL);
1606 lod_pool_putref(pool);
1612 * Verify LOV striping.
1614 * \param[in] d LOD device
1615 * \param[in] buf buffer with LOV EA to verify
1616 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1617 * 1 - from disk, do not allow
1618 * \param[in] start extent start for composite layout
1620 * \retval 0 if the striping is valid
1621 * \retval -EINVAL if striping is invalid
1623 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
1624 bool is_from_disk, __u64 start)
1626 struct lov_user_md_v1 *lum;
1627 struct lov_comp_md_v1 *comp_v1;
1634 if (buf->lb_len < sizeof(*lum)) {
1635 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1640 magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEF;
1641 if (magic != LOV_USER_MAGIC_V1 &&
1642 magic != LOV_USER_MAGIC_V3 &&
1643 magic != LOV_USER_MAGIC_SPECIFIC &&
1644 magic != LOV_USER_MAGIC_COMP_V1) {
1645 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1646 le32_to_cpu(lum->lmm_magic));
1650 if (magic == LOV_USER_MAGIC_COMP_V1) {
1651 struct lov_comp_md_entry_v1 *ent;
1652 struct lu_extent *ext;
1653 struct lov_desc *desc = &d->lod_desc;
1655 __u32 stripe_size = 0;
1656 __u64 prev_end = start;
1658 comp_v1 = buf->lb_buf;
1659 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1660 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1661 buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1665 if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1666 CDEBUG(D_LAYOUT, "entry count is zero\n");
1670 for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
1671 ent = &comp_v1->lcm_entries[i];
1672 ext = &ent->lcme_extent;
1675 (le32_to_cpu(ent->lcme_id) == 0 ||
1676 le32_to_cpu(ent->lcme_id) > LCME_ID_MAX)) {
1677 CDEBUG(D_LAYOUT, "invalid id %u\n",
1678 le32_to_cpu(ent->lcme_id));
1682 if (le64_to_cpu(ext->e_start) >=
1683 le64_to_cpu(ext->e_end)) {
1684 CDEBUG(D_LAYOUT, "invalid extent "
1686 le64_to_cpu(ext->e_start),
1687 le64_to_cpu(ext->e_end));
1691 /* first component must start with 0, and the next
1692 * must be adjacent with the previous one */
1693 if (le64_to_cpu(ext->e_start) != prev_end) {
1694 CDEBUG(D_LAYOUT, "invalid start "
1695 "actual:%llu, expect:%llu\n",
1696 le64_to_cpu(ext->e_start), prev_end);
1699 prev_end = le64_to_cpu(ext->e_end);
1701 tmp.lb_buf = (char *)comp_v1 +
1702 le32_to_cpu(ent->lcme_offset);
1703 tmp.lb_len = le32_to_cpu(ent->lcme_size);
1704 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
1710 /* extent end must be aligned with the stripe_size */
1711 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1712 if (stripe_size == 0)
1713 stripe_size = desc->ld_default_stripe_size;
1714 if (stripe_size == 0 ||
1715 (prev_end != LUSTRE_EOF &&
1716 (prev_end & (stripe_size - 1)))) {
1717 CDEBUG(D_LAYOUT, "stripe size isn't aligned. "
1718 " stripe_sz: %u, [%llu, %llu)\n",
1719 stripe_size, ext->e_start, prev_end);
1724 rc = lod_verify_v1v3(d, buf, is_from_disk);
1731 * set the default stripe size, if unset.
1733 * \param[in,out] val number of bytes per OST stripe
1735 * The minimum stripe size is 64KB to ensure that a single stripe is an
1736 * even multiple of a client PAGE_SIZE (IA64, PPC, etc). Otherwise, it
1737 * is difficult to split dirty pages across OSCs during writes.
1739 void lod_fix_desc_stripe_size(__u64 *val)
1741 if (*val < LOV_MIN_STRIPE_SIZE) {
1743 LCONSOLE_INFO("Increasing default stripe size to "
1744 "minimum value %u\n",
1745 LOV_DESC_STRIPE_SIZE_DEFAULT);
1746 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1747 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1748 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1749 LCONSOLE_WARN("Changing default stripe size to %llu (a "
1750 "multiple of %u)\n",
1751 *val, LOV_MIN_STRIPE_SIZE);
1756 * set the filesystem default number of stripes, if unset.
1758 * \param[in,out] val number of stripes
1760 * A value of "0" means "use the system-wide default stripe count", which
1761 * has either been inherited by now, or falls back to 1 stripe per file.
1762 * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
1763 * and is a valid value, so is left unchanged here.
1765 void lod_fix_desc_stripe_count(__u32 *val)
1772 * set the filesystem default layout pattern
1774 * \param[in,out] val LOV_PATTERN_* layout
1776 * A value of "0" means "use the system-wide default layout type", which
1777 * has either been inherited by now, or falls back to plain RAID0 striping.
1779 void lod_fix_desc_pattern(__u32 *val)
1781 /* from lov_setstripe */
1782 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1783 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1788 void lod_fix_desc_qos_maxage(__u32 *val)
1790 /* fix qos_maxage */
1792 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1796 * Used to fix insane default striping.
1798 * \param[in] desc striping description
1800 void lod_fix_desc(struct lov_desc *desc)
1802 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1803 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1804 lod_fix_desc_pattern(&desc->ld_pattern);
1805 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1809 * Initialize the structures used to store pools and default striping.
1811 * \param[in] lod LOD device
1812 * \param[in] lcfg configuration structure storing default striping.
1814 * \retval 0 if initialization succeeds
1815 * \retval negative error number on failure
1817 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1819 struct obd_device *obd;
1820 struct lov_desc *desc;
1824 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1825 LASSERT(obd != NULL);
1826 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1828 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1829 CERROR("LOD setup requires a descriptor\n");
1833 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1835 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1836 CERROR("descriptor size wrong: %d > %d\n",
1837 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1841 if (desc->ld_magic != LOV_DESC_MAGIC) {
1842 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1843 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1844 obd->obd_name, desc);
1845 lustre_swab_lov_desc(desc);
1847 CERROR("%s: Bad lov desc magic: %#x\n",
1848 obd->obd_name, desc->ld_magic);
1855 desc->ld_active_tgt_count = 0;
1856 lod->lod_desc = *desc;
1858 lod->lod_sp_me = LUSTRE_SP_CLI;
1860 /* Set up allocation policy (QoS and RR) */
1861 INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1862 init_rwsem(&lod->lod_qos.lq_rw_sem);
1863 lod->lod_qos.lq_dirty = 1;
1864 lod->lod_qos.lq_rr.lqr_dirty = 1;
1865 lod->lod_qos.lq_reset = 1;
1866 /* Default priority is toward free space balance */
1867 lod->lod_qos.lq_prio_free = 232;
1868 /* Default threshold for rr (roughly 17%) */
1869 lod->lod_qos.lq_threshold_rr = 43;
1871 /* Set up OST pool environment */
1872 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1873 HASH_POOLS_MAX_BITS,
1874 HASH_POOLS_BKT_BITS, 0,
1877 &pool_hash_operations,
1879 if (lod->lod_pools_hash_body == NULL)
1882 INIT_LIST_HEAD(&lod->lod_pool_list);
1883 lod->lod_pool_count = 0;
1884 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1887 lod_qos_rr_init(&lod->lod_qos.lq_rr);
1888 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1890 GOTO(out_pool_info, rc);
1895 lod_ost_pool_free(&lod->lod_pool_info);
1897 cfs_hash_putref(lod->lod_pools_hash_body);
1903 * Release the structures describing the pools.
1905 * \param[in] lod LOD device from which we release the structures
1909 int lod_pools_fini(struct lod_device *lod)
1911 struct obd_device *obd = lod2obd(lod);
1912 struct pool_desc *pool, *tmp;
1915 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
1916 /* free pool structs */
1917 CDEBUG(D_INFO, "delete pool %p\n", pool);
1918 /* In the function below, .hs_keycmp resolves to
1919 * pool_hashkey_keycmp() */
1920 /* coverity[overrun-buffer-val] */
1921 lod_pool_del(obd, pool->pool_name);
1924 cfs_hash_putref(lod->lod_pools_hash_body);
1925 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1926 lod_ost_pool_free(&lod->lod_pool_info);