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->lod_osts 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_device *lod)
53 down_read(&lod->lod_rw_sem);
54 mutex_lock(&lod->lod_mutex);
56 mutex_unlock(&lod->lod_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)
68 mutex_lock(&lod->lod_mutex);
70 if (lod->lod_refcount == 0 && lod->lod_death_row) {
71 struct lod_ost_desc *ost_desc, *tmp;
75 CDEBUG(D_CONFIG, "destroying %d lod desc\n",
78 cfs_foreach_bit(lod->lod_ost_bitmap, idx) {
79 ost_desc = OST_TGT(lod, idx);
82 if (!ost_desc->ltd_reap)
85 cfs_list_add(&ost_desc->ltd_kill, &kill);
87 lod_ost_pool_remove(&lod->lod_pool_info, idx);
88 OST_TGT(lod, idx) = NULL;
90 cfs_bitmap_clear(lod->lod_ost_bitmap, idx);
91 if (ost_desc->ltd_active)
92 lod->lod_desc.ld_active_tgt_count--;
95 mutex_unlock(&lod->lod_mutex);
96 up_read(&lod->lod_rw_sem);
98 cfs_list_for_each_entry_safe(ost_desc, tmp, &kill, ltd_kill) {
100 cfs_list_del(&ost_desc->ltd_kill);
101 /* remove from QoS structures */
102 rc = qos_del_tgt(lod, ost_desc);
104 CERROR("%s: qos_del_tgt(%s) failed: rc = %d\n",
105 lod2obd(lod)->obd_name,
106 obd_uuid2str(&ost_desc->ltd_uuid), rc);
108 rc = obd_disconnect(ost_desc->ltd_exp);
110 CERROR("%s: failed to disconnect %s: rc = %d\n",
111 lod2obd(lod)->obd_name,
112 obd_uuid2str(&ost_desc->ltd_uuid), rc);
113 OBD_FREE_PTR(ost_desc);
116 mutex_unlock(&lod->lod_mutex);
117 up_read(&lod->lod_rw_sem);
121 static int lod_bitmap_resize(struct lod_device *lod, __u32 newsize)
123 cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
127 /* grab write reference on the lod. Relocating the array requires
128 * exclusive access */
129 down_write(&lod->lod_rw_sem);
131 if (newsize <= lod->lod_osts_size)
132 /* someone else has already resize the array */
135 /* allocate new bitmap */
136 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
138 GOTO(out, rc = -ENOMEM);
140 if (lod->lod_osts_size > 0) {
141 /* the bitmap already exists, we need
142 * to copy data from old one */
143 cfs_bitmap_copy(new_bitmap, lod->lod_ost_bitmap);
144 old_bitmap = lod->lod_ost_bitmap;
147 lod->lod_osts_size = newsize;
148 lod->lod_ost_bitmap = new_bitmap;
151 CFS_FREE_BITMAP(old_bitmap);
153 CDEBUG(D_CONFIG, "ost size: %d\n", lod->lod_osts_size);
157 up_write(&lod->lod_rw_sem);
162 * Connect LOD to a new OSP and add it to the device table.
164 * \param env - is the environment passed by the caller
165 * \param lod - is the LOD device to be connected to the new OSP
166 * \param osp - is the name of OSP device name about to be added
167 * \param index - is the OSP index
168 * \param gen - is the generation number
170 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
171 char *osp, unsigned index, unsigned gen, int active)
173 struct obd_connect_data *data = NULL;
174 struct obd_export *exp = NULL;
175 struct obd_device *obd;
176 struct lu_device *ldev;
179 struct lod_ost_desc *ost_desc;
180 struct obd_uuid obd_uuid;
184 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
187 CERROR("request to add OBD %s with invalid generation: %d\n",
192 obd_str2uuid(&obd_uuid, osp);
194 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
195 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
197 CERROR("can't find %s device\n", osp);
205 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
206 data->ocd_version = LUSTRE_VERSION_CODE;
207 data->ocd_index = index;
209 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
212 CERROR("%s: cannot connect to next dev %s (%d)\n",
213 obd->obd_name, osp, rc);
217 LASSERT(obd->obd_lu_dev);
218 LASSERT(obd->obd_lu_dev->ld_site = lod->lod_dt_dev.dd_lu_dev.ld_site);
220 ldev = obd->obd_lu_dev;
223 /* Allocate ost descriptor and fill it */
224 OBD_ALLOC_PTR(ost_desc);
226 GOTO(out_conn, rc = -ENOMEM);
228 ost_desc->ltd_ost = d;
229 ost_desc->ltd_exp = exp;
230 ost_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
231 ost_desc->ltd_gen = gen;
232 ost_desc->ltd_index = index;
233 ost_desc->ltd_active = active;
236 if (index >= lod->lod_osts_size) {
237 /* we have to increase the size of the lod_osts array */
240 newsize = max(lod->lod_osts_size, (__u32)2);
241 while (newsize < index + 1)
242 newsize = newsize << 1;
244 /* lod_bitmap_resize() needs lod_rw_sem
245 * which we hold with th reference */
248 rc = lod_bitmap_resize(lod, newsize);
255 mutex_lock(&lod->lod_mutex);
256 if (cfs_bitmap_check(lod->lod_ost_bitmap, index)) {
257 CERROR("%s: device %d is registered already\n", obd->obd_name,
259 GOTO(out_mutex, rc = -EEXIST);
262 if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) {
263 OBD_ALLOC_PTR(lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK]);
264 if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) {
265 CERROR("can't allocate index to add %s\n",
267 GOTO(out_mutex, rc = -ENOMEM);
271 rc = lod_ost_pool_add(&lod->lod_pool_info, index, lod->lod_osts_size);
273 CERROR("%s: can't set up pool, failed with %d\n",
278 rc = qos_add_tgt(lod, ost_desc);
280 CERROR("%s: qos_add_tgt(%s) failed: rc = %d\n", obd->obd_name,
281 obd_uuid2str(&ost_desc->ltd_uuid), rc);
285 /* The new OST is now a full citizen */
286 if (index >= lod->lod_desc.ld_tgt_count)
287 lod->lod_desc.ld_tgt_count = index + 1;
289 lod->lod_desc.ld_active_tgt_count++;
290 OST_TGT(lod, index) = ost_desc;
291 cfs_bitmap_set(lod->lod_ost_bitmap, index);
293 mutex_unlock(&lod->lod_mutex);
296 if (lod->lod_recovery_completed)
297 ldev->ld_ops->ldo_recovery_complete(env, ldev);
302 lod_ost_pool_remove(&lod->lod_pool_info, index);
304 mutex_unlock(&lod->lod_mutex);
307 OBD_FREE_PTR(ost_desc);
315 * helper function to schedule OST removal from the device table
317 static void __lod_del_device(struct lod_device *lod, unsigned idx)
319 LASSERT(OST_TGT(lod,idx));
320 if (OST_TGT(lod,idx)->ltd_reap == 0) {
321 OST_TGT(lod,idx)->ltd_reap = 1;
322 lod->lod_death_row++;
327 * Add support for administratively disabled OST (through the MGS).
328 * Schedule a target for deletion. Disconnection and real removal from the
329 * table takes place in lod_putref() once the last table user release its
332 * \param env - is the environment passed by the caller
333 * \param lod - is the lod device currently connected to the OSP about to be
335 * \param osp - is the name of OSP device about to be removed
336 * \param idx - is the OSP index
337 * \param gen - is the generation number, not used currently
339 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
340 char *osp, unsigned idx, unsigned gen)
342 struct obd_device *obd;
344 struct obd_uuid uuid;
347 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
349 obd = class_name2obd(osp);
351 CERROR("can't find %s device\n", osp);
356 CERROR("%s: request to remove OBD %s with invalid generation %d"
357 "\n", obd->obd_name, osp, gen);
361 obd_str2uuid(&uuid, osp);
364 mutex_lock(&lod->lod_mutex);
365 /* check that the index is allocated in the bitmap */
366 if (!cfs_bitmap_check(lod->lod_ost_bitmap, idx) || !OST_TGT(lod,idx)) {
367 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
368 GOTO(out, rc = -EINVAL);
371 /* check that the UUID matches */
372 if (!obd_uuid_equals(&uuid, &OST_TGT(lod,idx)->ltd_uuid)) {
373 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
374 obd->obd_name, obd_uuid2str(&OST_TGT(lod,idx)->ltd_uuid),
376 GOTO(out, rc = -EINVAL);
379 __lod_del_device(lod, idx);
382 mutex_unlock(&lod->lod_mutex);
387 int lod_ea_store_resize(struct lod_thread_info *info, int size)
389 int round = size_roundup_power2(size);
391 LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
392 if (info->lti_ea_store) {
393 LASSERT(info->lti_ea_store_size);
394 LASSERT(info->lti_ea_store_size < round);
395 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
396 info->lti_ea_store_size, round);
397 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
398 info->lti_ea_store = NULL;
399 info->lti_ea_store_size = 0;
402 OBD_ALLOC_LARGE(info->lti_ea_store, round);
403 if (info->lti_ea_store == NULL)
405 info->lti_ea_store_size = round;
410 * generate and write LOV EA for given striped object
412 int lod_generate_and_set_lovea(const struct lu_env *env,
413 struct lod_object *lo, struct thandle *th)
415 struct lod_thread_info *info = lod_env_info(env);
416 struct dt_object *next = dt_object_child(&lo->ldo_obj);
417 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
418 struct lov_mds_md_v1 *lmm;
419 struct lov_ost_data_v1 *objs;
425 LASSERT(lo->ldo_stripenr > 0);
427 magic = lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
428 lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
429 if (info->lti_ea_store_size < lmm_size) {
430 rc = lod_ea_store_resize(info, lmm_size);
435 lmm = info->lti_ea_store;
437 lmm->lmm_magic = cpu_to_le32(magic);
438 lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
439 lmm->lmm_object_id = cpu_to_le64(fid_ver_oid(fid));
440 lmm->lmm_object_seq = cpu_to_le64(fid_seq(fid));
441 lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
442 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
443 lmm->lmm_layout_gen = 0;
444 if (magic == LOV_MAGIC_V1) {
445 objs = &lmm->lmm_objects[0];
447 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
448 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
449 objs = &v3->lmm_objects[0];
452 for (i = 0; i < lo->ldo_stripenr; i++) {
453 const struct lu_fid *fid;
454 struct lod_device *lod;
457 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
458 LASSERT(lo->ldo_stripe[i]);
459 fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
461 rc = fid_ostid_pack(fid, &info->lti_ostid);
464 objs[i].l_object_id = cpu_to_le64(info->lti_ostid.oi_id);
465 objs[i].l_object_seq = cpu_to_le64(info->lti_ostid.oi_seq);
466 objs[i].l_ost_gen = cpu_to_le32(0);
467 rc = lod_fld_lookup(env, lod, fid, &index, LU_SEQ_RANGE_OST);
469 CERROR("%s: Can not locate "DFID": rc = %d\n",
470 lod2obd(lod)->obd_name, PFID(fid), rc);
473 objs[i].l_ost_idx = cpu_to_le32(index);
476 info->lti_buf.lb_buf = lmm;
477 info->lti_buf.lb_len = lmm_size;
478 rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
484 int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
486 struct lod_thread_info *info = lod_env_info(env);
487 struct dt_object *next = dt_object_child(&lo->ldo_obj);
493 if (unlikely(info->lti_ea_store_size == 0)) {
494 /* just to enter in allocation block below */
498 info->lti_buf.lb_buf = info->lti_ea_store;
499 info->lti_buf.lb_len = info->lti_ea_store_size;
500 rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV,
503 /* if object is not striped or inaccessible */
508 /* EA doesn't fit, reallocate new buffer */
509 rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV,
517 rc = lod_ea_store_resize(info, rc);
526 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
529 struct lod_thread_info *info = lod_env_info(env);
530 struct lod_object *lo = lod_dt_obj(dt);
531 struct dt_object *next = dt_object_child(dt);
532 struct lov_user_md_v3 *v3;
536 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
539 * store striping defaults into new directory
540 * used to implement defaults inheritance
543 /* probably nothing to inherite */
544 if (lo->ldo_striping_cached == 0)
547 if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
548 lo->ldo_def_stripe_offset))
551 /* XXX: use thread info */
556 v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
557 v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
558 v3->lmm_object_id = 0;
559 v3->lmm_object_seq = 0;
560 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
561 v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
562 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
564 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
566 info->lti_buf.lb_buf = v3;
567 info->lti_buf.lb_len = sizeof(*v3);
568 rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
577 * allocate array of objects pointers, find/create objects
578 * stripenr and other fields should be initialized by this moment
580 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
581 struct lov_ost_data_v1 *objs)
583 struct lod_thread_info *info = lod_env_info(env);
584 struct lod_device *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
585 struct lu_object *o, *n;
586 struct lu_device *nd;
591 LASSERT(lo->ldo_stripe == NULL);
592 LASSERT(lo->ldo_stripenr > 0);
593 LASSERT(lo->ldo_stripe_size > 0);
595 i = sizeof(struct dt_object *) * lo->ldo_stripenr;
596 OBD_ALLOC(lo->ldo_stripe, i);
597 if (lo->ldo_stripe == NULL)
598 GOTO(out, rc = -ENOMEM);
599 lo->ldo_stripes_allocated = lo->ldo_stripenr;
601 for (i = 0; i < lo->ldo_stripenr; i++) {
602 info->lti_ostid.oi_id = le64_to_cpu(objs[i].l_object_id);
603 info->lti_ostid.oi_seq = le64_to_cpu(objs[i].l_object_seq);
604 idx = le64_to_cpu(objs[i].l_ost_idx);
605 fid_ostid_unpack(&info->lti_fid, &info->lti_ostid, idx);
606 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
607 PFID(&info->lti_fid));
609 * XXX: assertion is left for testing, to make
610 * sure we never process requests till configuration
611 * is completed. to be changed to -EINVAL
615 LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx));
616 LASSERT(OST_TGT(md,idx));
617 LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx);
618 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
621 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
623 GOTO(out, rc = PTR_ERR(o));
625 n = lu_object_locate(o->lo_header, nd->ld_type);
628 lo->ldo_stripe[i] = container_of(n, struct dt_object, do_lu);
636 * Parse striping information stored in lti_ea_store
638 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
639 const struct lu_buf *buf)
641 struct lov_mds_md_v1 *lmm;
642 struct lov_ost_data_v1 *objs;
648 LASSERT(buf->lb_buf);
649 LASSERT(buf->lb_len);
651 lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
652 magic = le32_to_cpu(lmm->lmm_magic);
654 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
655 GOTO(out, rc = -EINVAL);
656 if (le32_to_cpu(lmm->lmm_pattern) != LOV_PATTERN_RAID0)
657 GOTO(out, rc = -EINVAL);
659 lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
660 lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
661 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
663 LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
665 if (magic == LOV_MAGIC_V3) {
666 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
667 objs = &v3->lmm_objects[0];
668 lod_object_set_pool(lo, v3->lmm_pool_name);
670 objs = &lmm->lmm_objects[0];
673 rc = lod_initialize_objects(env, lo, objs);
680 * Load and parse striping information, create in-core representation for the
683 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
685 struct lod_thread_info *info = lod_env_info(env);
686 struct dt_object *next = dt_object_child(&lo->ldo_obj);
691 * currently this code is supposed to be called from declaration
692 * phase only, thus the object is not expected to be locked by caller
694 dt_write_lock(env, next, 0);
695 /* already initialized? */
696 if (lo->ldo_stripe) {
699 for (i = 0; i < lo->ldo_stripenr; i++)
700 LASSERTF(lo->ldo_stripe[i], "stripe %d is NULL\n", i);
704 if (!dt_object_exists(next))
707 /* only regular files can be striped */
708 if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG))
711 LASSERT(lo->ldo_stripenr == 0);
713 rc = lod_get_lov_ea(env, lo);
718 * there is LOV EA (striping information) in this object
719 * let's parse it and create in-core objects for the stripes
721 info->lti_buf.lb_buf = info->lti_ea_store;
722 info->lti_buf.lb_len = info->lti_ea_store_size;
723 rc = lod_parse_striping(env, lo, &info->lti_buf);
725 dt_write_unlock(env, next);
729 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
732 struct lov_user_md_v1 *lum;
733 struct lov_user_md_v3 *v3 = NULL;
734 struct pool_desc *pool = NULL;
740 if (lum->lmm_magic != LOV_USER_MAGIC_V1 &&
741 lum->lmm_magic != LOV_USER_MAGIC_V3 &&
742 lum->lmm_magic != LOV_MAGIC_V1_DEF &&
743 lum->lmm_magic != LOV_MAGIC_V3_DEF) {
744 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n",
749 if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
750 (specific == 0 && lum->lmm_pattern != 0)) {
751 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
756 /* 64kB is the largest common page size we see (ia64), and matches the
758 if (lum->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
759 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
760 lum->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
764 /* an offset of -1 is treated as a "special" valid offset */
765 if (lum->lmm_stripe_offset != (typeof(lum->lmm_stripe_offset))(-1)) {
766 /* if offset is not within valid range [0, osts_size) */
767 if (lum->lmm_stripe_offset >= d->lod_osts_size) {
768 CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
769 lum->lmm_stripe_offset, d->lod_osts_size);
773 /* if lmm_stripe_offset is *not* in bitmap */
774 if (!cfs_bitmap_check(d->lod_ost_bitmap,
775 lum->lmm_stripe_offset)) {
776 CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
777 lum->lmm_stripe_offset);
782 if (lum->lmm_magic == LOV_USER_MAGIC_V3)
786 pool = lod_find_pool(d, v3->lmm_pool_name);
789 __u16 offs = v3->lmm_stripe_offset;
791 if (offs != (typeof(v3->lmm_stripe_offset))(-1)) {
792 rc = lod_check_index_in_pool(offs, pool);
794 lod_pool_putref(pool);
799 if (specific && lum->lmm_stripe_count > pool_tgt_count(pool)) {
801 "stripe count %u > # OSTs %u in the pool\n",
802 lum->lmm_stripe_count, pool_tgt_count(pool));
803 lod_pool_putref(pool);
807 lod_pool_putref(pool);
813 void lod_fix_desc_stripe_size(__u64 *val)
815 if (*val < PTLRPC_MAX_BRW_SIZE) {
816 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
817 PTLRPC_MAX_BRW_SIZE);
818 *val = PTLRPC_MAX_BRW_SIZE;
819 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
820 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
821 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
823 *val, LOV_MIN_STRIPE_SIZE);
827 void lod_fix_desc_stripe_count(__u32 *val)
833 void lod_fix_desc_pattern(__u32 *val)
835 /* from lov_setstripe */
836 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
837 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
842 void lod_fix_desc_qos_maxage(__u32 *val)
846 *val = QOS_DEFAULT_MAXAGE;
849 void lod_fix_desc(struct lov_desc *desc)
851 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
852 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
853 lod_fix_desc_pattern(&desc->ld_pattern);
854 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
857 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
859 struct obd_device *obd;
860 struct lov_desc *desc;
864 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
865 LASSERT(obd != NULL);
866 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
868 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
869 CERROR("LOD setup requires a descriptor\n");
873 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
875 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
876 CERROR("descriptor size wrong: %d > %d\n",
877 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
881 if (desc->ld_magic != LOV_DESC_MAGIC) {
882 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
883 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
884 obd->obd_name, desc);
885 lustre_swab_lov_desc(desc);
887 CERROR("%s: Bad lov desc magic: %#x\n",
888 obd->obd_name, desc->ld_magic);
895 desc->ld_active_tgt_count = 0;
896 lod->lod_desc = *desc;
898 lod->lod_sp_me = LUSTRE_SP_CLI;
900 /* Set up allocation policy (QoS and RR) */
901 CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
902 init_rwsem(&lod->lod_qos.lq_rw_sem);
903 lod->lod_qos.lq_dirty = 1;
904 lod->lod_qos.lq_rr.lqr_dirty = 1;
905 lod->lod_qos.lq_reset = 1;
906 /* Default priority is toward free space balance */
907 lod->lod_qos.lq_prio_free = 232;
908 /* Default threshold for rr (roughly 17%) */
909 lod->lod_qos.lq_threshold_rr = 43;
910 /* Init statfs fields */
911 OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data);
912 if (NULL == lod->lod_qos.lq_statfs_data)
914 cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq);
916 /* Set up OST pool environment */
917 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
919 HASH_POOLS_BKT_BITS, 0,
922 &pool_hash_operations,
924 if (!lod->lod_pools_hash_body)
925 GOTO(out_statfs, rc = -ENOMEM);
926 CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
927 lod->lod_pool_count = 0;
928 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
931 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
933 GOTO(out_pool_info, rc);
935 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
936 * added to the LOD, see lod_add_device() */
937 lod->lod_ost_bitmap = NULL;
938 lod->lod_osts_size = 0;
941 lod->lod_death_row = 0;
942 lod->lod_refcount = 0;
947 lod_ost_pool_free(&lod->lod_pool_info);
949 cfs_hash_putref(lod->lod_pools_hash_body);
951 OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
955 int lod_pools_fini(struct lod_device *lod)
957 struct obd_device *obd = lod2obd(lod);
958 cfs_list_t *pos, *tmp;
959 struct pool_desc *pool;
962 cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
963 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
964 /* free pool structs */
965 CDEBUG(D_INFO, "delete pool %p\n", pool);
966 lod_pool_del(obd, pool->pool_name);
969 if (lod->lod_osts_size > 0) {
972 mutex_lock(&lod->lod_mutex);
973 cfs_foreach_bit(lod->lod_ost_bitmap, idx)
974 __lod_del_device(lod, idx);
975 mutex_unlock(&lod->lod_mutex);
977 CFS_FREE_BITMAP(lod->lod_ost_bitmap);
978 for (idx = 0; idx < OST_PTRS; idx++) {
979 if (lod->lod_ost_idx[idx])
980 OBD_FREE_PTR(lod->lod_ost_idx[idx]);
982 lod->lod_osts_size = 0;
985 cfs_hash_putref(lod->lod_pools_hash_body);
986 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
987 lod_ost_pool_free(&lod->lod_pool_info);
988 OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);