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, Intel Corporation.
29 * lustre/lod/lod_lov.c
31 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
35 # define EXPORT_SYMTAB
37 #define DEBUG_SUBSYSTEM S_MDS
39 #include <obd_class.h>
42 #include "lod_internal.h"
45 * Keep a refcount of lod->ltd_tgts usage to prevent racing with
46 * addition/deletion. Any function that expects lov_tgts to remain stationary
49 * \param lod - is the lod device from which we want to grab a reference
51 void lod_getref(struct lod_tgt_descs *ltd)
53 down_read(<d->ltd_rw_sem);
54 mutex_lock(<d->ltd_mutex);
56 mutex_unlock(<d->ltd_mutex);
60 * Companion of lod_getref() to release a reference on the lod table.
61 * If this is the last reference and the ost entry was scheduled for deletion,
62 * the descriptor is removed from the array.
64 * \param lod - is the lod device from which we release a reference
66 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
68 mutex_lock(<d->ltd_mutex);
70 if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
71 struct lod_tgt_desc *tgt_desc, *tmp;
75 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
78 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
79 tgt_desc = LTD_TGT(ltd, idx);
82 if (!tgt_desc->ltd_reap)
85 cfs_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 cfs_list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
102 cfs_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
175 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
176 char *osp, unsigned index, unsigned gen, 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 = &lod->lod_ost_descs;
186 struct obd_uuid obd_uuid;
189 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
192 CERROR("request to add OBD %s with invalid generation: %d\n",
197 obd_str2uuid(&obd_uuid, osp);
199 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
200 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
202 CERROR("can't find %s device\n", osp);
210 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
211 data->ocd_version = LUSTRE_VERSION_CODE;
212 data->ocd_index = index;
214 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
217 CERROR("%s: cannot connect to next dev %s (%d)\n",
218 obd->obd_name, osp, rc);
222 LASSERT(obd->obd_lu_dev);
223 LASSERT(obd->obd_lu_dev->ld_site = lod->lod_dt_dev.dd_lu_dev.ld_site);
225 ldev = obd->obd_lu_dev;
228 /* Allocate ost descriptor and fill it */
229 OBD_ALLOC_PTR(tgt_desc);
231 GOTO(out_conn, rc = -ENOMEM);
233 tgt_desc->ltd_tgt = d;
234 tgt_desc->ltd_exp = exp;
235 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
236 tgt_desc->ltd_gen = gen;
237 tgt_desc->ltd_index = index;
238 tgt_desc->ltd_active = active;
241 if (index >= ltd->ltd_tgts_size) {
242 /* we have to increase the size of the lod_osts array */
245 newsize = max(ltd->ltd_tgts_size, (__u32)2);
246 while (newsize < index + 1)
247 newsize = newsize << 1;
249 /* lod_bitmap_resize() needs lod_rw_sem
250 * which we hold with th reference */
251 lod_putref(lod, ltd);
253 rc = ltd_bitmap_resize(ltd, newsize);
260 mutex_lock(<d->ltd_mutex);
261 if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
262 CERROR("%s: device %d is registered already\n", obd->obd_name,
264 GOTO(out_mutex, rc = -EEXIST);
267 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
268 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
269 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
270 CERROR("can't allocate index to add %s\n",
272 GOTO(out_mutex, rc = -ENOMEM);
276 /* pool and qos are not supported for MDS stack yet */
277 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
280 CERROR("%s: can't set up pool, failed with %d\n",
285 rc = qos_add_tgt(lod, tgt_desc);
287 CERROR("%s: qos_add_tgt failed with %d\n",
292 /* The new OST is now a full citizen */
293 if (index >= lod->lod_desc.ld_tgt_count)
294 lod->lod_desc.ld_tgt_count = index + 1;
296 lod->lod_desc.ld_active_tgt_count++;
298 LTD_TGT(ltd, index) = tgt_desc;
299 cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
301 mutex_unlock(<d->ltd_mutex);
302 lod_putref(lod, ltd);
303 if (lod->lod_recovery_completed)
304 ldev->ld_ops->ldo_recovery_complete(env, ldev);
309 lod_ost_pool_remove(&lod->lod_pool_info, index);
311 mutex_unlock(<d->ltd_mutex);
312 lod_putref(lod, ltd);
314 OBD_FREE_PTR(tgt_desc);
322 * helper function to schedule OST removal from the device table
324 static void __lod_del_device(struct lod_tgt_descs *ltd,
327 LASSERT(LTD_TGT(ltd, idx));
328 if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
329 LTD_TGT(ltd, idx)->ltd_reap = 1;
330 ltd->ltd_death_row++;
334 int lod_fini_tgt(struct lod_device *lod, struct lod_tgt_descs *ltd)
338 if (ltd->ltd_tgts_size <= 0)
341 mutex_lock(<d->ltd_mutex);
342 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
343 __lod_del_device(ltd, idx);
344 mutex_unlock(<d->ltd_mutex);
345 lod_putref(lod, ltd);
346 CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
347 for (idx = 0; idx < TGT_PTRS; idx++) {
348 if (ltd->ltd_tgt_idx[idx])
349 OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
351 ltd->ltd_tgts_size = 0;
356 * Add support for administratively disabled OST (through the MGS).
357 * Schedule a target for deletion. Disconnection and real removal from the
358 * table takes place in lod_putref() once the last table user release its
361 * \param env - is the environment passed by the caller
362 * \param lod - is the lod device currently connected to the OSP about to be
364 * \param osp - is the name of OSP device about to be removed
365 * \param idx - is the OSP index
366 * \param gen - is the generation number, not used currently
368 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
369 struct lod_tgt_descs *ltd, char *osp, unsigned idx,
372 struct obd_device *obd;
374 struct obd_uuid uuid;
377 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
379 obd = class_name2obd(osp);
381 CERROR("can't find %s device\n", osp);
386 CERROR("%s: request to remove OBD %s with invalid generation %d"
387 "\n", obd->obd_name, osp, gen);
391 obd_str2uuid(&uuid, osp);
394 mutex_lock(<d->ltd_mutex);
395 /* check that the index is allocated in the bitmap */
396 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
397 !LTD_TGT(ltd, idx)) {
398 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
399 GOTO(out, rc = -EINVAL);
402 /* check that the UUID matches */
403 if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
404 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
405 obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
407 GOTO(out, rc = -EINVAL);
410 __lod_del_device(ltd, idx);
413 mutex_unlock(<d->ltd_mutex);
414 lod_putref(lod, ltd);
418 int lod_ea_store_resize(struct lod_thread_info *info, int size)
420 int round = size_roundup_power2(size);
422 LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
423 if (info->lti_ea_store) {
424 LASSERT(info->lti_ea_store_size);
425 LASSERT(info->lti_ea_store_size < round);
426 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
427 info->lti_ea_store_size, round);
428 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
429 info->lti_ea_store = NULL;
430 info->lti_ea_store_size = 0;
433 OBD_ALLOC_LARGE(info->lti_ea_store, round);
434 if (info->lti_ea_store == NULL)
436 info->lti_ea_store_size = round;
441 * generate and write LOV EA for given striped object
443 int lod_generate_and_set_lovea(const struct lu_env *env,
444 struct lod_object *lo, struct thandle *th)
446 struct lod_thread_info *info = lod_env_info(env);
447 struct dt_object *next = dt_object_child(&lo->ldo_obj);
448 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
449 struct lov_mds_md_v1 *lmm;
450 struct lov_ost_data_v1 *objs;
456 LASSERT(lo->ldo_stripenr > 0);
458 magic = lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
459 lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
460 if (info->lti_ea_store_size < lmm_size) {
461 rc = lod_ea_store_resize(info, lmm_size);
466 lmm = info->lti_ea_store;
468 lmm->lmm_magic = cpu_to_le32(magic);
469 lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
470 lmm->lmm_object_id = cpu_to_le64(fid_ver_oid(fid));
471 lmm->lmm_object_seq = cpu_to_le64(fid_seq(fid));
472 lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
473 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
474 lmm->lmm_layout_gen = 0;
475 if (magic == LOV_MAGIC_V1) {
476 objs = &lmm->lmm_objects[0];
478 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
479 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
480 objs = &v3->lmm_objects[0];
483 for (i = 0; i < lo->ldo_stripenr; i++) {
484 const struct lu_fid *fid;
485 struct lod_device *lod;
488 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
489 LASSERT(lo->ldo_stripe[i]);
490 fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
492 rc = fid_ostid_pack(fid, &info->lti_ostid);
495 objs[i].l_object_id = cpu_to_le64(info->lti_ostid.oi_id);
496 objs[i].l_object_seq = cpu_to_le64(info->lti_ostid.oi_seq);
497 objs[i].l_ost_gen = cpu_to_le32(0);
498 rc = lod_fld_lookup(env, lod, fid, &index, LU_SEQ_RANGE_OST);
500 CERROR("%s: Can not locate "DFID": rc = %d\n",
501 lod2obd(lod)->obd_name, PFID(fid), rc);
504 objs[i].l_ost_idx = cpu_to_le32(index);
507 info->lti_buf.lb_buf = lmm;
508 info->lti_buf.lb_len = lmm_size;
509 rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
515 int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
517 struct lod_thread_info *info = lod_env_info(env);
518 struct dt_object *next = dt_object_child(&lo->ldo_obj);
524 if (unlikely(info->lti_ea_store_size == 0)) {
525 /* just to enter in allocation block below */
529 info->lti_buf.lb_buf = info->lti_ea_store;
530 info->lti_buf.lb_len = info->lti_ea_store_size;
531 rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV,
534 /* if object is not striped or inaccessible */
539 /* EA doesn't fit, reallocate new buffer */
540 rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV,
548 rc = lod_ea_store_resize(info, rc);
557 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
560 struct lod_thread_info *info = lod_env_info(env);
561 struct lod_object *lo = lod_dt_obj(dt);
562 struct dt_object *next = dt_object_child(dt);
563 struct lov_user_md_v3 *v3;
567 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
570 * store striping defaults into new directory
571 * used to implement defaults inheritance
574 /* probably nothing to inherite */
575 if (lo->ldo_striping_cached == 0)
578 if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
579 lo->ldo_def_stripe_offset))
582 /* XXX: use thread info */
587 v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
588 v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
589 v3->lmm_object_id = 0;
590 v3->lmm_object_seq = 0;
591 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
592 v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
593 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
595 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
597 info->lti_buf.lb_buf = v3;
598 info->lti_buf.lb_len = sizeof(*v3);
599 rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
608 * allocate array of objects pointers, find/create objects
609 * stripenr and other fields should be initialized by this moment
611 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
612 struct lov_ost_data_v1 *objs)
614 struct lod_thread_info *info = lod_env_info(env);
615 struct lod_device *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
616 struct lu_object *o, *n;
617 struct lu_device *nd;
622 LASSERT(lo->ldo_stripe == NULL);
623 LASSERT(lo->ldo_stripenr > 0);
624 LASSERT(lo->ldo_stripe_size > 0);
626 i = sizeof(struct dt_object *) * lo->ldo_stripenr;
627 OBD_ALLOC(lo->ldo_stripe, i);
628 if (lo->ldo_stripe == NULL)
629 GOTO(out, rc = -ENOMEM);
630 lo->ldo_stripes_allocated = lo->ldo_stripenr;
632 for (i = 0; i < lo->ldo_stripenr; i++) {
633 info->lti_ostid.oi_id = le64_to_cpu(objs[i].l_object_id);
634 info->lti_ostid.oi_seq = le64_to_cpu(objs[i].l_object_seq);
635 idx = le64_to_cpu(objs[i].l_ost_idx);
636 fid_ostid_unpack(&info->lti_fid, &info->lti_ostid, idx);
637 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
638 PFID(&info->lti_fid));
640 * XXX: assertion is left for testing, to make
641 * sure we never process requests till configuration
642 * is completed. to be changed to -EINVAL
645 lod_getref(&md->lod_ost_descs);
646 LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx));
647 LASSERT(OST_TGT(md,idx));
648 LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx);
649 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
650 lod_putref(md, &md->lod_ost_descs);
652 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
654 GOTO(out, rc = PTR_ERR(o));
656 n = lu_object_locate(o->lo_header, nd->ld_type);
659 lo->ldo_stripe[i] = container_of(n, struct dt_object, do_lu);
667 * Parse striping information stored in lti_ea_store
669 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
670 const struct lu_buf *buf)
672 struct lov_mds_md_v1 *lmm;
673 struct lov_ost_data_v1 *objs;
679 LASSERT(buf->lb_buf);
680 LASSERT(buf->lb_len);
682 lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
683 magic = le32_to_cpu(lmm->lmm_magic);
685 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
686 GOTO(out, rc = -EINVAL);
687 if (le32_to_cpu(lmm->lmm_pattern) != LOV_PATTERN_RAID0)
688 GOTO(out, rc = -EINVAL);
690 lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
691 lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
692 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
694 LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
696 if (magic == LOV_MAGIC_V3) {
697 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
698 objs = &v3->lmm_objects[0];
699 lod_object_set_pool(lo, v3->lmm_pool_name);
701 objs = &lmm->lmm_objects[0];
704 rc = lod_initialize_objects(env, lo, objs);
711 * Load and parse striping information, create in-core representation for the
714 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
716 struct lod_thread_info *info = lod_env_info(env);
717 struct dt_object *next = dt_object_child(&lo->ldo_obj);
722 * currently this code is supposed to be called from declaration
723 * phase only, thus the object is not expected to be locked by caller
725 dt_write_lock(env, next, 0);
726 /* already initialized? */
727 if (lo->ldo_stripe) {
730 for (i = 0; i < lo->ldo_stripenr; i++)
731 LASSERTF(lo->ldo_stripe[i], "stripe %d is NULL\n", i);
735 if (!dt_object_exists(next))
738 /* only regular files can be striped */
739 if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG))
742 LASSERT(lo->ldo_stripenr == 0);
744 rc = lod_get_lov_ea(env, lo);
749 * there is LOV EA (striping information) in this object
750 * let's parse it and create in-core objects for the stripes
752 info->lti_buf.lb_buf = info->lti_ea_store;
753 info->lti_buf.lb_len = info->lti_ea_store_size;
754 rc = lod_parse_striping(env, lo, &info->lti_buf);
756 dt_write_unlock(env, next);
760 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
763 struct lov_user_md_v1 *lum;
764 struct lov_user_md_v3 *v3 = NULL;
765 struct pool_desc *pool = NULL;
771 if (lum->lmm_magic != LOV_USER_MAGIC_V1 &&
772 lum->lmm_magic != LOV_USER_MAGIC_V3 &&
773 lum->lmm_magic != LOV_MAGIC_V1_DEF &&
774 lum->lmm_magic != LOV_MAGIC_V3_DEF) {
775 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n",
780 if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
781 (specific == 0 && lum->lmm_pattern != 0)) {
782 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
787 /* 64kB is the largest common page size we see (ia64), and matches the
789 if (lum->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
790 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
791 lum->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
795 /* an offset of -1 is treated as a "special" valid offset */
796 if (lum->lmm_stripe_offset != (typeof(lum->lmm_stripe_offset))(-1)) {
797 /* if offset is not within valid range [0, osts_size) */
798 if (lum->lmm_stripe_offset >= d->lod_osts_size) {
799 CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
800 lum->lmm_stripe_offset, d->lod_osts_size);
804 /* if lmm_stripe_offset is *not* in bitmap */
805 if (!cfs_bitmap_check(d->lod_ost_bitmap,
806 lum->lmm_stripe_offset)) {
807 CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
808 lum->lmm_stripe_offset);
813 if (lum->lmm_magic == LOV_USER_MAGIC_V3)
817 pool = lod_find_pool(d, v3->lmm_pool_name);
820 __u16 offs = v3->lmm_stripe_offset;
822 if (offs != (typeof(v3->lmm_stripe_offset))(-1)) {
823 rc = lod_check_index_in_pool(offs, pool);
825 lod_pool_putref(pool);
830 if (specific && lum->lmm_stripe_count > pool_tgt_count(pool)) {
832 "stripe count %u > # OSTs %u in the pool\n",
833 lum->lmm_stripe_count, pool_tgt_count(pool));
834 lod_pool_putref(pool);
838 lod_pool_putref(pool);
844 void lod_fix_desc_stripe_size(__u64 *val)
846 if (*val < PTLRPC_MAX_BRW_SIZE) {
847 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
848 PTLRPC_MAX_BRW_SIZE);
849 *val = PTLRPC_MAX_BRW_SIZE;
850 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
851 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
852 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
854 *val, LOV_MIN_STRIPE_SIZE);
858 void lod_fix_desc_stripe_count(__u32 *val)
864 void lod_fix_desc_pattern(__u32 *val)
866 /* from lov_setstripe */
867 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
868 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
873 void lod_fix_desc_qos_maxage(__u32 *val)
877 *val = QOS_DEFAULT_MAXAGE;
880 void lod_fix_desc(struct lov_desc *desc)
882 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
883 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
884 lod_fix_desc_pattern(&desc->ld_pattern);
885 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
888 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
890 struct obd_device *obd;
891 struct lov_desc *desc;
895 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
896 LASSERT(obd != NULL);
897 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
899 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
900 CERROR("LOD setup requires a descriptor\n");
904 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
906 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
907 CERROR("descriptor size wrong: %d > %d\n",
908 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
912 if (desc->ld_magic != LOV_DESC_MAGIC) {
913 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
914 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
915 obd->obd_name, desc);
916 lustre_swab_lov_desc(desc);
918 CERROR("%s: Bad lov desc magic: %#x\n",
919 obd->obd_name, desc->ld_magic);
926 desc->ld_active_tgt_count = 0;
927 lod->lod_desc = *desc;
929 lod->lod_sp_me = LUSTRE_SP_CLI;
931 /* Set up allocation policy (QoS and RR) */
932 CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
933 init_rwsem(&lod->lod_qos.lq_rw_sem);
934 lod->lod_qos.lq_dirty = 1;
935 lod->lod_qos.lq_rr.lqr_dirty = 1;
936 lod->lod_qos.lq_reset = 1;
937 /* Default priority is toward free space balance */
938 lod->lod_qos.lq_prio_free = 232;
939 /* Default threshold for rr (roughly 17%) */
940 lod->lod_qos.lq_threshold_rr = 43;
941 /* Init statfs fields */
942 OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data);
943 if (NULL == lod->lod_qos.lq_statfs_data)
945 cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq);
947 /* Set up OST pool environment */
948 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
950 HASH_POOLS_BKT_BITS, 0,
953 &pool_hash_operations,
955 if (!lod->lod_pools_hash_body)
956 GOTO(out_statfs, rc = -ENOMEM);
957 CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
958 lod->lod_pool_count = 0;
959 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
962 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
964 GOTO(out_pool_info, rc);
969 lod_ost_pool_free(&lod->lod_pool_info);
971 cfs_hash_putref(lod->lod_pools_hash_body);
973 OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
977 int lod_pools_fini(struct lod_device *lod)
979 struct obd_device *obd = lod2obd(lod);
980 cfs_list_t *pos, *tmp;
981 struct pool_desc *pool;
984 cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
985 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
986 /* free pool structs */
987 CDEBUG(D_INFO, "delete pool %p\n", pool);
988 lod_pool_del(obd, pool->pool_name);
991 cfs_hash_putref(lod->lod_pools_hash_body);
992 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
993 lod_ost_pool_free(&lod->lod_pool_info);
994 OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);