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.
27 * lustre/lod/lod_lov.c
29 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
33 # define EXPORT_SYMTAB
35 #define DEBUG_SUBSYSTEM S_MDS
37 #include <obd_class.h>
40 #include "lod_internal.h"
43 * Keep a refcount of lod->lod_osts 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_device *lod)
51 cfs_down_read(&lod->lod_rw_sem);
52 cfs_mutex_lock(&lod->lod_mutex);
54 cfs_mutex_unlock(&lod->lod_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)
66 cfs_mutex_lock(&lod->lod_mutex);
68 if (lod->lod_refcount == 0 && lod->lod_death_row) {
69 struct lod_ost_desc *ost_desc, *tmp;
73 CDEBUG(D_CONFIG, "destroying %d lod desc\n",
76 cfs_foreach_bit(lod->lod_ost_bitmap, idx) {
77 ost_desc = OST_TGT(lod, idx);
80 if (!ost_desc->ltd_reap)
83 cfs_list_add(&ost_desc->ltd_kill, &kill);
85 lod_ost_pool_remove(&lod->lod_pool_info, idx);
86 OST_TGT(lod, idx) = NULL;
88 cfs_bitmap_clear(lod->lod_ost_bitmap, idx);
89 if (ost_desc->ltd_active)
90 lod->lod_desc.ld_active_tgt_count--;
93 cfs_mutex_unlock(&lod->lod_mutex);
94 cfs_up_read(&lod->lod_rw_sem);
96 cfs_list_for_each_entry_safe(ost_desc, tmp, &kill, ltd_kill) {
98 cfs_list_del(&ost_desc->ltd_kill);
99 /* XXX: remove from QoS structures */
100 /* disconnect from OSP */
101 rc = obd_disconnect(ost_desc->ltd_exp);
103 CERROR("%s: failed to disconnect %s (%d)\n",
104 lod2obd(lod)->obd_name,
105 obd_uuid2str(&ost_desc->ltd_uuid), rc);
106 OBD_FREE_PTR(ost_desc);
109 cfs_mutex_unlock(&lod->lod_mutex);
110 cfs_up_read(&lod->lod_rw_sem);
114 static int lod_bitmap_resize(struct lod_device *lod, __u32 newsize)
116 cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
120 /* grab write reference on the lod. Relocating the array requires
121 * exclusive access */
122 cfs_down_write(&lod->lod_rw_sem);
124 if (newsize <= lod->lod_osts_size)
125 /* someone else has already resize the array */
128 /* allocate new bitmap */
129 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
131 GOTO(out, rc = -ENOMEM);
133 if (lod->lod_osts_size > 0) {
134 /* the bitmap already exists, we need
135 * to copy data from old one */
136 cfs_bitmap_copy(new_bitmap, lod->lod_ost_bitmap);
137 old_bitmap = lod->lod_ost_bitmap;
140 lod->lod_osts_size = newsize;
141 lod->lod_ost_bitmap = new_bitmap;
144 CFS_FREE_BITMAP(old_bitmap);
146 CDEBUG(D_CONFIG, "ost size: %d\n", lod->lod_osts_size);
150 cfs_up_write(&lod->lod_rw_sem);
155 * Connect LOD to a new OSP and add it to the device table.
157 * \param env - is the environment passed by the caller
158 * \param lod - is the LOD device to be connected to the new OSP
159 * \param osp - is the name of OSP device name about to be added
160 * \param index - is the OSP index
161 * \param gen - is the generation number
163 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
164 char *osp, unsigned index, unsigned gen, int active)
166 struct obd_connect_data *data = NULL;
167 struct obd_export *exp = NULL;
168 struct obd_device *obd;
169 struct lu_device *ldev;
172 struct lod_ost_desc *ost_desc;
173 struct obd_uuid obd_uuid;
177 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
180 CERROR("request to add OBD %s with invalid generation: %d\n",
185 obd_str2uuid(&obd_uuid, osp);
187 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
188 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
190 CERROR("can't find %s device\n", osp);
198 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
199 data->ocd_version = LUSTRE_VERSION_CODE;
200 data->ocd_index = index;
202 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
205 CERROR("%s: cannot connect to next dev %s (%d)\n",
206 obd->obd_name, osp, rc);
210 LASSERT(obd->obd_lu_dev);
211 LASSERT(obd->obd_lu_dev->ld_site = lod->lod_dt_dev.dd_lu_dev.ld_site);
213 ldev = obd->obd_lu_dev;
216 /* Allocate ost descriptor and fill it */
217 OBD_ALLOC_PTR(ost_desc);
219 GOTO(out_conn, rc = -ENOMEM);
221 ost_desc->ltd_ost = d;
222 ost_desc->ltd_exp = exp;
223 ost_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
224 ost_desc->ltd_gen = gen;
225 ost_desc->ltd_index = index;
226 ost_desc->ltd_active = active;
229 if (index >= lod->lod_osts_size) {
230 /* we have to increase the size of the lod_osts array */
233 newsize = max(lod->lod_osts_size, (__u32)2);
234 while (newsize < index + 1)
235 newsize = newsize << 1;
237 /* lod_bitmap_resize() needs lod_rw_sem
238 * which we hold with th reference */
241 rc = lod_bitmap_resize(lod, newsize);
248 cfs_mutex_lock(&lod->lod_mutex);
249 if (cfs_bitmap_check(lod->lod_ost_bitmap, index)) {
250 CERROR("%s: device %d is registered already\n", obd->obd_name,
252 GOTO(out_mutex, rc = -EEXIST);
255 if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) {
256 OBD_ALLOC_PTR(lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK]);
257 if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) {
258 CERROR("can't allocate index to add %s\n",
260 GOTO(out_mutex, rc = -ENOMEM);
264 rc = lod_ost_pool_add(&lod->lod_pool_info, index, lod->lod_osts_size);
266 CERROR("%s: can't set up pool, failed with %d\n",
271 /* The new OST is now a full citizen */
272 if (index >= lod->lod_desc.ld_tgt_count)
273 lod->lod_desc.ld_tgt_count = index + 1;
275 lod->lod_desc.ld_active_tgt_count++;
276 OST_TGT(lod, index) = ost_desc;
277 cfs_bitmap_set(lod->lod_ost_bitmap, index);
279 cfs_mutex_unlock(&lod->lod_mutex);
282 if (lod->lod_recovery_completed)
283 ldev->ld_ops->ldo_recovery_complete(env, ldev);
287 lod_ost_pool_remove(&lod->lod_pool_info, index);
289 cfs_mutex_unlock(&lod->lod_mutex);
292 OBD_FREE_PTR(ost_desc);
300 * helper function to schedule OST removal from the device table
302 static void __lod_del_device(struct lod_device *lod, unsigned idx)
304 LASSERT(OST_TGT(lod,idx));
305 if (OST_TGT(lod,idx)->ltd_reap == 0) {
306 OST_TGT(lod,idx)->ltd_reap = 1;
307 lod->lod_death_row++;
312 * Add support for administratively disabled OST (through the MGS).
313 * Schedule a target for deletion. Disconnection and real removal from the
314 * table takes place in lod_putref() once the last table user release its
317 * \param env - is the environment passed by the caller
318 * \param lod - is the lod device currently connected to the OSP about to be
320 * \param osp - is the name of OSP device about to be removed
321 * \param idx - is the OSP index
322 * \param gen - is the generation number, not used currently
324 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
325 char *osp, unsigned idx, unsigned gen)
327 struct obd_device *obd;
329 struct obd_uuid uuid;
332 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
334 obd = class_name2obd(osp);
336 CERROR("can't find %s device\n", osp);
341 CERROR("%s: request to remove OBD %s with invalid generation %d"
342 "\n", obd->obd_name, osp, gen);
346 obd_str2uuid(&uuid, osp);
349 cfs_mutex_lock(&lod->lod_mutex);
350 /* check that the index is allocated in the bitmap */
351 if (!cfs_bitmap_check(lod->lod_ost_bitmap, idx) || !OST_TGT(lod,idx)) {
352 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
353 GOTO(out, rc = -EINVAL);
356 /* check that the UUID matches */
357 if (!obd_uuid_equals(&uuid, &OST_TGT(lod,idx)->ltd_uuid)) {
358 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
359 obd->obd_name, obd_uuid2str(&OST_TGT(lod,idx)->ltd_uuid),
361 GOTO(out, rc = -EINVAL);
364 __lod_del_device(lod, idx);
367 cfs_mutex_unlock(&lod->lod_mutex);
372 void lod_fix_desc_stripe_size(__u64 *val)
374 if (*val < PTLRPC_MAX_BRW_SIZE) {
375 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
376 PTLRPC_MAX_BRW_SIZE);
377 *val = PTLRPC_MAX_BRW_SIZE;
378 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
379 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
380 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
382 *val, LOV_MIN_STRIPE_SIZE);
386 void lod_fix_desc_stripe_count(__u32 *val)
392 void lod_fix_desc_pattern(__u32 *val)
394 /* from lov_setstripe */
395 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
396 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
401 void lod_fix_desc_qos_maxage(__u32 *val)
405 *val = QOS_DEFAULT_MAXAGE;
408 void lod_fix_desc(struct lov_desc *desc)
410 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
411 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
412 lod_fix_desc_pattern(&desc->ld_pattern);
413 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
416 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
418 struct lprocfs_static_vars lvars = { 0 };
419 struct obd_device *obd;
420 struct lov_desc *desc;
424 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
425 LASSERT(obd != NULL);
426 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
428 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
429 CERROR("LOD setup requires a descriptor\n");
433 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
435 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
436 CERROR("descriptor size wrong: %d > %d\n",
437 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
441 if (desc->ld_magic != LOV_DESC_MAGIC) {
442 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
443 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
444 obd->obd_name, desc);
445 lustre_swab_lov_desc(desc);
447 CERROR("%s: Bad lov desc magic: %#x\n",
448 obd->obd_name, desc->ld_magic);
455 desc->ld_active_tgt_count = 0;
456 lod->lod_desc = *desc;
458 lod->lod_sp_me = LUSTRE_SP_CLI;
460 /* Set up allocation policy (QoS and RR) */
461 CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
462 cfs_init_rwsem(&lod->lod_qos.lq_rw_sem);
463 lod->lod_qos.lq_dirty = 1;
464 lod->lod_qos.lq_rr.lqr_dirty = 1;
465 lod->lod_qos.lq_reset = 1;
466 /* Default priority is toward free space balance */
467 lod->lod_qos.lq_prio_free = 232;
468 /* Default threshold for rr (roughly 17%) */
469 lod->lod_qos.lq_threshold_rr = 43;
470 /* Init statfs fields */
471 OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data);
472 if (NULL == lod->lod_qos.lq_statfs_data)
474 cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq);
476 /* Set up OST pool environment */
477 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
479 HASH_POOLS_BKT_BITS, 0,
482 &pool_hash_operations,
484 if (!lod->lod_pools_hash_body)
485 GOTO(out_statfs, rc = -ENOMEM);
486 CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
487 lod->lod_pool_count = 0;
488 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
491 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
493 GOTO(out_pool_info, rc);
495 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
496 * added to the LOD, see lod_add_device() */
497 lod->lod_ost_bitmap = NULL;
498 lod->lod_osts_size = 0;
501 lod->lod_death_row = 0;
502 lod->lod_refcount = 0;
504 lprocfs_lod_init_vars(&lvars);
505 lprocfs_obd_setup(obd, lvars.obd_vars);
508 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
509 0444, &lod_proc_target_fops, obd);
511 CWARN("%s: Error adding the target_obd file %d\n",
515 lod->lod_pool_proc_entry = lprocfs_register("pools",
518 if (IS_ERR(lod->lod_pool_proc_entry)) {
519 int ret = PTR_ERR(lod->lod_pool_proc_entry);
520 lod->lod_pool_proc_entry = NULL;
521 CWARN("%s: Failed to create pool proc file %d\n",
523 rc = lod_pools_fini(lod);
531 lprocfs_obd_cleanup(obd);
532 lod_ost_pool_free(&lod->lod_qos.lq_rr.lqr_pool);
534 lod_ost_pool_free(&lod->lod_pool_info);
536 cfs_hash_putref(lod->lod_pools_hash_body);
538 OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
542 int lod_pools_fini(struct lod_device *lod)
544 struct obd_device *obd = lod2obd(lod);
545 cfs_list_t *pos, *tmp;
546 struct pool_desc *pool;
549 cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
550 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
551 /* free pool structs */
552 CDEBUG(D_INFO, "delete pool %p\n", pool);
553 lod_pool_del(obd, pool->pool_name);
556 if (lod->lod_osts_size > 0) {
559 cfs_mutex_lock(&lod->lod_mutex);
560 cfs_foreach_bit(lod->lod_ost_bitmap, idx)
561 __lod_del_device(lod, idx);
562 cfs_mutex_unlock(&lod->lod_mutex);
564 CFS_FREE_BITMAP(lod->lod_ost_bitmap);
565 for (idx = 0; idx < OST_PTRS; idx++) {
566 if (lod->lod_ost_idx[idx])
567 OBD_FREE_PTR(lod->lod_ost_idx[idx]);
569 lod->lod_osts_size = 0;
572 cfs_hash_putref(lod->lod_pools_hash_body);
573 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
574 lod_ost_pool_free(&lod->lod_pool_info);
576 /* clear pools parent proc entry only after all pools are killed */
577 if (lod->lod_pool_proc_entry) {
578 lprocfs_remove(&lod->lod_pool_proc_entry);
579 lod->lod_pool_proc_entry = NULL;
582 lprocfs_obd_cleanup(obd);
584 OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);