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>
39 #include "lod_internal.h"
42 * Keep a refcount of lod->ltd_tgts usage to prevent racing with
43 * addition/deletion. Any function that expects lov_tgts to remain stationary
46 * \param lod - is the lod device from which we want to grab a reference
48 void lod_getref(struct lod_tgt_descs *ltd)
50 down_read(<d->ltd_rw_sem);
51 mutex_lock(<d->ltd_mutex);
53 mutex_unlock(<d->ltd_mutex);
57 * Companion of lod_getref() to release a reference on the lod table.
58 * If this is the last reference and the ost entry was scheduled for deletion,
59 * the descriptor is removed from the array.
61 * \param lod - is the lod device from which we release a reference
63 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
65 mutex_lock(<d->ltd_mutex);
67 if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
68 struct lod_tgt_desc *tgt_desc, *tmp;
72 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
75 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
76 tgt_desc = LTD_TGT(ltd, idx);
79 if (!tgt_desc->ltd_reap)
82 cfs_list_add(&tgt_desc->ltd_kill, &kill);
83 LTD_TGT(ltd, idx) = NULL;
84 /*FIXME: only support ost pool for now */
85 if (ltd == &lod->lod_ost_descs) {
86 lod_ost_pool_remove(&lod->lod_pool_info, idx);
87 if (tgt_desc->ltd_active)
88 lod->lod_desc.ld_active_tgt_count--;
91 cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
94 mutex_unlock(<d->ltd_mutex);
95 up_read(<d->ltd_rw_sem);
97 cfs_list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
99 cfs_list_del(&tgt_desc->ltd_kill);
100 if (ltd == &lod->lod_ost_descs) {
101 /* remove from QoS structures */
102 rc = qos_del_tgt(lod, tgt_desc);
104 CERROR("%s: qos_del_tgt(%s) failed:"
106 lod2obd(lod)->obd_name,
107 obd_uuid2str(&tgt_desc->ltd_uuid),
110 rc = obd_disconnect(tgt_desc->ltd_exp);
112 CERROR("%s: failed to disconnect %s: rc = %d\n",
113 lod2obd(lod)->obd_name,
114 obd_uuid2str(&tgt_desc->ltd_uuid), rc);
115 OBD_FREE_PTR(tgt_desc);
118 mutex_unlock(<d->ltd_mutex);
119 up_read(<d->ltd_rw_sem);
123 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
125 cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
129 /* grab write reference on the lod. Relocating the array requires
130 * exclusive access */
132 down_write(<d->ltd_rw_sem);
133 if (newsize <= ltd->ltd_tgts_size)
134 /* someone else has already resize the array */
137 /* allocate new bitmap */
138 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
140 GOTO(out, rc = -ENOMEM);
142 if (ltd->ltd_tgts_size > 0) {
143 /* the bitmap already exists, we need
144 * to copy data from old one */
145 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
146 old_bitmap = ltd->ltd_tgt_bitmap;
149 ltd->ltd_tgts_size = newsize;
150 ltd->ltd_tgt_bitmap = new_bitmap;
153 CFS_FREE_BITMAP(old_bitmap);
155 CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
159 up_write(<d->ltd_rw_sem);
164 * Connect LOD to a new OSP and add it to the device table.
166 * \param env - is the environment passed by the caller
167 * \param lod - is the LOD device to be connected to the new OSP
168 * \param osp - is the name of OSP device name about to be added
169 * \param index - is the OSP index
170 * \param gen - is the generation number
171 * \param tgt_index - is the group of the OSP.
172 * \param type - is the type of device (mdc or osc)
174 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
175 char *osp, unsigned index, unsigned gen, int tgt_index,
176 char *type, int active)
178 struct obd_connect_data *data = NULL;
179 struct obd_export *exp = NULL;
180 struct obd_device *obd;
181 struct lu_device *ldev;
184 struct lod_tgt_desc *tgt_desc;
185 struct lod_tgt_descs *ltd;
186 struct obd_uuid obd_uuid;
190 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
193 CERROR("request to add OBD %s with invalid generation: %d\n",
198 obd_str2uuid(&obd_uuid, osp);
200 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
201 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
203 CERROR("can't find %s device\n", osp);
211 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
212 data->ocd_version = LUSTRE_VERSION_CODE;
213 data->ocd_index = index;
215 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
217 data->ocd_connect_flags |= OBD_CONNECT_AT |
220 #ifdef HAVE_LRU_RESIZE_SUPPORT
221 OBD_CONNECT_LRU_RESIZE |
224 OBD_CONNECT_OSS_CAPA |
225 OBD_CONNECT_REQPORTAL |
226 OBD_CONNECT_SKIP_ORPHAN |
228 OBD_CONNECT_LVB_TYPE |
229 OBD_CONNECT_VERSION |
230 OBD_CONNECT_PINGLESS |
233 data->ocd_group = tgt_index;
234 ltd = &lod->lod_ost_descs;
236 struct obd_import *imp = obd->u.cli.cl_import;
239 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
240 data->ocd_connect_flags |= OBD_CONNECT_ACL |
241 OBD_CONNECT_MDS_CAPA |
242 OBD_CONNECT_OSS_CAPA |
244 OBD_CONNECT_MDS_MDS |
249 /* XXX set MDS-MDS flags, remove this when running this
251 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
252 spin_lock(&imp->imp_lock);
253 imp->imp_server_timeout = 1;
254 spin_unlock(&imp->imp_lock);
255 imp->imp_client->cli_request_portal = OUT_PORTAL;
256 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
258 ltd = &lod->lod_mdt_descs;
261 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
264 CERROR("%s: cannot connect to next dev %s (%d)\n",
265 obd->obd_name, osp, rc);
269 LASSERT(obd->obd_lu_dev);
270 LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
272 ldev = obd->obd_lu_dev;
275 /* Allocate ost descriptor and fill it */
276 OBD_ALLOC_PTR(tgt_desc);
278 GOTO(out_conn, rc = -ENOMEM);
280 tgt_desc->ltd_tgt = d;
281 tgt_desc->ltd_exp = exp;
282 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
283 tgt_desc->ltd_gen = gen;
284 tgt_desc->ltd_index = index;
285 tgt_desc->ltd_active = active;
288 if (index >= ltd->ltd_tgts_size) {
289 /* we have to increase the size of the lod_osts array */
292 newsize = max(ltd->ltd_tgts_size, (__u32)2);
293 while (newsize < index + 1)
294 newsize = newsize << 1;
296 /* lod_bitmap_resize() needs lod_rw_sem
297 * which we hold with th reference */
298 lod_putref(lod, ltd);
300 rc = ltd_bitmap_resize(ltd, newsize);
307 mutex_lock(<d->ltd_mutex);
308 if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
309 CERROR("%s: device %d is registered already\n", obd->obd_name,
311 GOTO(out_mutex, rc = -EEXIST);
314 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
315 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
316 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
317 CERROR("can't allocate index to add %s\n",
319 GOTO(out_mutex, rc = -ENOMEM);
323 if (!strcmp(LUSTRE_OSC_NAME, type)) {
324 /* pool and qos are not supported for MDS stack yet */
325 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
328 CERROR("%s: can't set up pool, failed with %d\n",
333 rc = qos_add_tgt(lod, tgt_desc);
335 CERROR("%s: qos_add_tgt failed with %d\n",
340 /* The new OST is now a full citizen */
341 if (index >= lod->lod_desc.ld_tgt_count)
342 lod->lod_desc.ld_tgt_count = index + 1;
344 lod->lod_desc.ld_active_tgt_count++;
347 LTD_TGT(ltd, index) = tgt_desc;
348 cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
350 mutex_unlock(<d->ltd_mutex);
351 lod_putref(lod, ltd);
352 if (lod->lod_recovery_completed)
353 ldev->ld_ops->ldo_recovery_complete(env, ldev);
355 rc = lfsck_add_target(env, lod->lod_child, d, exp, index, for_ost);
357 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
358 "index = %u, rc = %d\n", osp, type, index, rc);
363 lod_ost_pool_remove(&lod->lod_pool_info, index);
365 mutex_unlock(<d->ltd_mutex);
366 lod_putref(lod, ltd);
368 OBD_FREE_PTR(tgt_desc);
376 * helper function to schedule OST removal from the device table
378 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
379 struct lod_tgt_descs *ltd, unsigned idx,
382 LASSERT(LTD_TGT(ltd, idx));
384 lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
387 if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
388 LTD_TGT(ltd, idx)->ltd_reap = 1;
389 ltd->ltd_death_row++;
393 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
394 struct lod_tgt_descs *ltd, bool for_ost)
398 if (ltd->ltd_tgts_size <= 0)
401 mutex_lock(<d->ltd_mutex);
402 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
403 __lod_del_device(env, lod, ltd, idx, for_ost);
404 mutex_unlock(<d->ltd_mutex);
405 lod_putref(lod, ltd);
406 CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
407 for (idx = 0; idx < TGT_PTRS; idx++) {
408 if (ltd->ltd_tgt_idx[idx])
409 OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
411 ltd->ltd_tgts_size = 0;
416 * Add support for administratively disabled OST (through the MGS).
417 * Schedule a target for deletion. Disconnection and real removal from the
418 * table takes place in lod_putref() once the last table user release its
421 * \param env - is the environment passed by the caller
422 * \param lod - is the lod device currently connected to the OSP about to be
424 * \param osp - is the name of OSP device about to be removed
425 * \param idx - is the OSP index
426 * \param gen - is the generation number, not used currently
428 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
429 struct lod_tgt_descs *ltd, char *osp, unsigned idx,
430 unsigned gen, bool for_ost)
432 struct obd_device *obd;
434 struct obd_uuid uuid;
437 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
439 obd_str2uuid(&uuid, osp);
441 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
442 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
444 CERROR("can't find %s device\n", osp);
449 CERROR("%s: request to remove OBD %s with invalid generation %d"
450 "\n", obd->obd_name, osp, gen);
454 obd_str2uuid(&uuid, osp);
457 mutex_lock(<d->ltd_mutex);
458 /* check that the index is allocated in the bitmap */
459 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
460 !LTD_TGT(ltd, idx)) {
461 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
462 GOTO(out, rc = -EINVAL);
465 /* check that the UUID matches */
466 if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
467 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
468 obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
470 GOTO(out, rc = -EINVAL);
473 __lod_del_device(env, lod, ltd, idx, for_ost);
476 mutex_unlock(<d->ltd_mutex);
477 lod_putref(lod, ltd);
481 int lod_ea_store_resize(struct lod_thread_info *info, int size)
483 int round = size_roundup_power2(size);
486 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
487 if (info->lti_ea_store) {
488 LASSERT(info->lti_ea_store_size);
489 LASSERT(info->lti_ea_store_size < round);
490 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
491 info->lti_ea_store_size, round);
492 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
493 info->lti_ea_store = NULL;
494 info->lti_ea_store_size = 0;
497 OBD_ALLOC_LARGE(info->lti_ea_store, round);
498 if (info->lti_ea_store == NULL)
500 info->lti_ea_store_size = round;
505 * generate and write LOV EA for given striped object
507 int lod_generate_and_set_lovea(const struct lu_env *env,
508 struct lod_object *lo, struct thandle *th)
510 struct lod_thread_info *info = lod_env_info(env);
511 struct dt_object *next = dt_object_child(&lo->ldo_obj);
512 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
513 struct lov_mds_md_v1 *lmm;
514 struct lov_ost_data_v1 *objs;
522 magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
523 lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
524 if (info->lti_ea_store_size < lmm_size) {
525 rc = lod_ea_store_resize(info, lmm_size);
530 if (lo->ldo_pattern == 0) /* default striping */
531 lo->ldo_pattern = LOV_PATTERN_RAID0;
533 lmm = info->lti_ea_store;
535 lmm->lmm_magic = cpu_to_le32(magic);
536 lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
537 fid_to_lmm_oi(fid, &lmm->lmm_oi);
538 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
539 lmm->lmm_oi.oi.oi_id++;
540 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
541 lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
542 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
543 if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
544 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
545 lmm->lmm_layout_gen = 0;
546 if (magic == LOV_MAGIC_V1) {
547 objs = &lmm->lmm_objects[0];
549 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
550 cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
551 sizeof(v3->lmm_pool_name));
552 if (cplen >= sizeof(v3->lmm_pool_name))
554 objs = &v3->lmm_objects[0];
557 for (i = 0; i < lo->ldo_stripenr; i++) {
558 struct lu_fid *fid = &info->lti_fid;
559 struct lod_device *lod;
561 int type = LU_SEQ_RANGE_OST;
563 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
564 LASSERT(lo->ldo_stripe[i]);
566 *fid = *lu_object_fid(&lo->ldo_stripe[i]->do_lu);
567 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) {
568 if (cfs_fail_val == 0)
569 cfs_fail_val = fid->f_oid;
571 fid->f_oid = cfs_fail_val;
574 rc = fid_to_ostid(fid, &info->lti_ostid);
577 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
578 objs[i].l_ost_gen = cpu_to_le32(0);
579 rc = lod_fld_lookup(env, lod, fid, &index, &type);
581 CERROR("%s: Can not locate "DFID": rc = %d\n",
582 lod2obd(lod)->obd_name, PFID(fid), rc);
583 lod_object_free_striping(env, lo);
586 objs[i].l_ost_idx = cpu_to_le32(index);
589 info->lti_buf.lb_buf = lmm;
590 info->lti_buf.lb_len = lmm_size;
591 rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
594 lod_object_free_striping(env, lo);
599 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
602 struct lod_thread_info *info = lod_env_info(env);
603 struct dt_object *next = dt_object_child(&lo->ldo_obj);
609 if (unlikely(info->lti_ea_store == NULL)) {
610 /* just to enter in allocation block below */
614 info->lti_buf.lb_buf = info->lti_ea_store;
615 info->lti_buf.lb_len = info->lti_ea_store_size;
616 rc = dt_xattr_get(env, next, &info->lti_buf, name, BYPASS_CAPA);
619 /* if object is not striped or inaccessible */
620 if (rc == -ENODATA || rc == -ENOENT)
624 /* EA doesn't fit, reallocate new buffer */
625 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name,
627 if (rc == -ENODATA || rc == -ENOENT)
633 rc = lod_ea_store_resize(info, rc);
642 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
645 struct lod_thread_info *info = lod_env_info(env);
646 struct lod_object *lo = lod_dt_obj(dt);
647 struct dt_object *next = dt_object_child(dt);
648 struct lov_user_md_v3 *v3;
652 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
655 * store striping defaults into new directory
656 * used to implement defaults inheritance
659 /* probably nothing to inherite */
660 if (lo->ldo_striping_cached == 0)
663 if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
664 lo->ldo_def_stripe_offset))
667 v3 = info->lti_ea_store;
668 if (info->lti_ea_store_size < sizeof(*v3)) {
669 rc = lod_ea_store_resize(info, sizeof(*v3));
672 v3 = info->lti_ea_store;
674 memset(v3, 0, sizeof(*v3));
675 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
676 v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
677 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
678 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
679 if (lo->ldo_pool != NULL)
680 strlcpy(v3->lmm_pool_name, lo->ldo_pool,
681 sizeof(v3->lmm_pool_name));
682 info->lti_buf.lb_buf = v3;
683 info->lti_buf.lb_len = sizeof(*v3);
684 rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
690 static int validate_lod_and_idx(struct lod_device *md, int idx)
692 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
693 !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
694 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
695 md->lod_ost_descs.ltd_tgts_size);
699 if (unlikely(OST_TGT(md, idx) == NULL)) {
700 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
701 lod2obd(md)->obd_name, idx);
705 if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
706 CERROR("%s: invalid lod device, for idx: %d\n",
707 lod2obd(md)->obd_name , idx);
715 * allocate array of objects pointers, find/create objects
716 * stripenr and other fields should be initialized by this moment
718 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
719 struct lov_ost_data_v1 *objs)
721 struct lod_thread_info *info = lod_env_info(env);
722 struct lod_device *md;
723 struct lu_object *o, *n;
724 struct lu_device *nd;
725 struct dt_object **stripe;
731 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
732 LASSERT(lo->ldo_stripe == NULL);
733 LASSERT(lo->ldo_stripenr > 0);
734 LASSERT(lo->ldo_stripe_size > 0);
736 stripe_len = lo->ldo_stripenr;
737 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
741 for (i = 0; i < lo->ldo_stripenr; i++) {
742 if (unlikely(lovea_slot_is_dummy(&objs[i])))
745 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
746 idx = le32_to_cpu(objs[i].l_ost_idx);
747 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
750 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
751 PFID(&info->lti_fid));
752 lod_getref(&md->lod_ost_descs);
754 rc = validate_lod_and_idx(md, idx);
755 if (unlikely(rc != 0)) {
756 lod_putref(md, &md->lod_ost_descs);
760 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
761 lod_putref(md, &md->lod_ost_descs);
763 /* In the function below, .hs_keycmp resolves to
764 * u_obj_hop_keycmp() */
765 /* coverity[overrun-buffer-val] */
766 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
768 GOTO(out, rc = PTR_ERR(o));
770 n = lu_object_locate(o->lo_header, nd->ld_type);
773 stripe[i] = container_of(n, struct dt_object, do_lu);
778 for (i = 0; i < stripe_len; i++)
779 if (stripe[i] != NULL)
780 lu_object_put(env, &stripe[i]->do_lu);
782 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
783 lo->ldo_stripenr = 0;
785 lo->ldo_stripe = stripe;
786 lo->ldo_stripes_allocated = stripe_len;
793 * Parse striping information stored in lti_ea_store
795 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
796 const struct lu_buf *buf)
798 struct lov_mds_md_v1 *lmm;
799 struct lov_ost_data_v1 *objs;
806 LASSERT(buf->lb_buf);
807 LASSERT(buf->lb_len);
809 lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
810 magic = le32_to_cpu(lmm->lmm_magic);
811 pattern = le32_to_cpu(lmm->lmm_pattern);
813 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
814 GOTO(out, rc = -EINVAL);
815 if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
816 GOTO(out, rc = -EINVAL);
818 lo->ldo_pattern = pattern;
819 lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
820 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
821 lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
822 /* released file stripenr fixup. */
823 if (pattern & LOV_PATTERN_F_RELEASED)
824 lo->ldo_stripenr = 0;
826 LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
828 if (magic == LOV_MAGIC_V3) {
829 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
830 objs = &v3->lmm_objects[0];
831 lod_object_set_pool(lo, v3->lmm_pool_name);
833 objs = &lmm->lmm_objects[0];
836 if (lo->ldo_stripenr > 0)
837 rc = lod_initialize_objects(env, lo, objs);
843 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
845 struct lod_thread_info *info = lod_env_info(env);
846 struct dt_object *next = dt_object_child(&lo->ldo_obj);
850 /* already initialized? */
851 if (lo->ldo_stripe != NULL)
854 if (!dt_object_exists(next))
857 /* Do not load stripe for slaves of striped dir */
858 if (lo->ldo_dir_slave_stripe)
861 if (S_ISREG(lu_object_attr(lod2lu_obj(lo)))) {
862 rc = lod_get_lov_ea(env, lo);
866 * there is LOV EA (striping information) in this object
867 * let's parse it and create in-core objects for the stripes
869 info->lti_buf.lb_buf = info->lti_ea_store;
870 info->lti_buf.lb_len = info->lti_ea_store_size;
871 rc = lod_parse_striping(env, lo, &info->lti_buf);
872 } else if (S_ISDIR(lu_object_attr(lod2lu_obj(lo)))) {
873 rc = lod_get_lmv_ea(env, lo);
877 * there is LOV EA (striping information) in this object
878 * let's parse it and create in-core objects for the stripes
880 info->lti_buf.lb_buf = info->lti_ea_store;
881 info->lti_buf.lb_len = info->lti_ea_store_size;
882 rc = lod_parse_dir_striping(env, lo, &info->lti_buf);
889 * Load and parse striping information, create in-core representation for the
892 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
894 struct dt_object *next = dt_object_child(&lo->ldo_obj);
897 /* currently this code is supposed to be called from declaration
898 * phase only, thus the object is not expected to be locked by caller */
899 dt_write_lock(env, next, 0);
900 rc = lod_load_striping_locked(env, lo);
901 dt_write_unlock(env, next);
905 /* verify the striping information for directory */
906 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
909 struct lov_user_md_v1 *lum;
910 struct lov_user_md_v3 *lum3;
911 struct pool_desc *pool = NULL;
922 LASSERT(sizeof(*lum) < sizeof(*lum3));
924 if (buf->lb_len < sizeof(*lum)) {
925 CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md\n",
927 GOTO(out, rc = -EINVAL);
930 magic = le32_to_cpu(lum->lmm_magic);
931 if (magic != LOV_USER_MAGIC_V1 &&
932 magic != LOV_USER_MAGIC_V3 &&
933 magic != LOV_MAGIC_V1_DEF &&
934 magic != LOV_MAGIC_V3_DEF) {
935 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
936 GOTO(out, rc = -EINVAL);
939 /* the user uses "0" for default stripe pattern normally. */
940 if (!is_from_disk && lum->lmm_pattern == 0)
941 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
943 if (le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) {
944 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
945 le32_to_cpu(lum->lmm_pattern));
946 GOTO(out, rc = -EINVAL);
949 /* 64kB is the largest common page size we see (ia64), and matches the
951 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
952 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
953 CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
954 stripe_size, LOV_MIN_STRIPE_SIZE);
955 GOTO(out, rc = -EINVAL);
958 /* an offset of -1 is treated as a "special" valid offset */
959 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
960 if (stripe_offset != (typeof(stripe_offset))-1) {
961 /* if offset is not within valid range [0, osts_size) */
962 if (stripe_offset >= d->lod_osts_size) {
963 CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
964 stripe_offset, d->lod_osts_size);
965 GOTO(out, rc = -EINVAL);
968 /* if lmm_stripe_offset is *not* in bitmap */
969 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
970 CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
972 GOTO(out, rc = -EINVAL);
976 if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
977 lum_size = offsetof(struct lov_user_md_v1,
979 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
980 lum_size = offsetof(struct lov_user_md_v3,
983 GOTO(out, rc = -EINVAL);
985 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
986 if (buf->lb_len != lum_size) {
987 CDEBUG(D_IOCTL, "invalid buf len %zd for lov_user_md with "
988 "magic %#x and stripe_count %u\n",
989 buf->lb_len, magic, stripe_count);
990 GOTO(out, rc = -EINVAL);
993 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
997 if (buf->lb_len < sizeof(*lum3)) {
998 CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md_v3\n",
1000 GOTO(out, rc = -EINVAL);
1003 /* In the function below, .hs_keycmp resolves to
1004 * pool_hashkey_keycmp() */
1005 /* coverity[overrun-buffer-val] */
1006 pool = lod_find_pool(d, lum3->lmm_pool_name);
1010 if (stripe_offset != (typeof(stripe_offset))-1) {
1011 rc = lod_check_index_in_pool(stripe_offset, pool);
1013 GOTO(out, rc = -EINVAL);
1016 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1018 "stripe count %u > # OSTs %u in the pool\n",
1019 stripe_count, pool_tgt_count(pool));
1020 GOTO(out, rc = -EINVAL);
1025 lod_pool_putref(pool);
1030 void lod_fix_desc_stripe_size(__u64 *val)
1032 if (*val < LOV_MIN_STRIPE_SIZE) {
1034 LCONSOLE_INFO("Increasing default stripe size to "
1035 "minimum value %u\n",
1036 LOV_DESC_STRIPE_SIZE_DEFAULT);
1037 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1038 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1039 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1040 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
1041 "multiple of %u)\n",
1042 *val, LOV_MIN_STRIPE_SIZE);
1046 void lod_fix_desc_stripe_count(__u32 *val)
1052 void lod_fix_desc_pattern(__u32 *val)
1054 /* from lov_setstripe */
1055 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1056 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1061 void lod_fix_desc_qos_maxage(__u32 *val)
1063 /* fix qos_maxage */
1065 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1068 void lod_fix_desc(struct lov_desc *desc)
1070 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1071 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1072 lod_fix_desc_pattern(&desc->ld_pattern);
1073 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1076 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1078 struct obd_device *obd;
1079 struct lov_desc *desc;
1083 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1084 LASSERT(obd != NULL);
1085 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1087 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1088 CERROR("LOD setup requires a descriptor\n");
1092 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1094 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1095 CERROR("descriptor size wrong: %d > %d\n",
1096 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1100 if (desc->ld_magic != LOV_DESC_MAGIC) {
1101 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1102 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1103 obd->obd_name, desc);
1104 lustre_swab_lov_desc(desc);
1106 CERROR("%s: Bad lov desc magic: %#x\n",
1107 obd->obd_name, desc->ld_magic);
1114 desc->ld_active_tgt_count = 0;
1115 lod->lod_desc = *desc;
1117 lod->lod_sp_me = LUSTRE_SP_CLI;
1119 /* Set up allocation policy (QoS and RR) */
1120 CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1121 init_rwsem(&lod->lod_qos.lq_rw_sem);
1122 lod->lod_qos.lq_dirty = 1;
1123 lod->lod_qos.lq_rr.lqr_dirty = 1;
1124 lod->lod_qos.lq_reset = 1;
1125 /* Default priority is toward free space balance */
1126 lod->lod_qos.lq_prio_free = 232;
1127 /* Default threshold for rr (roughly 17%) */
1128 lod->lod_qos.lq_threshold_rr = 43;
1130 /* Set up OST pool environment */
1131 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1132 HASH_POOLS_MAX_BITS,
1133 HASH_POOLS_BKT_BITS, 0,
1136 &pool_hash_operations,
1138 if (lod->lod_pools_hash_body == NULL)
1141 CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
1142 lod->lod_pool_count = 0;
1143 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1146 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1148 GOTO(out_pool_info, rc);
1153 lod_ost_pool_free(&lod->lod_pool_info);
1155 cfs_hash_putref(lod->lod_pools_hash_body);
1160 int lod_pools_fini(struct lod_device *lod)
1162 struct obd_device *obd = lod2obd(lod);
1163 cfs_list_t *pos, *tmp;
1164 struct pool_desc *pool;
1167 cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
1168 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
1169 /* free pool structs */
1170 CDEBUG(D_INFO, "delete pool %p\n", pool);
1171 /* In the function below, .hs_keycmp resolves to
1172 * pool_hashkey_keycmp() */
1173 /* coverity[overrun-buffer-val] */
1174 lod_pool_del(obd, pool->pool_name);
1177 cfs_hash_putref(lod->lod_pools_hash_body);
1178 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1179 lod_ost_pool_free(&lod->lod_pool_info);