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, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see [sun.com URL with a
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * This file is part of Lustre, http://www.lustre.org/
32 * Lustre is a trademark of Sun Microsystems, Inc.
34 * lustre/lov/lov_pool.c
38 * Author: Jacques-Charles LAFOUCRIERE <jc.lafoucriere@cea.fr>
39 * Author: Alex Lyashkov <Alexey.Lyashkov@Sun.COM>
40 * Author: Nathaniel Rutman <Nathan.Rutman@Sun.COM>
43 #define DEBUG_SUBSYSTEM S_LOV
46 #include <libcfs/libcfs.h>
48 #include <liblustre.h>
52 #include "lov_internal.h"
54 static void lov_pool_getref(struct pool_desc *pool)
56 CDEBUG(D_INFO, "pool %p\n", pool);
57 cfs_atomic_inc(&pool->pool_refcount);
60 void lov_pool_putref(struct pool_desc *pool)
62 CDEBUG(D_INFO, "pool %p\n", pool);
63 if (cfs_atomic_dec_and_test(&pool->pool_refcount)) {
64 LASSERT(cfs_hlist_unhashed(&pool->pool_hash));
65 LASSERT(cfs_list_empty(&pool->pool_list));
66 LASSERT(pool->pool_proc_entry == NULL);
67 lov_ost_pool_free(&(pool->pool_rr.lqr_pool));
68 lov_ost_pool_free(&(pool->pool_obds));
74 void lov_pool_putref_locked(struct pool_desc *pool)
76 CDEBUG(D_INFO, "pool %p\n", pool);
77 LASSERT(cfs_atomic_read(&pool->pool_refcount) > 1);
79 cfs_atomic_dec(&pool->pool_refcount);
83 * hash function using a Rotating Hash algorithm
84 * Knuth, D. The Art of Computer Programming,
85 * Volume 3: Sorting and Searching,
87 * Addison Wesley, 1973
89 static __u32 pool_hashfn(cfs_hash_t *hash_body, const void *key, unsigned mask)
96 poolname = (char *)key;
97 for (i = 0; i < LOV_MAXPOOLNAME; i++) {
98 if (poolname[i] == '\0')
100 result = (result << 4)^(result >> 28) ^ poolname[i];
102 return (result % mask);
105 static void *pool_key(cfs_hlist_node_t *hnode)
107 struct pool_desc *pool;
109 pool = cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
110 return (pool->pool_name);
113 static int pool_hashkey_keycmp(const void *key, cfs_hlist_node_t *compared_hnode)
116 struct pool_desc *pool;
118 pool_name = (char *)key;
119 pool = cfs_hlist_entry(compared_hnode, struct pool_desc, pool_hash);
120 return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
123 static void *pool_hashobject(cfs_hlist_node_t *hnode)
125 return cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
128 static void pool_hashrefcount_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
130 struct pool_desc *pool;
132 pool = cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
133 lov_pool_getref(pool);
136 static void pool_hashrefcount_put_locked(cfs_hash_t *hs,
137 cfs_hlist_node_t *hnode)
139 struct pool_desc *pool;
141 pool = cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
142 lov_pool_putref_locked(pool);
145 cfs_hash_ops_t pool_hash_operations = {
146 .hs_hash = pool_hashfn,
148 .hs_keycmp = pool_hashkey_keycmp,
149 .hs_object = pool_hashobject,
150 .hs_get = pool_hashrefcount_get,
151 .hs_put_locked = pool_hashrefcount_put_locked,
156 /* ifdef needed for liblustre support */
158 * pool /proc seq_file methods
161 * iterator is used to go through the target pool entries
162 * index is the current entry index in the lp_array[] array
163 * index >= pos returned to the seq_file interface
164 * pos is from 0 to (pool->pool_obds.op_count - 1)
166 #define POOL_IT_MAGIC 0xB001CEA0
167 struct pool_iterator {
169 struct pool_desc *pool;
170 int idx; /* from 0 to pool_tgt_size - 1 */
173 static void *pool_proc_next(struct seq_file *s, void *v, loff_t *pos)
175 struct pool_iterator *iter = (struct pool_iterator *)s->private;
178 LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
180 /* test if end of file */
181 if (*pos >= pool_tgt_count(iter->pool))
184 /* iterate to find a non empty entry */
185 prev_idx = iter->idx;
186 cfs_down_read(&pool_tgt_rw_sem(iter->pool));
188 if (iter->idx == pool_tgt_count(iter->pool)) {
189 iter->idx = prev_idx; /* we stay on the last entry */
190 cfs_up_read(&pool_tgt_rw_sem(iter->pool));
193 cfs_up_read(&pool_tgt_rw_sem(iter->pool));
195 /* return != NULL to continue */
199 static void *pool_proc_start(struct seq_file *s, loff_t *pos)
201 struct pool_desc *pool = (struct pool_desc *)s->private;
202 struct pool_iterator *iter;
204 lov_pool_getref(pool);
205 if ((pool_tgt_count(pool) == 0) ||
206 (*pos >= pool_tgt_count(pool))) {
207 /* iter is not created, so stop() has no way to
208 * find pool to dec ref */
209 lov_pool_putref(pool);
215 return ERR_PTR(-ENOMEM);
216 iter->magic = POOL_IT_MAGIC;
220 /* we use seq_file private field to memorized iterator so
221 * we can free it at stop() */
222 /* /!\ do not forget to restore it to pool before freeing it */
230 ptr = pool_proc_next(s, &iter, &i);
231 } while ((i < *pos) && (ptr != NULL));
237 static void pool_proc_stop(struct seq_file *s, void *v)
239 struct pool_iterator *iter = (struct pool_iterator *)s->private;
241 /* in some cases stop() method is called 2 times, without
242 * calling start() method (see seq_read() from fs/seq_file.c)
243 * we have to free only if s->private is an iterator */
244 if ((iter) && (iter->magic == POOL_IT_MAGIC)) {
245 /* we restore s->private so next call to pool_proc_start()
247 s->private = iter->pool;
248 lov_pool_putref(iter->pool);
254 static int pool_proc_show(struct seq_file *s, void *v)
256 struct pool_iterator *iter = (struct pool_iterator *)v;
257 struct lov_tgt_desc *tgt;
259 LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
260 LASSERT(iter->pool != NULL);
261 LASSERT(iter->idx <= pool_tgt_count(iter->pool));
263 cfs_down_read(&pool_tgt_rw_sem(iter->pool));
264 tgt = pool_tgt(iter->pool, iter->idx);
265 cfs_up_read(&pool_tgt_rw_sem(iter->pool));
267 seq_printf(s, "%s\n", obd_uuid2str(&(tgt->ltd_uuid)));
272 static struct seq_operations pool_proc_ops = {
273 .start = pool_proc_start,
274 .next = pool_proc_next,
275 .stop = pool_proc_stop,
276 .show = pool_proc_show,
279 static int pool_proc_open(struct inode *inode, struct file *file)
283 rc = seq_open(file, &pool_proc_ops);
285 struct seq_file *s = file->private_data;
286 s->private = PROC_I(inode)->pde->data;
291 static struct file_operations pool_proc_operations = {
292 .open = pool_proc_open,
295 .release = seq_release,
299 void lov_dump_pool(int level, struct pool_desc *pool)
303 lov_pool_getref(pool);
305 CDEBUG(level, "pool "LOV_POOLNAMEF" has %d members\n",
306 pool->pool_name, pool->pool_obds.op_count);
307 cfs_down_read(&pool_tgt_rw_sem(pool));
309 for (i = 0; i < pool_tgt_count(pool) ; i++) {
310 if (!pool_tgt(pool, i) || !(pool_tgt(pool, i))->ltd_exp)
312 CDEBUG(level, "pool "LOV_POOLNAMEF"[%d] = %s\n",
314 obd_uuid2str(&((pool_tgt(pool, i))->ltd_uuid)));
317 cfs_up_read(&pool_tgt_rw_sem(pool));
318 lov_pool_putref(pool);
321 #define LOV_POOL_INIT_COUNT 2
322 int lov_ost_pool_init(struct ost_pool *op, unsigned int count)
327 count = LOV_POOL_INIT_COUNT;
330 cfs_init_rwsem(&op->op_rw_sem);
332 OBD_ALLOC(op->op_array, op->op_size * sizeof(op->op_array[0]));
333 if (op->op_array == NULL) {
341 /* Caller must hold write op_rwlock */
342 int lov_ost_pool_extend(struct ost_pool *op, unsigned int min_count)
347 LASSERT(min_count != 0);
349 if (op->op_count < op->op_size)
352 new_size = max(min_count, 2 * op->op_size);
353 OBD_ALLOC(new, new_size * sizeof(op->op_array[0]));
357 /* copy old array to new one */
358 memcpy(new, op->op_array, op->op_size * sizeof(op->op_array[0]));
359 OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
361 op->op_size = new_size;
365 int lov_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count)
370 cfs_down_write(&op->op_rw_sem);
372 rc = lov_ost_pool_extend(op, min_count);
376 /* search ost in pool array */
377 for (i = 0; i < op->op_count; i++) {
378 if (op->op_array[i] == idx)
379 GOTO(out, rc = -EEXIST);
381 /* ost not found we add it */
382 op->op_array[op->op_count] = idx;
386 cfs_up_write(&op->op_rw_sem);
390 int lov_ost_pool_remove(struct ost_pool *op, __u32 idx)
395 cfs_down_write(&op->op_rw_sem);
397 for (i = 0; i < op->op_count; i++) {
398 if (op->op_array[i] == idx) {
399 memmove(&op->op_array[i], &op->op_array[i + 1],
400 (op->op_count - i - 1) * sizeof(op->op_array[0]));
402 cfs_up_write(&op->op_rw_sem);
408 cfs_up_write(&op->op_rw_sem);
412 int lov_ost_pool_free(struct ost_pool *op)
416 if (op->op_size == 0)
419 cfs_down_write(&op->op_rw_sem);
421 OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
426 cfs_up_write(&op->op_rw_sem);
431 int lov_pool_new(struct obd_device *obd, char *poolname)
434 struct pool_desc *new_pool;
440 if (strlen(poolname) > LOV_MAXPOOLNAME)
441 RETURN(-ENAMETOOLONG);
443 OBD_ALLOC_PTR(new_pool);
444 if (new_pool == NULL)
447 strncpy(new_pool->pool_name, poolname, LOV_MAXPOOLNAME);
448 new_pool->pool_name[LOV_MAXPOOLNAME] = '\0';
449 new_pool->pool_lov = lov;
450 /* ref count init to 1 because when created a pool is always used
453 cfs_atomic_set(&new_pool->pool_refcount, 1);
454 rc = lov_ost_pool_init(&new_pool->pool_obds, 0);
458 memset(&(new_pool->pool_rr), 0, sizeof(struct lov_qos_rr));
459 rc = lov_ost_pool_init(&new_pool->pool_rr.lqr_pool, 0);
461 GOTO(out_free_pool_obds, rc);
463 CFS_INIT_HLIST_NODE(&new_pool->pool_hash);
466 /* we need this assert seq_file is not implementated for liblustre */
467 /* get ref for /proc file */
468 lov_pool_getref(new_pool);
469 new_pool->pool_proc_entry = lprocfs_add_simple(lov->lov_pool_proc_entry,
470 poolname, NULL, NULL,
472 &pool_proc_operations);
473 if (IS_ERR(new_pool->pool_proc_entry)) {
474 CWARN("Cannot add proc pool entry "LOV_POOLNAMEF"\n", poolname);
475 new_pool->pool_proc_entry = NULL;
476 lov_pool_putref(new_pool);
478 CDEBUG(D_INFO, "pool %p - proc %p\n", new_pool, new_pool->pool_proc_entry);
481 cfs_spin_lock(&obd->obd_dev_lock);
482 cfs_list_add_tail(&new_pool->pool_list, &lov->lov_pool_list);
483 lov->lov_pool_count++;
484 cfs_spin_unlock(&obd->obd_dev_lock);
486 /* add to find only when it fully ready */
487 rc = cfs_hash_add_unique(lov->lov_pools_hash_body, poolname,
488 &new_pool->pool_hash);
490 GOTO(out_err, rc = -EEXIST);
492 CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
493 poolname, lov->lov_pool_count);
498 cfs_spin_lock(&obd->obd_dev_lock);
499 cfs_list_del_init(&new_pool->pool_list);
500 lov->lov_pool_count--;
501 cfs_spin_unlock(&obd->obd_dev_lock);
503 lprocfs_remove(&new_pool->pool_proc_entry);
505 lov_ost_pool_free(&new_pool->pool_rr.lqr_pool);
507 lov_ost_pool_free(&new_pool->pool_obds);
508 OBD_FREE_PTR(new_pool);
512 int lov_pool_del(struct obd_device *obd, char *poolname)
515 struct pool_desc *pool;
520 /* lookup and kill hash reference */
521 pool = cfs_hash_del_key(lov->lov_pools_hash_body, poolname);
525 if (pool->pool_proc_entry != NULL) {
526 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_proc_entry);
527 lprocfs_remove(&pool->pool_proc_entry);
528 lov_pool_putref(pool);
531 cfs_spin_lock(&obd->obd_dev_lock);
532 cfs_list_del_init(&pool->pool_list);
533 lov->lov_pool_count--;
534 cfs_spin_unlock(&obd->obd_dev_lock);
536 /* release last reference */
537 lov_pool_putref(pool);
543 int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
545 struct obd_uuid ost_uuid;
547 struct pool_desc *pool;
548 unsigned int lov_idx;
554 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
558 obd_str2uuid(&ost_uuid, ostname);
561 /* search ost in lov array */
563 for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
564 if (!lov->lov_tgts[lov_idx])
566 if (obd_uuid_equals(&ost_uuid,
567 &(lov->lov_tgts[lov_idx]->ltd_uuid)))
570 /* test if ost found in lov */
571 if (lov_idx == lov->desc.ld_tgt_count)
572 GOTO(out, rc = -EINVAL);
574 rc = lov_ost_pool_add(&pool->pool_obds, lov_idx, lov->lov_tgt_size);
578 pool->pool_rr.lqr_dirty = 1;
580 CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
581 ostname, poolname, pool_tgt_count(pool));
586 lov_pool_putref(pool);
590 int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
592 struct obd_uuid ost_uuid;
594 struct pool_desc *pool;
595 unsigned int lov_idx;
601 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
605 obd_str2uuid(&ost_uuid, ostname);
608 /* search ost in lov array, to get index */
609 for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
610 if (!lov->lov_tgts[lov_idx])
613 if (obd_uuid_equals(&ost_uuid,
614 &(lov->lov_tgts[lov_idx]->ltd_uuid)))
618 /* test if ost found in lov */
619 if (lov_idx == lov->desc.ld_tgt_count)
620 GOTO(out, rc = -EINVAL);
622 lov_ost_pool_remove(&pool->pool_obds, lov_idx);
624 pool->pool_rr.lqr_dirty = 1;
626 CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
632 lov_pool_putref(pool);
636 int lov_check_index_in_pool(__u32 idx, struct pool_desc *pool)
641 /* caller may no have a ref on pool if it got the pool
642 * without calling lov_find_pool() (e.g. go through the lov pool
645 lov_pool_getref(pool);
647 cfs_down_read(&pool_tgt_rw_sem(pool));
649 for (i = 0; i < pool_tgt_count(pool); i++) {
650 if (pool_tgt_array(pool)[i] == idx)
656 cfs_up_read(&pool_tgt_rw_sem(pool));
658 lov_pool_putref(pool);
662 struct pool_desc *lov_find_pool(struct lov_obd *lov, char *poolname)
664 struct pool_desc *pool;
667 if (poolname[0] != '\0') {
668 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
670 CWARN("Request for an unknown pool ("LOV_POOLNAMEF")\n",
672 if ((pool != NULL) && (pool_tgt_count(pool) == 0)) {
673 CWARN("Request for an empty pool ("LOV_POOLNAMEF")\n",
675 /* pool is ignored, so we remove ref on it */
676 lov_pool_putref(pool);