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 2008 Sun Microsystems, Inc. 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/lod/lod_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
45 #include <libcfs/libcfs.h>
47 #include "lod_internal.h"
49 #define pool_tgt(_p, _i) \
50 OST_TGT(lu2lod_dev((_p)->pool_lobd->obd_lu_dev),(_p)->pool_obds.op_array[_i])
52 static void lod_pool_getref(struct pool_desc *pool)
54 CDEBUG(D_INFO, "pool %p\n", pool);
55 atomic_inc(&pool->pool_refcount);
58 void lod_pool_putref(struct pool_desc *pool)
60 CDEBUG(D_INFO, "pool %p\n", pool);
61 if (atomic_dec_and_test(&pool->pool_refcount)) {
62 LASSERT(cfs_hlist_unhashed(&pool->pool_hash));
63 LASSERT(cfs_list_empty(&pool->pool_list));
64 LASSERT(pool->pool_proc_entry == NULL);
65 lod_ost_pool_free(&(pool->pool_rr.lqr_pool));
66 lod_ost_pool_free(&(pool->pool_obds));
72 void lod_pool_putref_locked(struct pool_desc *pool)
74 CDEBUG(D_INFO, "pool %p\n", pool);
75 LASSERT(cfs_atomic_read(&pool->pool_refcount) > 1);
77 cfs_atomic_dec(&pool->pool_refcount);
82 * hash function using a Rotating Hash algorithm
83 * Knuth, D. The Art of Computer Programming,
84 * Volume 3: Sorting and Searching,
86 * Addison Wesley, 1973
88 static __u32 pool_hashfn(cfs_hash_t *hash_body, const void *key, unsigned mask)
95 poolname = (char *)key;
96 for (i = 0; i < LOV_MAXPOOLNAME; i++) {
97 if (poolname[i] == '\0')
99 result = (result << 4)^(result >> 28) ^ poolname[i];
101 return (result % mask);
104 static void *pool_key(cfs_hlist_node_t *hnode)
106 struct pool_desc *pool;
108 pool = cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
109 return (pool->pool_name);
112 static int pool_hashkey_keycmp(const void *key, cfs_hlist_node_t *compared_hnode)
115 struct pool_desc *pool;
117 pool_name = (char *)key;
118 pool = cfs_hlist_entry(compared_hnode, struct pool_desc, pool_hash);
119 return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
122 static void *pool_hashobject(cfs_hlist_node_t *hnode)
124 return cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
127 static void pool_hashrefcount_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
129 struct pool_desc *pool;
131 pool = cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
132 lod_pool_getref(pool);
135 static void pool_hashrefcount_put_locked(cfs_hash_t *hs,
136 cfs_hlist_node_t *hnode)
138 struct pool_desc *pool;
140 pool = cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
141 lod_pool_putref_locked(pool);
144 cfs_hash_ops_t pool_hash_operations = {
145 .hs_hash = pool_hashfn,
147 .hs_keycmp = pool_hashkey_keycmp,
148 .hs_object = pool_hashobject,
149 .hs_get = pool_hashrefcount_get,
150 .hs_put_locked = pool_hashrefcount_put_locked,
154 /* ifdef needed for liblustre support */
156 * pool /proc seq_file methods
159 * iterator is used to go through the target pool entries
160 * index is the current entry index in the lp_array[] array
161 * index >= pos returned to the seq_file interface
162 * pos is from 0 to (pool->pool_obds.op_count - 1)
164 #define POOL_IT_MAGIC 0xB001CEA0
165 struct pool_iterator {
167 struct pool_desc *pool;
168 int idx; /* from 0 to pool_tgt_size - 1 */
171 static void *pool_proc_next(struct seq_file *s, void *v, loff_t *pos)
173 struct pool_iterator *iter = (struct pool_iterator *)s->private;
176 LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
178 /* test if end of file */
179 if (*pos >= pool_tgt_count(iter->pool))
182 /* iterate to find a non empty entry */
183 prev_idx = iter->idx;
184 down_read(&pool_tgt_rw_sem(iter->pool));
186 if (iter->idx == pool_tgt_count(iter->pool)) {
187 iter->idx = prev_idx; /* we stay on the last entry */
188 up_read(&pool_tgt_rw_sem(iter->pool));
191 up_read(&pool_tgt_rw_sem(iter->pool));
193 /* return != NULL to continue */
197 static void *pool_proc_start(struct seq_file *s, loff_t *pos)
199 struct pool_desc *pool = (struct pool_desc *)s->private;
200 struct pool_iterator *iter;
202 lod_pool_getref(pool);
203 if ((pool_tgt_count(pool) == 0) ||
204 (*pos >= pool_tgt_count(pool))) {
205 /* iter is not created, so stop() has no way to
206 * find pool to dec ref */
207 lod_pool_putref(pool);
213 return ERR_PTR(-ENOMEM);
214 iter->magic = POOL_IT_MAGIC;
218 /* we use seq_file private field to memorized iterator so
219 * we can free it at stop() */
220 /* /!\ do not forget to restore it to pool before freeing it */
228 ptr = pool_proc_next(s, &iter, &i);
229 } while ((i < *pos) && (ptr != NULL));
235 static void pool_proc_stop(struct seq_file *s, void *v)
237 struct pool_iterator *iter = (struct pool_iterator *)s->private;
239 /* in some cases stop() method is called 2 times, without
240 * calling start() method (see seq_read() from fs/seq_file.c)
241 * we have to free only if s->private is an iterator */
242 if ((iter) && (iter->magic == POOL_IT_MAGIC)) {
243 /* we restore s->private so next call to pool_proc_start()
245 s->private = iter->pool;
246 lod_pool_putref(iter->pool);
252 static int pool_proc_show(struct seq_file *s, void *v)
254 struct pool_iterator *iter = (struct pool_iterator *)v;
255 struct lod_ost_desc *osc_desc;
257 LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
258 LASSERT(iter->pool != NULL);
259 LASSERT(iter->idx <= pool_tgt_count(iter->pool));
261 down_read(&pool_tgt_rw_sem(iter->pool));
262 osc_desc = pool_tgt(iter->pool, iter->idx);
263 up_read(&pool_tgt_rw_sem(iter->pool));
265 seq_printf(s, "%s\n", obd_uuid2str(&(osc_desc->ltd_uuid)));
270 static struct seq_operations pool_proc_ops = {
271 .start = pool_proc_start,
272 .next = pool_proc_next,
273 .stop = pool_proc_stop,
274 .show = pool_proc_show,
277 static int pool_proc_open(struct inode *inode, struct file *file)
281 rc = seq_open(file, &pool_proc_ops);
283 struct seq_file *s = file->private_data;
284 s->private = PROC_I(inode)->pde->data;
289 static struct file_operations pool_proc_operations = {
290 .open = pool_proc_open,
293 .release = seq_release,
297 void lod_dump_pool(int level, struct pool_desc *pool)
301 lod_pool_getref(pool);
303 CDEBUG(level, "pool "LOV_POOLNAMEF" has %d members\n",
304 pool->pool_name, pool->pool_obds.op_count);
305 down_read(&pool_tgt_rw_sem(pool));
307 for (i = 0; i < pool_tgt_count(pool) ; i++) {
308 if (!pool_tgt(pool, i) || !(pool_tgt(pool, i))->ltd_exp)
310 CDEBUG(level, "pool "LOV_POOLNAMEF"[%d] = %s\n",
312 obd_uuid2str(&((pool_tgt(pool, i))->ltd_uuid)));
315 up_read(&pool_tgt_rw_sem(pool));
316 lod_pool_putref(pool);
319 #define LOD_POOL_INIT_COUNT 2
320 int lod_ost_pool_init(struct ost_pool *op, unsigned int count)
325 count = LOD_POOL_INIT_COUNT;
328 init_rwsem(&op->op_rw_sem);
330 OBD_ALLOC(op->op_array, op->op_size * sizeof(op->op_array[0]));
331 if (op->op_array == NULL) {
339 /* Caller must hold write op_rwlock */
340 int lod_ost_pool_extend(struct ost_pool *op, unsigned int min_count)
345 LASSERT(min_count != 0);
347 if (op->op_count < op->op_size)
350 new_size = max(min_count, 2 * op->op_size);
351 OBD_ALLOC(new, new_size * sizeof(op->op_array[0]));
355 /* copy old array to new one */
356 memcpy(new, op->op_array, op->op_size * sizeof(op->op_array[0]));
357 OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
359 op->op_size = new_size;
363 int lod_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count)
368 down_write(&op->op_rw_sem);
370 rc = lod_ost_pool_extend(op, min_count);
374 /* search ost in pool array */
375 for (i = 0; i < op->op_count; i++) {
376 if (op->op_array[i] == idx)
377 GOTO(out, rc = -EEXIST);
379 /* ost not found we add it */
380 op->op_array[op->op_count] = idx;
384 up_write(&op->op_rw_sem);
388 int lod_ost_pool_remove(struct ost_pool *op, __u32 idx)
393 down_write(&op->op_rw_sem);
395 for (i = 0; i < op->op_count; i++) {
396 if (op->op_array[i] == idx) {
397 memmove(&op->op_array[i], &op->op_array[i + 1],
398 (op->op_count - i - 1) * sizeof(op->op_array[0]));
400 up_write(&op->op_rw_sem);
406 up_write(&op->op_rw_sem);
410 int lod_ost_pool_free(struct ost_pool *op)
414 if (op->op_size == 0)
417 down_write(&op->op_rw_sem);
419 OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
424 up_write(&op->op_rw_sem);
428 int lod_pool_new(struct obd_device *obd, char *poolname)
430 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
431 struct pool_desc *new_pool;
435 if (strlen(poolname) > LOV_MAXPOOLNAME)
436 RETURN(-ENAMETOOLONG);
438 OBD_ALLOC_PTR(new_pool);
439 if (new_pool == NULL)
442 strncpy(new_pool->pool_name, poolname, LOV_MAXPOOLNAME);
443 new_pool->pool_name[LOV_MAXPOOLNAME] = '\0';
444 new_pool->pool_lobd = obd;
445 /* ref count init to 1 because when created a pool is always used
448 atomic_set(&new_pool->pool_refcount, 1);
449 rc = lod_ost_pool_init(&new_pool->pool_obds, 0);
453 memset(&(new_pool->pool_rr), 0, sizeof(struct lov_qos_rr));
454 rc = lod_ost_pool_init(&new_pool->pool_rr.lqr_pool, 0);
456 GOTO(out_free_pool_obds, rc);
458 INIT_HLIST_NODE(&new_pool->pool_hash);
461 lod_pool_getref(new_pool);
462 new_pool->pool_proc_entry = lprocfs_add_simple(lod->lod_pool_proc_entry,
463 poolname, NULL, NULL,
465 &pool_proc_operations);
466 if (IS_ERR(new_pool->pool_proc_entry)) {
467 CWARN("Cannot add proc pool entry "LOV_POOLNAMEF"\n", poolname);
468 new_pool->pool_proc_entry = NULL;
469 lod_pool_putref(new_pool);
471 CDEBUG(D_INFO, "pool %p - proc %p\n", new_pool, new_pool->pool_proc_entry);
474 spin_lock(&obd->obd_dev_lock);
475 cfs_list_add_tail(&new_pool->pool_list, &lod->lod_pool_list);
476 lod->lod_pool_count++;
477 spin_unlock(&obd->obd_dev_lock);
479 /* add to find only when it fully ready */
480 rc = cfs_hash_add_unique(lod->lod_pools_hash_body, poolname,
481 &new_pool->pool_hash);
483 GOTO(out_err, rc = -EEXIST);
485 CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
486 poolname, lod->lod_pool_count);
491 spin_lock(&obd->obd_dev_lock);
492 cfs_list_del_init(&new_pool->pool_list);
493 lod->lod_pool_count--;
494 spin_unlock(&obd->obd_dev_lock);
496 lprocfs_remove(&new_pool->pool_proc_entry);
498 lod_ost_pool_free(&new_pool->pool_rr.lqr_pool);
500 lod_ost_pool_free(&new_pool->pool_obds);
501 OBD_FREE_PTR(new_pool);
505 int lod_pool_del(struct obd_device *obd, char *poolname)
507 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
508 struct pool_desc *pool;
511 /* lookup and kill hash reference */
512 pool = cfs_hash_del_key(lod->lod_pools_hash_body, poolname);
516 if (pool->pool_proc_entry != NULL) {
517 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_proc_entry);
518 lprocfs_remove(&pool->pool_proc_entry);
519 lod_pool_putref(pool);
522 spin_lock(&obd->obd_dev_lock);
523 cfs_list_del_init(&pool->pool_list);
524 lod->lod_pool_count--;
525 spin_unlock(&obd->obd_dev_lock);
527 /* release last reference */
528 lod_pool_putref(pool);
534 int lod_pool_add(struct obd_device *obd, char *poolname, char *ostname)
536 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
537 struct obd_uuid ost_uuid;
538 struct pool_desc *pool;
543 pool = cfs_hash_lookup(lod->lod_pools_hash_body, poolname);
547 obd_str2uuid(&ost_uuid, ostname);
549 /* search ost in lod array */
551 lod_foreach_ost(lod, idx) {
552 if (obd_uuid_equals(&ost_uuid, &OST_TGT(lod, idx)->ltd_uuid)) {
561 rc = lod_ost_pool_add(&pool->pool_obds, idx, lod->lod_osts_size);
565 pool->pool_rr.lqr_dirty = 1;
567 CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
568 ostname, poolname, pool_tgt_count(pool));
573 lod_pool_putref(pool);
577 int lod_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
579 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
580 struct obd_uuid ost_uuid;
581 struct pool_desc *pool;
586 pool = cfs_hash_lookup(lod->lod_pools_hash_body, poolname);
590 obd_str2uuid(&ost_uuid, ostname);
593 /* search ost in lod array, to get index */
594 cfs_foreach_bit(lod->lod_ost_bitmap, idx) {
595 if (obd_uuid_equals(&ost_uuid, &OST_TGT(lod, idx)->ltd_uuid)) {
601 /* test if ost found in lod array */
605 lod_ost_pool_remove(&pool->pool_obds, idx);
607 pool->pool_rr.lqr_dirty = 1;
609 CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
615 lod_pool_putref(pool);
619 int lod_check_index_in_pool(__u32 idx, struct pool_desc *pool)
624 /* caller may no have a ref on pool if it got the pool
625 * without calling lod_find_pool() (e.g. go through the lod pool
628 lod_pool_getref(pool);
630 down_read(&pool_tgt_rw_sem(pool));
632 for (i = 0; i < pool_tgt_count(pool); i++) {
633 if (pool_tgt_array(pool)[i] == idx)
639 up_read(&pool_tgt_rw_sem(pool));
641 lod_pool_putref(pool);
645 struct pool_desc *lod_find_pool(struct lod_device *lod, char *poolname)
647 struct pool_desc *pool;
650 if (poolname[0] != '\0') {
651 pool = cfs_hash_lookup(lod->lod_pools_hash_body, poolname);
653 CWARN("Request for an unknown pool ("LOV_POOLNAMEF")\n",
655 if ((pool != NULL) && (pool_tgt_count(pool) == 0)) {
656 CWARN("Request for an empty pool ("LOV_POOLNAMEF")\n",
658 /* pool is ignored, so we remove ref on it */
659 lod_pool_putref(pool);