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>
38 #include <lustre_lfsck.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;
73 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
76 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
77 tgt_desc = LTD_TGT(ltd, idx);
80 if (!tgt_desc->ltd_reap)
83 cfs_list_add(&tgt_desc->ltd_kill, &kill);
84 LTD_TGT(ltd, idx) = NULL;
85 /*FIXME: only support ost pool for now */
86 if (ltd == &lod->lod_ost_descs) {
87 lod_ost_pool_remove(&lod->lod_pool_info, idx);
88 if (tgt_desc->ltd_active)
89 lod->lod_desc.ld_active_tgt_count--;
92 cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
95 mutex_unlock(<d->ltd_mutex);
96 up_read(<d->ltd_rw_sem);
98 cfs_list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
100 cfs_list_del(&tgt_desc->ltd_kill);
101 if (ltd == &lod->lod_ost_descs) {
102 /* remove from QoS structures */
103 rc = qos_del_tgt(lod, tgt_desc);
105 CERROR("%s: qos_del_tgt(%s) failed:"
107 lod2obd(lod)->obd_name,
108 obd_uuid2str(&tgt_desc->ltd_uuid),
111 rc = obd_disconnect(tgt_desc->ltd_exp);
113 CERROR("%s: failed to disconnect %s: rc = %d\n",
114 lod2obd(lod)->obd_name,
115 obd_uuid2str(&tgt_desc->ltd_uuid), rc);
116 OBD_FREE_PTR(tgt_desc);
119 mutex_unlock(<d->ltd_mutex);
120 up_read(<d->ltd_rw_sem);
124 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
126 cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
130 /* grab write reference on the lod. Relocating the array requires
131 * exclusive access */
133 down_write(<d->ltd_rw_sem);
134 if (newsize <= ltd->ltd_tgts_size)
135 /* someone else has already resize the array */
138 /* allocate new bitmap */
139 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
141 GOTO(out, rc = -ENOMEM);
143 if (ltd->ltd_tgts_size > 0) {
144 /* the bitmap already exists, we need
145 * to copy data from old one */
146 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
147 old_bitmap = ltd->ltd_tgt_bitmap;
150 ltd->ltd_tgts_size = newsize;
151 ltd->ltd_tgt_bitmap = new_bitmap;
154 CFS_FREE_BITMAP(old_bitmap);
156 CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
160 up_write(<d->ltd_rw_sem);
165 * Connect LOD to a new OSP and add it to the device table.
167 * \param env - is the environment passed by the caller
168 * \param lod - is the LOD device to be connected to the new OSP
169 * \param osp - is the name of OSP device name about to be added
170 * \param index - is the OSP index
171 * \param gen - is the generation number
172 * \param tgt_index - is the group of the OSP.
173 * \param type - is the type of device (mdc or osc)
175 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
176 char *osp, unsigned index, unsigned gen, int tgt_index,
177 char *type, int active)
179 struct obd_connect_data *data = NULL;
180 struct obd_export *exp = NULL;
181 struct obd_device *obd;
182 struct lu_device *ldev;
185 struct lod_tgt_desc *tgt_desc;
186 struct lod_tgt_descs *ltd;
187 struct obd_uuid obd_uuid;
191 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
194 CERROR("request to add OBD %s with invalid generation: %d\n",
199 obd_str2uuid(&obd_uuid, osp);
201 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
202 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
204 CERROR("can't find %s device\n", osp);
212 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
213 data->ocd_version = LUSTRE_VERSION_CODE;
214 data->ocd_index = index;
216 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
218 data->ocd_connect_flags |= OBD_CONNECT_AT |
221 #ifdef HAVE_LRU_RESIZE_SUPPORT
222 OBD_CONNECT_LRU_RESIZE |
225 OBD_CONNECT_OSS_CAPA |
226 OBD_CONNECT_REQPORTAL |
227 OBD_CONNECT_SKIP_ORPHAN |
229 OBD_CONNECT_LVB_TYPE |
230 OBD_CONNECT_VERSION |
231 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 |
248 /* XXX set MDS-MDS flags, remove this when running this
250 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
251 spin_lock(&imp->imp_lock);
252 imp->imp_server_timeout = 1;
253 spin_unlock(&imp->imp_lock);
254 imp->imp_client->cli_request_portal = OUT_PORTAL;
255 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
257 ltd = &lod->lod_mdt_descs;
260 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
263 CERROR("%s: cannot connect to next dev %s (%d)\n",
264 obd->obd_name, osp, rc);
268 LASSERT(obd->obd_lu_dev);
269 LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
271 ldev = obd->obd_lu_dev;
274 /* Allocate ost descriptor and fill it */
275 OBD_ALLOC_PTR(tgt_desc);
277 GOTO(out_conn, rc = -ENOMEM);
279 tgt_desc->ltd_tgt = d;
280 tgt_desc->ltd_exp = exp;
281 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
282 tgt_desc->ltd_gen = gen;
283 tgt_desc->ltd_index = index;
284 tgt_desc->ltd_active = active;
287 if (index >= ltd->ltd_tgts_size) {
288 /* we have to increase the size of the lod_osts array */
291 newsize = max(ltd->ltd_tgts_size, (__u32)2);
292 while (newsize < index + 1)
293 newsize = newsize << 1;
295 /* lod_bitmap_resize() needs lod_rw_sem
296 * which we hold with th reference */
297 lod_putref(lod, ltd);
299 rc = ltd_bitmap_resize(ltd, newsize);
306 mutex_lock(<d->ltd_mutex);
307 if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
308 CERROR("%s: device %d is registered already\n", obd->obd_name,
310 GOTO(out_mutex, rc = -EEXIST);
313 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
314 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
315 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
316 CERROR("can't allocate index to add %s\n",
318 GOTO(out_mutex, rc = -ENOMEM);
322 if (!strcmp(LUSTRE_OSC_NAME, type)) {
323 /* pool and qos are not supported for MDS stack yet */
324 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
327 CERROR("%s: can't set up pool, failed with %d\n",
332 rc = qos_add_tgt(lod, tgt_desc);
334 CERROR("%s: qos_add_tgt failed with %d\n",
339 /* The new OST is now a full citizen */
340 if (index >= lod->lod_desc.ld_tgt_count)
341 lod->lod_desc.ld_tgt_count = index + 1;
343 lod->lod_desc.ld_active_tgt_count++;
346 LTD_TGT(ltd, index) = tgt_desc;
347 cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
349 mutex_unlock(<d->ltd_mutex);
350 lod_putref(lod, ltd);
351 if (lod->lod_recovery_completed)
352 ldev->ld_ops->ldo_recovery_complete(env, ldev);
354 rc = lfsck_add_target(env, lod->lod_child, d, exp, index, for_ost);
356 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
357 "index = %u, rc = %d\n", osp, type, index, rc);
362 lod_ost_pool_remove(&lod->lod_pool_info, index);
364 mutex_unlock(<d->ltd_mutex);
365 lod_putref(lod, ltd);
367 OBD_FREE_PTR(tgt_desc);
375 * helper function to schedule OST removal from the device table
377 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
378 struct lod_tgt_descs *ltd, unsigned idx,
381 LASSERT(LTD_TGT(ltd, idx));
383 lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
386 if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
387 LTD_TGT(ltd, idx)->ltd_reap = 1;
388 ltd->ltd_death_row++;
392 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
393 struct lod_tgt_descs *ltd, bool for_ost)
397 if (ltd->ltd_tgts_size <= 0)
400 mutex_lock(<d->ltd_mutex);
401 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
402 __lod_del_device(env, lod, ltd, idx, for_ost);
403 mutex_unlock(<d->ltd_mutex);
404 lod_putref(lod, ltd);
405 CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
406 for (idx = 0; idx < TGT_PTRS; idx++) {
407 if (ltd->ltd_tgt_idx[idx])
408 OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
410 ltd->ltd_tgts_size = 0;
415 * Add support for administratively disabled OST (through the MGS).
416 * Schedule a target for deletion. Disconnection and real removal from the
417 * table takes place in lod_putref() once the last table user release its
420 * \param env - is the environment passed by the caller
421 * \param lod - is the lod device currently connected to the OSP about to be
423 * \param osp - is the name of OSP device about to be removed
424 * \param idx - is the OSP index
425 * \param gen - is the generation number, not used currently
427 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
428 struct lod_tgt_descs *ltd, char *osp, unsigned idx,
429 unsigned gen, bool for_ost)
431 struct obd_device *obd;
433 struct obd_uuid uuid;
436 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
438 obd_str2uuid(&uuid, osp);
440 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
441 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
443 CERROR("can't find %s device\n", osp);
448 CERROR("%s: request to remove OBD %s with invalid generation %d"
449 "\n", obd->obd_name, osp, gen);
453 obd_str2uuid(&uuid, osp);
456 mutex_lock(<d->ltd_mutex);
457 /* check that the index is allocated in the bitmap */
458 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
459 !LTD_TGT(ltd, idx)) {
460 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
461 GOTO(out, rc = -EINVAL);
464 /* check that the UUID matches */
465 if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
466 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
467 obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
469 GOTO(out, rc = -EINVAL);
472 __lod_del_device(env, lod, ltd, idx, for_ost);
475 mutex_unlock(<d->ltd_mutex);
476 lod_putref(lod, ltd);
480 int lod_ea_store_resize(struct lod_thread_info *info, int size)
482 int round = size_roundup_power2(size);
484 LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
485 if (info->lti_ea_store) {
486 LASSERT(info->lti_ea_store_size);
487 LASSERT(info->lti_ea_store_size < round);
488 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
489 info->lti_ea_store_size, round);
490 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
491 info->lti_ea_store = NULL;
492 info->lti_ea_store_size = 0;
495 OBD_ALLOC_LARGE(info->lti_ea_store, round);
496 if (info->lti_ea_store == NULL)
498 info->lti_ea_store_size = round;
503 * generate and write LOV EA for given striped object
505 int lod_generate_and_set_lovea(const struct lu_env *env,
506 struct lod_object *lo, struct thandle *th)
508 struct lod_thread_info *info = lod_env_info(env);
509 struct dt_object *next = dt_object_child(&lo->ldo_obj);
510 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
511 struct lov_mds_md_v1 *lmm;
512 struct lov_ost_data_v1 *objs;
520 magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
521 lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
522 if (info->lti_ea_store_size < lmm_size) {
523 rc = lod_ea_store_resize(info, lmm_size);
528 if (lo->ldo_pattern == 0) /* default striping */
529 lo->ldo_pattern = LOV_PATTERN_RAID0;
531 lmm = info->lti_ea_store;
533 lmm->lmm_magic = cpu_to_le32(magic);
534 lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
535 fid_to_lmm_oi(fid, &lmm->lmm_oi);
536 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
537 lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
538 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
539 if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
540 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
541 lmm->lmm_layout_gen = 0;
542 if (magic == LOV_MAGIC_V1) {
543 objs = &lmm->lmm_objects[0];
545 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
546 cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
547 sizeof(v3->lmm_pool_name));
548 if (cplen >= sizeof(v3->lmm_pool_name))
550 objs = &v3->lmm_objects[0];
553 for (i = 0; i < lo->ldo_stripenr; i++) {
554 const struct lu_fid *fid;
555 struct lod_device *lod;
558 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
559 LASSERT(lo->ldo_stripe[i]);
560 fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
562 rc = fid_to_ostid(fid, &info->lti_ostid);
565 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
566 objs[i].l_ost_gen = cpu_to_le32(0);
567 rc = lod_fld_lookup(env, lod, fid, &index, LU_SEQ_RANGE_OST);
569 lod_object_free_striping(env, lo);
570 CERROR("%s: Can not locate "DFID": rc = %d\n",
571 lod2obd(lod)->obd_name, PFID(fid), rc);
574 objs[i].l_ost_idx = cpu_to_le32(index);
577 info->lti_buf.lb_buf = lmm;
578 info->lti_buf.lb_len = lmm_size;
579 rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
585 int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
587 struct lod_thread_info *info = lod_env_info(env);
588 struct dt_object *next = dt_object_child(&lo->ldo_obj);
594 if (unlikely(info->lti_ea_store_size == 0)) {
595 /* just to enter in allocation block below */
599 info->lti_buf.lb_buf = info->lti_ea_store;
600 info->lti_buf.lb_len = info->lti_ea_store_size;
601 rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV,
604 /* if object is not striped or inaccessible */
609 /* EA doesn't fit, reallocate new buffer */
610 rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV,
618 rc = lod_ea_store_resize(info, rc);
627 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
630 struct lod_thread_info *info = lod_env_info(env);
631 struct lod_object *lo = lod_dt_obj(dt);
632 struct dt_object *next = dt_object_child(dt);
633 struct lov_user_md_v3 *v3;
638 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
641 * store striping defaults into new directory
642 * used to implement defaults inheritance
645 /* probably nothing to inherite */
646 if (lo->ldo_striping_cached == 0)
649 if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
650 lo->ldo_def_stripe_offset))
653 /* XXX: use thread info */
658 v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
659 v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
660 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
661 v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
662 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
664 cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
665 sizeof(v3->lmm_pool_name));
666 if (cplen >= sizeof(v3->lmm_pool_name)) {
672 info->lti_buf.lb_buf = v3;
673 info->lti_buf.lb_len = sizeof(*v3);
674 rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
682 static int validate_lod_and_idx(struct lod_device *md, int idx)
684 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
685 !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
686 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
687 md->lod_ost_descs.ltd_tgts_size);
691 if (unlikely(OST_TGT(md, idx) == NULL)) {
692 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
693 lod2obd(md)->obd_name, idx);
697 if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
698 CERROR("%s: invalid lod device, for idx: %d\n",
699 lod2obd(md)->obd_name , idx);
707 * allocate array of objects pointers, find/create objects
708 * stripenr and other fields should be initialized by this moment
710 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
711 struct lov_ost_data_v1 *objs)
713 struct lod_thread_info *info = lod_env_info(env);
714 struct lod_device *md;
715 struct lu_object *o, *n;
716 struct lu_device *nd;
717 struct dt_object **stripe;
723 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
724 LASSERT(lo->ldo_stripe == NULL);
725 LASSERT(lo->ldo_stripenr > 0);
726 LASSERT(lo->ldo_stripe_size > 0);
728 stripe_len = lo->ldo_stripenr;
729 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
733 for (i = 0; i < lo->ldo_stripenr; i++) {
734 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
735 idx = le32_to_cpu(objs[i].l_ost_idx);
736 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
739 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
740 PFID(&info->lti_fid));
741 lod_getref(&md->lod_ost_descs);
743 rc = validate_lod_and_idx(md, idx);
744 if (unlikely(rc != 0)) {
745 lod_putref(md, &md->lod_ost_descs);
749 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
750 lod_putref(md, &md->lod_ost_descs);
752 /* In the function below, .hs_keycmp resolves to
753 * u_obj_hop_keycmp() */
754 /* coverity[overrun-buffer-val] */
755 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
757 GOTO(out, rc = PTR_ERR(o));
759 n = lu_object_locate(o->lo_header, nd->ld_type);
762 stripe[i] = container_of(n, struct dt_object, do_lu);
767 for (i = 0; i < stripe_len; i++)
768 if (stripe[i] != NULL)
769 lu_object_put(env, &stripe[i]->do_lu);
771 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
773 lo->ldo_stripe = stripe;
774 lo->ldo_stripes_allocated = stripe_len;
781 * Parse striping information stored in lti_ea_store
783 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
784 const struct lu_buf *buf)
786 struct lov_mds_md_v1 *lmm;
787 struct lov_ost_data_v1 *objs;
794 LASSERT(buf->lb_buf);
795 LASSERT(buf->lb_len);
797 lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
798 magic = le32_to_cpu(lmm->lmm_magic);
799 pattern = le32_to_cpu(lmm->lmm_pattern);
801 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
802 GOTO(out, rc = -EINVAL);
803 if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
804 GOTO(out, rc = -EINVAL);
806 lo->ldo_pattern = pattern;
807 lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
808 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
809 lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
810 /* released file stripenr fixup. */
811 if (pattern & LOV_PATTERN_F_RELEASED)
812 lo->ldo_stripenr = 0;
814 LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
816 if (magic == LOV_MAGIC_V3) {
817 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
818 objs = &v3->lmm_objects[0];
819 lod_object_set_pool(lo, v3->lmm_pool_name);
821 objs = &lmm->lmm_objects[0];
824 if (lo->ldo_stripenr > 0)
825 rc = lod_initialize_objects(env, lo, objs);
832 * Load and parse striping information, create in-core representation for the
835 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
837 struct lod_thread_info *info = lod_env_info(env);
838 struct dt_object *next = dt_object_child(&lo->ldo_obj);
843 * currently this code is supposed to be called from declaration
844 * phase only, thus the object is not expected to be locked by caller
846 dt_write_lock(env, next, 0);
847 /* already initialized? */
848 if (lo->ldo_stripe != NULL)
851 if (!dt_object_exists(next))
854 /* only regular files can be striped */
855 if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG))
858 rc = lod_get_lov_ea(env, lo);
863 * there is LOV EA (striping information) in this object
864 * let's parse it and create in-core objects for the stripes
866 info->lti_buf.lb_buf = info->lti_ea_store;
867 info->lti_buf.lb_len = info->lti_ea_store_size;
868 rc = lod_parse_striping(env, lo, &info->lti_buf);
870 dt_write_unlock(env, next);
874 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
877 struct lov_user_md_v1 *lum;
878 struct lov_user_md_v3 *lum3;
879 struct pool_desc *pool = NULL;
890 LASSERT(sizeof(*lum) < sizeof(*lum3));
892 if (buf->lb_len < sizeof(*lum)) {
893 CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md\n",
895 GOTO(out, rc = -EINVAL);
898 magic = le32_to_cpu(lum->lmm_magic);
899 if (magic != LOV_USER_MAGIC_V1 &&
900 magic != LOV_USER_MAGIC_V3 &&
901 magic != LOV_MAGIC_V1_DEF &&
902 magic != LOV_MAGIC_V3_DEF) {
903 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
904 GOTO(out, rc = -EINVAL);
907 if ((specific && le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) ||
908 (!specific && lum->lmm_pattern != 0)) {
909 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
910 le32_to_cpu(lum->lmm_pattern));
911 GOTO(out, rc = -EINVAL);
914 /* 64kB is the largest common page size we see (ia64), and matches the
916 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
917 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
918 CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
919 stripe_size, LOV_MIN_STRIPE_SIZE);
920 GOTO(out, rc = -EINVAL);
923 /* an offset of -1 is treated as a "special" valid offset */
924 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
925 if (stripe_offset != (typeof(stripe_offset))-1) {
926 /* if offset is not within valid range [0, osts_size) */
927 if (stripe_offset >= d->lod_osts_size) {
928 CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
929 stripe_offset, d->lod_osts_size);
930 GOTO(out, rc = -EINVAL);
933 /* if lmm_stripe_offset is *not* in bitmap */
934 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
935 CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
937 GOTO(out, rc = -EINVAL);
941 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
942 if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
943 lum_size = offsetof(struct lov_user_md_v1,
944 lmm_objects[stripe_count]);
945 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
946 lum_size = offsetof(struct lov_user_md_v3,
947 lmm_objects[stripe_count]);
951 if (specific && buf->lb_len != lum_size) {
952 CDEBUG(D_IOCTL, "invalid buf len %zd for lov_user_md with "
953 "magic %#x and stripe_count %u\n",
954 buf->lb_len, magic, stripe_count);
955 GOTO(out, rc = -EINVAL);
958 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
962 if (buf->lb_len < sizeof(*lum3)) {
963 CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md_v3\n",
965 GOTO(out, rc = -EINVAL);
968 /* In the function below, .hs_keycmp resolves to
969 * pool_hashkey_keycmp() */
970 /* coverity[overrun-buffer-val] */
971 pool = lod_find_pool(d, lum3->lmm_pool_name);
975 if (stripe_offset != (typeof(stripe_offset))-1) {
976 rc = lod_check_index_in_pool(stripe_offset, pool);
978 GOTO(out, rc = -EINVAL);
981 if (specific && stripe_count > pool_tgt_count(pool)) {
983 "stripe count %u > # OSTs %u in the pool\n",
984 stripe_count, pool_tgt_count(pool));
985 GOTO(out, rc = -EINVAL);
990 lod_pool_putref(pool);
995 void lod_fix_desc_stripe_size(__u64 *val)
997 if (*val < LOV_MIN_STRIPE_SIZE) {
999 LCONSOLE_INFO("Increasing default stripe size to "
1000 "minimum value %u\n",
1001 LOV_DEFAULT_STRIPE_SIZE);
1002 *val = LOV_DEFAULT_STRIPE_SIZE;
1003 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1004 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1005 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
1006 "multiple of %u)\n",
1007 *val, LOV_MIN_STRIPE_SIZE);
1011 void lod_fix_desc_stripe_count(__u32 *val)
1017 void lod_fix_desc_pattern(__u32 *val)
1019 /* from lov_setstripe */
1020 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1021 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1026 void lod_fix_desc_qos_maxage(__u32 *val)
1028 /* fix qos_maxage */
1030 *val = QOS_DEFAULT_MAXAGE;
1033 void lod_fix_desc(struct lov_desc *desc)
1035 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1036 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1037 lod_fix_desc_pattern(&desc->ld_pattern);
1038 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1041 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1043 struct obd_device *obd;
1044 struct lov_desc *desc;
1048 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1049 LASSERT(obd != NULL);
1050 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1052 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1053 CERROR("LOD setup requires a descriptor\n");
1057 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1059 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1060 CERROR("descriptor size wrong: %d > %d\n",
1061 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1065 if (desc->ld_magic != LOV_DESC_MAGIC) {
1066 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1067 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1068 obd->obd_name, desc);
1069 lustre_swab_lov_desc(desc);
1071 CERROR("%s: Bad lov desc magic: %#x\n",
1072 obd->obd_name, desc->ld_magic);
1079 desc->ld_active_tgt_count = 0;
1080 lod->lod_desc = *desc;
1082 lod->lod_sp_me = LUSTRE_SP_CLI;
1084 /* Set up allocation policy (QoS and RR) */
1085 CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1086 init_rwsem(&lod->lod_qos.lq_rw_sem);
1087 lod->lod_qos.lq_dirty = 1;
1088 lod->lod_qos.lq_rr.lqr_dirty = 1;
1089 lod->lod_qos.lq_reset = 1;
1090 /* Default priority is toward free space balance */
1091 lod->lod_qos.lq_prio_free = 232;
1092 /* Default threshold for rr (roughly 17%) */
1093 lod->lod_qos.lq_threshold_rr = 43;
1095 /* Set up OST pool environment */
1096 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1097 HASH_POOLS_MAX_BITS,
1098 HASH_POOLS_BKT_BITS, 0,
1101 &pool_hash_operations,
1103 if (lod->lod_pools_hash_body == NULL)
1106 CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
1107 lod->lod_pool_count = 0;
1108 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1111 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1113 GOTO(out_pool_info, rc);
1118 lod_ost_pool_free(&lod->lod_pool_info);
1120 cfs_hash_putref(lod->lod_pools_hash_body);
1125 int lod_pools_fini(struct lod_device *lod)
1127 struct obd_device *obd = lod2obd(lod);
1128 cfs_list_t *pos, *tmp;
1129 struct pool_desc *pool;
1132 cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
1133 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
1134 /* free pool structs */
1135 CDEBUG(D_INFO, "delete pool %p\n", pool);
1136 /* In the function below, .hs_keycmp resolves to
1137 * pool_hashkey_keycmp() */
1138 /* coverity[overrun-buffer-val] */
1139 lod_pool_del(obd, pool->pool_name);
1142 cfs_hash_putref(lod->lod_pools_hash_body);
1143 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1144 lod_ost_pool_free(&lod->lod_pool_info);