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 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2013, Intel Corporation.
29 * lustre/lod/lod_lov.c
31 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
34 #define DEBUG_SUBSYSTEM S_MDS
36 #include <obd_class.h>
37 #include <lustre_lfsck.h>
38 #include <lustre_lmv.h>
40 #include "lod_internal.h"
43 * Keep a refcount of lod->ltd_tgts usage to prevent racing with
44 * addition/deletion. Any function that expects lov_tgts to remain stationary
47 * \param lod - is the lod device from which we want to grab a reference
49 void lod_getref(struct lod_tgt_descs *ltd)
51 down_read(<d->ltd_rw_sem);
52 mutex_lock(<d->ltd_mutex);
54 mutex_unlock(<d->ltd_mutex);
58 * Companion of lod_getref() to release a reference on the lod table.
59 * If this is the last reference and the ost entry was scheduled for deletion,
60 * the descriptor is removed from the array.
62 * \param lod - is the lod device from which we release a reference
64 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
66 mutex_lock(<d->ltd_mutex);
68 if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
69 struct lod_tgt_desc *tgt_desc, *tmp;
70 struct list_head kill;
73 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
76 INIT_LIST_HEAD(&kill);
78 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
79 tgt_desc = LTD_TGT(ltd, idx);
82 if (!tgt_desc->ltd_reap)
85 list_add(&tgt_desc->ltd_kill, &kill);
86 LTD_TGT(ltd, idx) = NULL;
87 /*FIXME: only support ost pool for now */
88 if (ltd == &lod->lod_ost_descs) {
89 lod_ost_pool_remove(&lod->lod_pool_info, idx);
90 if (tgt_desc->ltd_active)
91 lod->lod_desc.ld_active_tgt_count--;
94 cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
97 mutex_unlock(<d->ltd_mutex);
98 up_read(<d->ltd_rw_sem);
100 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
102 list_del(&tgt_desc->ltd_kill);
103 if (ltd == &lod->lod_ost_descs) {
104 /* remove from QoS structures */
105 rc = qos_del_tgt(lod, tgt_desc);
107 CERROR("%s: qos_del_tgt(%s) failed:"
109 lod2obd(lod)->obd_name,
110 obd_uuid2str(&tgt_desc->ltd_uuid),
113 rc = obd_disconnect(tgt_desc->ltd_exp);
115 CERROR("%s: failed to disconnect %s: rc = %d\n",
116 lod2obd(lod)->obd_name,
117 obd_uuid2str(&tgt_desc->ltd_uuid), rc);
118 OBD_FREE_PTR(tgt_desc);
121 mutex_unlock(<d->ltd_mutex);
122 up_read(<d->ltd_rw_sem);
126 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
128 cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
132 /* grab write reference on the lod. Relocating the array requires
133 * exclusive access */
135 down_write(<d->ltd_rw_sem);
136 if (newsize <= ltd->ltd_tgts_size)
137 /* someone else has already resize the array */
140 /* allocate new bitmap */
141 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
143 GOTO(out, rc = -ENOMEM);
145 if (ltd->ltd_tgts_size > 0) {
146 /* the bitmap already exists, we need
147 * to copy data from old one */
148 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
149 old_bitmap = ltd->ltd_tgt_bitmap;
152 ltd->ltd_tgts_size = newsize;
153 ltd->ltd_tgt_bitmap = new_bitmap;
156 CFS_FREE_BITMAP(old_bitmap);
158 CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
162 up_write(<d->ltd_rw_sem);
167 * Connect LOD to a new OSP and add it to the device table.
169 * \param env - is the environment passed by the caller
170 * \param lod - is the LOD device to be connected to the new OSP
171 * \param osp - is the name of OSP device name about to be added
172 * \param index - is the OSP index
173 * \param gen - is the generation number
174 * \param tgt_index - is the group of the OSP.
175 * \param type - is the type of device (mdc or osc)
177 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
178 char *osp, unsigned index, unsigned gen, int tgt_index,
179 char *type, int active)
181 struct obd_connect_data *data = NULL;
182 struct obd_export *exp = NULL;
183 struct obd_device *obd;
184 struct lu_device *ldev;
187 struct lod_tgt_desc *tgt_desc;
188 struct lod_tgt_descs *ltd;
189 struct obd_uuid obd_uuid;
193 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
196 CERROR("request to add OBD %s with invalid generation: %d\n",
201 obd_str2uuid(&obd_uuid, osp);
203 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
204 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
206 CERROR("can't find %s device\n", osp);
214 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
215 data->ocd_version = LUSTRE_VERSION_CODE;
216 data->ocd_index = index;
218 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
220 data->ocd_connect_flags |= OBD_CONNECT_AT |
223 #ifdef HAVE_LRU_RESIZE_SUPPORT
224 OBD_CONNECT_LRU_RESIZE |
227 OBD_CONNECT_OSS_CAPA |
228 OBD_CONNECT_REQPORTAL |
229 OBD_CONNECT_SKIP_ORPHAN |
231 OBD_CONNECT_LVB_TYPE |
232 OBD_CONNECT_VERSION |
233 OBD_CONNECT_PINGLESS |
236 data->ocd_group = tgt_index;
237 ltd = &lod->lod_ost_descs;
239 struct obd_import *imp = obd->u.cli.cl_import;
242 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
243 data->ocd_connect_flags |= OBD_CONNECT_ACL |
244 OBD_CONNECT_MDS_CAPA |
245 OBD_CONNECT_OSS_CAPA |
247 OBD_CONNECT_MDS_MDS |
252 /* XXX set MDS-MDS flags, remove this when running this
254 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
255 spin_lock(&imp->imp_lock);
256 imp->imp_server_timeout = 1;
257 spin_unlock(&imp->imp_lock);
258 imp->imp_client->cli_request_portal = OUT_PORTAL;
259 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
261 ltd = &lod->lod_mdt_descs;
264 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
267 CERROR("%s: cannot connect to next dev %s (%d)\n",
268 obd->obd_name, osp, rc);
272 LASSERT(obd->obd_lu_dev);
273 LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
275 ldev = obd->obd_lu_dev;
278 /* Allocate ost descriptor and fill it */
279 OBD_ALLOC_PTR(tgt_desc);
281 GOTO(out_conn, rc = -ENOMEM);
283 tgt_desc->ltd_tgt = d;
284 tgt_desc->ltd_exp = exp;
285 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
286 tgt_desc->ltd_gen = gen;
287 tgt_desc->ltd_index = index;
288 tgt_desc->ltd_active = active;
291 if (index >= ltd->ltd_tgts_size) {
292 /* we have to increase the size of the lod_osts array */
295 newsize = max(ltd->ltd_tgts_size, (__u32)2);
296 while (newsize < index + 1)
297 newsize = newsize << 1;
299 /* lod_bitmap_resize() needs lod_rw_sem
300 * which we hold with th reference */
301 lod_putref(lod, ltd);
303 rc = ltd_bitmap_resize(ltd, newsize);
310 mutex_lock(<d->ltd_mutex);
311 if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
312 CERROR("%s: device %d is registered already\n", obd->obd_name,
314 GOTO(out_mutex, rc = -EEXIST);
317 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
318 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
319 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
320 CERROR("can't allocate index to add %s\n",
322 GOTO(out_mutex, rc = -ENOMEM);
326 if (!strcmp(LUSTRE_OSC_NAME, type)) {
327 /* pool and qos are not supported for MDS stack yet */
328 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
331 CERROR("%s: can't set up pool, failed with %d\n",
336 rc = qos_add_tgt(lod, tgt_desc);
338 CERROR("%s: qos_add_tgt failed with %d\n",
343 /* The new OST is now a full citizen */
344 if (index >= lod->lod_desc.ld_tgt_count)
345 lod->lod_desc.ld_tgt_count = index + 1;
347 lod->lod_desc.ld_active_tgt_count++;
350 LTD_TGT(ltd, index) = tgt_desc;
351 cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
353 mutex_unlock(<d->ltd_mutex);
354 lod_putref(lod, ltd);
355 if (lod->lod_recovery_completed)
356 ldev->ld_ops->ldo_recovery_complete(env, ldev);
358 rc = lfsck_add_target(env, lod->lod_child, d, exp, index, for_ost);
360 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
361 "index = %u, rc = %d\n", osp, type, index, rc);
366 lod_ost_pool_remove(&lod->lod_pool_info, index);
368 mutex_unlock(<d->ltd_mutex);
369 lod_putref(lod, ltd);
371 OBD_FREE_PTR(tgt_desc);
379 * helper function to schedule OST removal from the device table
381 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
382 struct lod_tgt_descs *ltd, unsigned idx,
385 LASSERT(LTD_TGT(ltd, idx));
387 lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
390 if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
391 LTD_TGT(ltd, idx)->ltd_reap = 1;
392 ltd->ltd_death_row++;
396 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
397 struct lod_tgt_descs *ltd, bool for_ost)
401 if (ltd->ltd_tgts_size <= 0)
404 mutex_lock(<d->ltd_mutex);
405 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
406 __lod_del_device(env, lod, ltd, idx, for_ost);
407 mutex_unlock(<d->ltd_mutex);
408 lod_putref(lod, ltd);
409 CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
410 for (idx = 0; idx < TGT_PTRS; idx++) {
411 if (ltd->ltd_tgt_idx[idx])
412 OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
414 ltd->ltd_tgts_size = 0;
419 * Add support for administratively disabled OST (through the MGS).
420 * Schedule a target for deletion. Disconnection and real removal from the
421 * table takes place in lod_putref() once the last table user release its
424 * \param env - is the environment passed by the caller
425 * \param lod - is the lod device currently connected to the OSP about to be
427 * \param osp - is the name of OSP device about to be removed
428 * \param idx - is the OSP index
429 * \param gen - is the generation number, not used currently
431 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
432 struct lod_tgt_descs *ltd, char *osp, unsigned idx,
433 unsigned gen, bool for_ost)
435 struct obd_device *obd;
437 struct obd_uuid uuid;
440 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
442 obd_str2uuid(&uuid, osp);
444 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
445 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
447 CERROR("can't find %s device\n", osp);
452 CERROR("%s: request to remove OBD %s with invalid generation %d"
453 "\n", obd->obd_name, osp, gen);
457 obd_str2uuid(&uuid, osp);
460 mutex_lock(<d->ltd_mutex);
461 /* check that the index is allocated in the bitmap */
462 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
463 !LTD_TGT(ltd, idx)) {
464 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
465 GOTO(out, rc = -EINVAL);
468 /* check that the UUID matches */
469 if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
470 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
471 obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
473 GOTO(out, rc = -EINVAL);
476 __lod_del_device(env, lod, ltd, idx, for_ost);
479 mutex_unlock(<d->ltd_mutex);
480 lod_putref(lod, ltd);
484 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
486 __u32 round = size_roundup_power2(size);
489 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
490 if (info->lti_ea_store) {
491 LASSERT(info->lti_ea_store_size);
492 LASSERT(info->lti_ea_store_size < round);
493 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
494 info->lti_ea_store_size, round);
495 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
496 info->lti_ea_store = NULL;
497 info->lti_ea_store_size = 0;
500 OBD_ALLOC_LARGE(info->lti_ea_store, round);
501 if (info->lti_ea_store == NULL)
503 info->lti_ea_store_size = round;
508 * generate and write LOV EA for given striped object
510 int lod_generate_and_set_lovea(const struct lu_env *env,
511 struct lod_object *lo, struct thandle *th)
513 struct lod_thread_info *info = lod_env_info(env);
514 struct dt_object *next = dt_object_child(&lo->ldo_obj);
515 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
516 struct lov_mds_md_v1 *lmm;
517 struct lov_ost_data_v1 *objs;
525 magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
526 lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
527 if (info->lti_ea_store_size < lmm_size) {
528 rc = lod_ea_store_resize(info, lmm_size);
533 if (lo->ldo_pattern == 0) /* default striping */
534 lo->ldo_pattern = LOV_PATTERN_RAID0;
536 lmm = info->lti_ea_store;
538 lmm->lmm_magic = cpu_to_le32(magic);
539 lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
540 fid_to_lmm_oi(fid, &lmm->lmm_oi);
541 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
542 lmm->lmm_oi.oi.oi_id++;
543 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
544 lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
545 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
546 if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
547 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
548 lmm->lmm_layout_gen = 0;
549 if (magic == LOV_MAGIC_V1) {
550 objs = &lmm->lmm_objects[0];
552 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
553 size_t cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
554 sizeof(v3->lmm_pool_name));
555 if (cplen >= sizeof(v3->lmm_pool_name))
557 objs = &v3->lmm_objects[0];
560 for (i = 0; i < lo->ldo_stripenr; i++) {
561 struct lu_fid *fid = &info->lti_fid;
562 struct lod_device *lod;
564 int type = LU_SEQ_RANGE_OST;
566 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
567 LASSERT(lo->ldo_stripe[i]);
569 *fid = *lu_object_fid(&lo->ldo_stripe[i]->do_lu);
570 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) {
571 if (cfs_fail_val == 0)
572 cfs_fail_val = fid->f_oid;
574 fid->f_oid = cfs_fail_val;
577 rc = fid_to_ostid(fid, &info->lti_ostid);
580 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
581 objs[i].l_ost_gen = cpu_to_le32(0);
582 rc = lod_fld_lookup(env, lod, fid, &index, &type);
584 CERROR("%s: Can not locate "DFID": rc = %d\n",
585 lod2obd(lod)->obd_name, PFID(fid), rc);
586 lod_object_free_striping(env, lo);
589 objs[i].l_ost_idx = cpu_to_le32(index);
592 info->lti_buf.lb_buf = lmm;
593 info->lti_buf.lb_len = lmm_size;
594 rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
597 lod_object_free_striping(env, lo);
602 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
605 struct lod_thread_info *info = lod_env_info(env);
606 struct dt_object *next = dt_object_child(&lo->ldo_obj);
612 if (unlikely(info->lti_ea_store == NULL)) {
613 /* just to enter in allocation block below */
617 info->lti_buf.lb_buf = info->lti_ea_store;
618 info->lti_buf.lb_len = info->lti_ea_store_size;
619 rc = dt_xattr_get(env, next, &info->lti_buf, name, BYPASS_CAPA);
622 /* if object is not striped or inaccessible */
623 if (rc == -ENODATA || rc == -ENOENT)
627 /* EA doesn't fit, reallocate new buffer */
628 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name,
630 if (rc == -ENODATA || rc == -ENOENT)
636 rc = lod_ea_store_resize(info, rc);
645 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
648 struct lod_thread_info *info = lod_env_info(env);
649 struct lod_object *lo = lod_dt_obj(dt);
650 struct dt_object *next = dt_object_child(dt);
651 struct lov_user_md_v3 *v3;
655 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
658 * store striping defaults into new directory
659 * used to implement defaults inheritance
662 /* probably nothing to inherite */
663 if (lo->ldo_striping_cached == 0)
666 if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
667 lo->ldo_def_stripe_offset))
670 v3 = info->lti_ea_store;
671 if (info->lti_ea_store_size < sizeof(*v3)) {
672 rc = lod_ea_store_resize(info, sizeof(*v3));
675 v3 = info->lti_ea_store;
677 memset(v3, 0, sizeof(*v3));
678 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
679 v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
680 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
681 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
682 if (lo->ldo_pool != NULL)
683 strlcpy(v3->lmm_pool_name, lo->ldo_pool,
684 sizeof(v3->lmm_pool_name));
685 info->lti_buf.lb_buf = v3;
686 info->lti_buf.lb_len = sizeof(*v3);
687 rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
693 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
695 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
696 !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
697 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
698 md->lod_ost_descs.ltd_tgts_size);
702 if (unlikely(OST_TGT(md, idx) == NULL)) {
703 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
704 lod2obd(md)->obd_name, idx);
708 if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
709 CERROR("%s: invalid lod device, for idx: %d\n",
710 lod2obd(md)->obd_name , idx);
718 * allocate array of objects pointers, find/create objects
719 * stripenr and other fields should be initialized by this moment
721 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
722 struct lov_ost_data_v1 *objs)
724 struct lod_thread_info *info = lod_env_info(env);
725 struct lod_device *md;
726 struct lu_object *o, *n;
727 struct lu_device *nd;
728 struct dt_object **stripe;
735 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
736 LASSERT(lo->ldo_stripe == NULL);
737 LASSERT(lo->ldo_stripenr > 0);
738 LASSERT(lo->ldo_stripe_size > 0);
740 stripe_len = lo->ldo_stripenr;
741 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
745 for (i = 0; i < lo->ldo_stripenr; i++) {
746 if (unlikely(lovea_slot_is_dummy(&objs[i])))
749 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
750 idx = le32_to_cpu(objs[i].l_ost_idx);
751 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
754 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
755 PFID(&info->lti_fid));
756 lod_getref(&md->lod_ost_descs);
758 rc = validate_lod_and_idx(md, idx);
759 if (unlikely(rc != 0)) {
760 lod_putref(md, &md->lod_ost_descs);
764 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
765 lod_putref(md, &md->lod_ost_descs);
767 /* In the function below, .hs_keycmp resolves to
768 * u_obj_hop_keycmp() */
769 /* coverity[overrun-buffer-val] */
770 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
772 GOTO(out, rc = PTR_ERR(o));
774 n = lu_object_locate(o->lo_header, nd->ld_type);
777 stripe[i] = container_of(n, struct dt_object, do_lu);
782 for (i = 0; i < stripe_len; i++)
783 if (stripe[i] != NULL)
784 lu_object_put(env, &stripe[i]->do_lu);
786 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
787 lo->ldo_stripenr = 0;
789 lo->ldo_stripe = stripe;
790 lo->ldo_stripes_allocated = stripe_len;
797 * Parse striping information stored in lti_ea_store
799 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
800 const struct lu_buf *buf)
802 struct lov_mds_md_v1 *lmm;
803 struct lov_ost_data_v1 *objs;
810 LASSERT(buf->lb_buf);
811 LASSERT(buf->lb_len);
813 lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
814 magic = le32_to_cpu(lmm->lmm_magic);
815 pattern = le32_to_cpu(lmm->lmm_pattern);
817 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
818 GOTO(out, rc = -EINVAL);
819 if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
820 GOTO(out, rc = -EINVAL);
822 lo->ldo_pattern = pattern;
823 lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
824 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
825 lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
826 /* released file stripenr fixup. */
827 if (pattern & LOV_PATTERN_F_RELEASED)
828 lo->ldo_stripenr = 0;
830 LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
832 if (magic == LOV_MAGIC_V3) {
833 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
834 objs = &v3->lmm_objects[0];
835 lod_object_set_pool(lo, v3->lmm_pool_name);
837 objs = &lmm->lmm_objects[0];
840 if (lo->ldo_stripenr > 0)
841 rc = lod_initialize_objects(env, lo, objs);
847 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
849 struct lod_thread_info *info = lod_env_info(env);
850 struct lu_buf *buf = &info->lti_buf;
851 struct dt_object *next = dt_object_child(&lo->ldo_obj);
855 /* already initialized? */
856 if (lo->ldo_stripe != NULL)
859 if (!dt_object_exists(next))
862 /* Do not load stripe for slaves of striped dir */
863 if (lo->ldo_dir_slave_stripe)
866 if (S_ISREG(lu_object_attr(lod2lu_obj(lo)))) {
867 rc = lod_get_lov_ea(env, lo);
871 * there is LOV EA (striping information) in this object
872 * let's parse it and create in-core objects for the stripes
874 buf->lb_buf = info->lti_ea_store;
875 buf->lb_len = info->lti_ea_store_size;
876 rc = lod_parse_striping(env, lo, buf);
877 } else if (S_ISDIR(lu_object_attr(lod2lu_obj(lo)))) {
878 rc = lod_get_lmv_ea(env, lo);
879 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1))
880 GOTO(out, rc = rc > 0 ? -EINVAL : rc);
882 buf->lb_buf = info->lti_ea_store;
883 buf->lb_len = info->lti_ea_store_size;
884 if (rc == sizeof(struct lmv_mds_md_v1)) {
885 rc = lod_load_lmv_shards(env, lo, buf, true);
886 if (buf->lb_buf != info->lti_ea_store) {
887 OBD_FREE_LARGE(info->lti_ea_store,
888 info->lti_ea_store_size);
889 info->lti_ea_store = buf->lb_buf;
890 info->lti_ea_store_size = buf->lb_len;
898 * there is LOV EA (striping information) in this object
899 * let's parse it and create in-core objects for the stripes
901 rc = lod_parse_dir_striping(env, lo, buf);
908 * Load and parse striping information, create in-core representation for the
911 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
913 struct dt_object *next = dt_object_child(&lo->ldo_obj);
916 /* currently this code is supposed to be called from declaration
917 * phase only, thus the object is not expected to be locked by caller */
918 dt_write_lock(env, next, 0);
919 rc = lod_load_striping_locked(env, lo);
920 dt_write_unlock(env, next);
924 /* verify the striping information for directory */
925 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
928 struct lov_user_md_v1 *lum;
929 struct lov_user_md_v3 *lum3;
930 struct pool_desc *pool = NULL;
941 LASSERT(sizeof(*lum) < sizeof(*lum3));
943 if (buf->lb_len < sizeof(*lum)) {
944 CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md\n",
946 GOTO(out, rc = -EINVAL);
949 magic = le32_to_cpu(lum->lmm_magic);
950 if (magic != LOV_USER_MAGIC_V1 &&
951 magic != LOV_USER_MAGIC_V3 &&
952 magic != LOV_MAGIC_V1_DEF &&
953 magic != LOV_MAGIC_V3_DEF) {
954 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
955 GOTO(out, rc = -EINVAL);
958 /* the user uses "0" for default stripe pattern normally. */
959 if (!is_from_disk && lum->lmm_pattern == 0)
960 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
962 if (le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) {
963 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
964 le32_to_cpu(lum->lmm_pattern));
965 GOTO(out, rc = -EINVAL);
968 /* 64kB is the largest common page size we see (ia64), and matches the
970 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
971 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
972 CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
973 stripe_size, LOV_MIN_STRIPE_SIZE);
974 GOTO(out, rc = -EINVAL);
977 /* an offset of -1 is treated as a "special" valid offset */
978 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
979 if (stripe_offset != (typeof(stripe_offset))-1) {
980 /* if offset is not within valid range [0, osts_size) */
981 if (stripe_offset >= d->lod_osts_size) {
982 CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
983 stripe_offset, d->lod_osts_size);
984 GOTO(out, rc = -EINVAL);
987 /* if lmm_stripe_offset is *not* in bitmap */
988 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
989 CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
991 GOTO(out, rc = -EINVAL);
995 if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
996 lum_size = offsetof(struct lov_user_md_v1,
998 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
999 lum_size = offsetof(struct lov_user_md_v3,
1002 GOTO(out, rc = -EINVAL);
1004 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1005 if (buf->lb_len != lum_size) {
1006 CDEBUG(D_IOCTL, "invalid buf len %zd for lov_user_md with "
1007 "magic %#x and stripe_count %u\n",
1008 buf->lb_len, magic, stripe_count);
1009 GOTO(out, rc = -EINVAL);
1012 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
1016 if (buf->lb_len < sizeof(*lum3)) {
1017 CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md_v3\n",
1019 GOTO(out, rc = -EINVAL);
1022 /* In the function below, .hs_keycmp resolves to
1023 * pool_hashkey_keycmp() */
1024 /* coverity[overrun-buffer-val] */
1025 pool = lod_find_pool(d, lum3->lmm_pool_name);
1029 if (stripe_offset != (typeof(stripe_offset))-1) {
1030 rc = lod_check_index_in_pool(stripe_offset, pool);
1032 GOTO(out, rc = -EINVAL);
1035 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1037 "stripe count %u > # OSTs %u in the pool\n",
1038 stripe_count, pool_tgt_count(pool));
1039 GOTO(out, rc = -EINVAL);
1044 lod_pool_putref(pool);
1049 void lod_fix_desc_stripe_size(__u64 *val)
1051 if (*val < LOV_MIN_STRIPE_SIZE) {
1053 LCONSOLE_INFO("Increasing default stripe size to "
1054 "minimum value %u\n",
1055 LOV_DESC_STRIPE_SIZE_DEFAULT);
1056 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1057 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1058 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1059 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
1060 "multiple of %u)\n",
1061 *val, LOV_MIN_STRIPE_SIZE);
1065 void lod_fix_desc_stripe_count(__u32 *val)
1071 void lod_fix_desc_pattern(__u32 *val)
1073 /* from lov_setstripe */
1074 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1075 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1080 void lod_fix_desc_qos_maxage(__u32 *val)
1082 /* fix qos_maxage */
1084 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1087 void lod_fix_desc(struct lov_desc *desc)
1089 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1090 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1091 lod_fix_desc_pattern(&desc->ld_pattern);
1092 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1095 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1097 struct obd_device *obd;
1098 struct lov_desc *desc;
1102 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1103 LASSERT(obd != NULL);
1104 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1106 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1107 CERROR("LOD setup requires a descriptor\n");
1111 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1113 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1114 CERROR("descriptor size wrong: %d > %d\n",
1115 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1119 if (desc->ld_magic != LOV_DESC_MAGIC) {
1120 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1121 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1122 obd->obd_name, desc);
1123 lustre_swab_lov_desc(desc);
1125 CERROR("%s: Bad lov desc magic: %#x\n",
1126 obd->obd_name, desc->ld_magic);
1133 desc->ld_active_tgt_count = 0;
1134 lod->lod_desc = *desc;
1136 lod->lod_sp_me = LUSTRE_SP_CLI;
1138 /* Set up allocation policy (QoS and RR) */
1139 INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1140 init_rwsem(&lod->lod_qos.lq_rw_sem);
1141 lod->lod_qos.lq_dirty = 1;
1142 lod->lod_qos.lq_rr.lqr_dirty = 1;
1143 lod->lod_qos.lq_reset = 1;
1144 /* Default priority is toward free space balance */
1145 lod->lod_qos.lq_prio_free = 232;
1146 /* Default threshold for rr (roughly 17%) */
1147 lod->lod_qos.lq_threshold_rr = 43;
1149 /* Set up OST pool environment */
1150 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1151 HASH_POOLS_MAX_BITS,
1152 HASH_POOLS_BKT_BITS, 0,
1155 &pool_hash_operations,
1157 if (lod->lod_pools_hash_body == NULL)
1160 INIT_LIST_HEAD(&lod->lod_pool_list);
1161 lod->lod_pool_count = 0;
1162 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1165 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1167 GOTO(out_pool_info, rc);
1172 lod_ost_pool_free(&lod->lod_pool_info);
1174 cfs_hash_putref(lod->lod_pools_hash_body);
1179 int lod_pools_fini(struct lod_device *lod)
1181 struct obd_device *obd = lod2obd(lod);
1182 struct pool_desc *pool, *tmp;
1185 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
1186 /* free pool structs */
1187 CDEBUG(D_INFO, "delete pool %p\n", pool);
1188 /* In the function below, .hs_keycmp resolves to
1189 * pool_hashkey_keycmp() */
1190 /* coverity[overrun-buffer-val] */
1191 lod_pool_del(obd, pool->pool_name);
1194 cfs_hash_putref(lod->lod_pools_hash_body);
1195 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1196 lod_ost_pool_free(&lod->lod_pool_info);