Whamcloud - gitweb
LU-657 test: limit the write size in run_dd
[fs/lustre-release.git] / lustre / lod / lod_pool.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9  *
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).
15  *
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
18  * copy of GPLv2].
19  *
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
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2012, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/lod/lod_pool.c
37  *
38  * OST pool methods
39  *
40  * Author: Jacques-Charles LAFOUCRIERE <jc.lafoucriere@cea.fr>
41  * Author: Alex Lyashkov <Alexey.Lyashkov@Sun.COM>
42  * Author: Nathaniel Rutman <Nathan.Rutman@Sun.COM>
43  */
44
45 #define DEBUG_SUBSYSTEM S_LOV
46
47 #include <libcfs/libcfs.h>
48 #include <obd.h>
49 #include "lod_internal.h"
50
51 #define pool_tgt(_p, _i) \
52         OST_TGT(lu2lod_dev((_p)->pool_lobd->obd_lu_dev),(_p)->pool_obds.op_array[_i])
53
54 static void lod_pool_getref(struct pool_desc *pool)
55 {
56         CDEBUG(D_INFO, "pool %p\n", pool);
57         atomic_inc(&pool->pool_refcount);
58 }
59
60 void lod_pool_putref(struct pool_desc *pool)
61 {
62         CDEBUG(D_INFO, "pool %p\n", pool);
63         if (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                 lod_ost_pool_free(&(pool->pool_rr.lqr_pool));
68                 lod_ost_pool_free(&(pool->pool_obds));
69                 OBD_FREE_PTR(pool);
70                 EXIT;
71         }
72 }
73
74 void lod_pool_putref_locked(struct pool_desc *pool)
75 {
76         CDEBUG(D_INFO, "pool %p\n", pool);
77         LASSERT(cfs_atomic_read(&pool->pool_refcount) > 1);
78
79         cfs_atomic_dec(&pool->pool_refcount);
80 }
81
82
83 /*
84  * hash function using a Rotating Hash algorithm
85  * Knuth, D. The Art of Computer Programming,
86  * Volume 3: Sorting and Searching,
87  * Chapter 6.4.
88  * Addison Wesley, 1973
89  */
90 static __u32 pool_hashfn(cfs_hash_t *hash_body, const void *key, unsigned mask)
91 {
92         int i;
93         __u32 result;
94         char *poolname;
95
96         result = 0;
97         poolname = (char *)key;
98         for (i = 0; i < LOV_MAXPOOLNAME; i++) {
99                 if (poolname[i] == '\0')
100                         break;
101                 result = (result << 4)^(result >> 28) ^  poolname[i];
102         }
103         return (result % mask);
104 }
105
106 static void *pool_key(cfs_hlist_node_t *hnode)
107 {
108         struct pool_desc *pool;
109
110         pool = cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
111         return (pool->pool_name);
112 }
113
114 static int pool_hashkey_keycmp(const void *key, cfs_hlist_node_t *compared_hnode)
115 {
116         char *pool_name;
117         struct pool_desc *pool;
118
119         pool_name = (char *)key;
120         pool = cfs_hlist_entry(compared_hnode, struct pool_desc, pool_hash);
121         return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
122 }
123
124 static void *pool_hashobject(cfs_hlist_node_t *hnode)
125 {
126         return cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
127 }
128
129 static void pool_hashrefcount_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
130 {
131         struct pool_desc *pool;
132
133         pool = cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
134         lod_pool_getref(pool);
135 }
136
137 static void pool_hashrefcount_put_locked(cfs_hash_t *hs,
138                 cfs_hlist_node_t *hnode)
139 {
140         struct pool_desc *pool;
141
142         pool = cfs_hlist_entry(hnode, struct pool_desc, pool_hash);
143         lod_pool_putref_locked(pool);
144 }
145
146 cfs_hash_ops_t pool_hash_operations = {
147         .hs_hash        = pool_hashfn,
148         .hs_key         = pool_key,
149         .hs_keycmp      = pool_hashkey_keycmp,
150         .hs_object      = pool_hashobject,
151         .hs_get         = pool_hashrefcount_get,
152         .hs_put_locked  = pool_hashrefcount_put_locked,
153 };
154
155 #ifdef LPROCFS
156 /* ifdef needed for liblustre support */
157 /*
158  * pool /proc seq_file methods
159  */
160 /*
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)
165  */
166 #define POOL_IT_MAGIC 0xB001CEA0
167 struct pool_iterator {
168         int magic;
169         struct pool_desc *pool;
170         int idx;        /* from 0 to pool_tgt_size - 1 */
171 };
172
173 static void *pool_proc_next(struct seq_file *s, void *v, loff_t *pos)
174 {
175         struct pool_iterator *iter = (struct pool_iterator *)s->private;
176         int prev_idx;
177
178         LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
179
180         /* test if end of file */
181         if (*pos >= pool_tgt_count(iter->pool))
182                 return NULL;
183
184         /* iterate to find a non empty entry */
185         prev_idx = iter->idx;
186         down_read(&pool_tgt_rw_sem(iter->pool));
187         iter->idx++;
188         if (iter->idx == pool_tgt_count(iter->pool)) {
189                 iter->idx = prev_idx; /* we stay on the last entry */
190                 up_read(&pool_tgt_rw_sem(iter->pool));
191                 return NULL;
192         }
193         up_read(&pool_tgt_rw_sem(iter->pool));
194         (*pos)++;
195         /* return != NULL to continue */
196         return iter;
197 }
198
199 static void *pool_proc_start(struct seq_file *s, loff_t *pos)
200 {
201         struct pool_desc *pool = (struct pool_desc *)s->private;
202         struct pool_iterator *iter;
203
204         lod_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                 lod_pool_putref(pool);
210                 return NULL;
211         }
212
213         OBD_ALLOC_PTR(iter);
214         if (!iter)
215                 return ERR_PTR(-ENOMEM);
216         iter->magic = POOL_IT_MAGIC;
217         iter->pool = pool;
218         iter->idx = 0;
219
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 */
223         s->private = iter;
224         if (*pos > 0) {
225                 loff_t i;
226                 void *ptr;
227
228                 i = 0;
229                 do {
230                         ptr = pool_proc_next(s, &iter, &i);
231                 } while ((i < *pos) && (ptr != NULL));
232                 return ptr;
233         }
234         return iter;
235 }
236
237 static void pool_proc_stop(struct seq_file *s, void *v)
238 {
239         struct pool_iterator *iter = (struct pool_iterator *)s->private;
240
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()
246                  * will work */
247                 s->private = iter->pool;
248                 lod_pool_putref(iter->pool);
249                 OBD_FREE_PTR(iter);
250         }
251         return;
252 }
253
254 static int pool_proc_show(struct seq_file *s, void *v)
255 {
256         struct pool_iterator *iter = (struct pool_iterator *)v;
257         struct lod_ost_desc  *osc_desc;
258
259         LASSERTF(iter->magic == POOL_IT_MAGIC, "%08X", iter->magic);
260         LASSERT(iter->pool != NULL);
261         LASSERT(iter->idx <= pool_tgt_count(iter->pool));
262
263         down_read(&pool_tgt_rw_sem(iter->pool));
264         osc_desc = pool_tgt(iter->pool, iter->idx);
265         up_read(&pool_tgt_rw_sem(iter->pool));
266         if (osc_desc)
267                 seq_printf(s, "%s\n", obd_uuid2str(&(osc_desc->ltd_uuid)));
268
269         return 0;
270 }
271
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,
277 };
278
279 static int pool_proc_open(struct inode *inode, struct file *file)
280 {
281         int rc;
282
283         rc = seq_open(file, &pool_proc_ops);
284         if (!rc) {
285                 struct seq_file *s = file->private_data;
286                 s->private = PROC_I(inode)->pde->data;
287         }
288         return rc;
289 }
290
291 static struct file_operations pool_proc_operations = {
292         .open           = pool_proc_open,
293         .read           = seq_read,
294         .llseek         = seq_lseek,
295         .release        = seq_release,
296 };
297 #endif /* LPROCFS */
298
299 void lod_dump_pool(int level, struct pool_desc *pool)
300 {
301         int i;
302
303         lod_pool_getref(pool);
304
305         CDEBUG(level, "pool "LOV_POOLNAMEF" has %d members\n",
306                pool->pool_name, pool->pool_obds.op_count);
307         down_read(&pool_tgt_rw_sem(pool));
308
309         for (i = 0; i < pool_tgt_count(pool) ; i++) {
310                 if (!pool_tgt(pool, i) || !(pool_tgt(pool, i))->ltd_exp)
311                         continue;
312                 CDEBUG(level, "pool "LOV_POOLNAMEF"[%d] = %s\n",
313                        pool->pool_name, i,
314                        obd_uuid2str(&((pool_tgt(pool, i))->ltd_uuid)));
315         }
316
317         up_read(&pool_tgt_rw_sem(pool));
318         lod_pool_putref(pool);
319 }
320
321 #define LOD_POOL_INIT_COUNT 2
322 int lod_ost_pool_init(struct ost_pool *op, unsigned int count)
323 {
324         ENTRY;
325
326         if (count == 0)
327                 count = LOD_POOL_INIT_COUNT;
328         op->op_array = NULL;
329         op->op_count = 0;
330         init_rwsem(&op->op_rw_sem);
331         op->op_size = count;
332         OBD_ALLOC(op->op_array, op->op_size * sizeof(op->op_array[0]));
333         if (op->op_array == NULL) {
334                 op->op_size = 0;
335                 RETURN(-ENOMEM);
336         }
337         EXIT;
338         return 0;
339 }
340
341 /* Caller must hold write op_rwlock */
342 int lod_ost_pool_extend(struct ost_pool *op, unsigned int min_count)
343 {
344         __u32 *new;
345         int new_size;
346
347         LASSERT(min_count != 0);
348
349         if (op->op_count < op->op_size)
350                 return 0;
351
352         new_size = max(min_count, 2 * op->op_size);
353         OBD_ALLOC(new, new_size * sizeof(op->op_array[0]));
354         if (new == NULL)
355                 return -ENOMEM;
356
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]));
360         op->op_array = new;
361         op->op_size = new_size;
362         return 0;
363 }
364
365 int lod_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count)
366 {
367         int rc = 0, i;
368         ENTRY;
369
370         down_write(&op->op_rw_sem);
371
372         rc = lod_ost_pool_extend(op, min_count);
373         if (rc)
374                 GOTO(out, rc);
375
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);
380         }
381         /* ost not found we add it */
382         op->op_array[op->op_count] = idx;
383         op->op_count++;
384         EXIT;
385 out:
386         up_write(&op->op_rw_sem);
387         return rc;
388 }
389
390 int lod_ost_pool_remove(struct ost_pool *op, __u32 idx)
391 {
392         int i;
393         ENTRY;
394
395         down_write(&op->op_rw_sem);
396
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]));
401                         op->op_count--;
402                         up_write(&op->op_rw_sem);
403                         EXIT;
404                         return 0;
405                 }
406         }
407
408         up_write(&op->op_rw_sem);
409         RETURN(-EINVAL);
410 }
411
412 int lod_ost_pool_free(struct ost_pool *op)
413 {
414         ENTRY;
415
416         if (op->op_size == 0)
417                 RETURN(0);
418
419         down_write(&op->op_rw_sem);
420
421         OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
422         op->op_array = NULL;
423         op->op_count = 0;
424         op->op_size = 0;
425
426         up_write(&op->op_rw_sem);
427         RETURN(0);
428 }
429
430 int lod_pool_new(struct obd_device *obd, char *poolname)
431 {
432         struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
433         struct pool_desc  *new_pool;
434         int rc;
435         ENTRY;
436
437         if (strlen(poolname) > LOV_MAXPOOLNAME)
438                 RETURN(-ENAMETOOLONG);
439
440         OBD_ALLOC_PTR(new_pool);
441         if (new_pool == NULL)
442                 RETURN(-ENOMEM);
443
444         strncpy(new_pool->pool_name, poolname, LOV_MAXPOOLNAME);
445         new_pool->pool_name[LOV_MAXPOOLNAME] = '\0';
446         new_pool->pool_lobd = obd;
447         /* ref count init to 1 because when created a pool is always used
448          * up to deletion
449          */
450         atomic_set(&new_pool->pool_refcount, 1);
451         rc = lod_ost_pool_init(&new_pool->pool_obds, 0);
452         if (rc)
453                 GOTO(out_err, rc);
454
455         memset(&(new_pool->pool_rr), 0, sizeof(struct lov_qos_rr));
456         rc = lod_ost_pool_init(&new_pool->pool_rr.lqr_pool, 0);
457         if (rc)
458                 GOTO(out_free_pool_obds, rc);
459
460         INIT_HLIST_NODE(&new_pool->pool_hash);
461
462 #ifdef LPROCFS
463         lod_pool_getref(new_pool);
464         new_pool->pool_proc_entry = lprocfs_add_simple(lod->lod_pool_proc_entry,
465                                                        poolname, NULL, NULL,
466                                                        new_pool,
467                                                        &pool_proc_operations);
468         if (IS_ERR(new_pool->pool_proc_entry)) {
469                 CWARN("Cannot add proc pool entry "LOV_POOLNAMEF"\n", poolname);
470                 new_pool->pool_proc_entry = NULL;
471                 lod_pool_putref(new_pool);
472         }
473         CDEBUG(D_INFO, "pool %p - proc %p\n", new_pool, new_pool->pool_proc_entry);
474 #endif
475
476         spin_lock(&obd->obd_dev_lock);
477         cfs_list_add_tail(&new_pool->pool_list, &lod->lod_pool_list);
478         lod->lod_pool_count++;
479         spin_unlock(&obd->obd_dev_lock);
480
481         /* add to find only when it fully ready  */
482         rc = cfs_hash_add_unique(lod->lod_pools_hash_body, poolname,
483                                  &new_pool->pool_hash);
484         if (rc)
485                 GOTO(out_err, rc = -EEXIST);
486
487         CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
488                         poolname, lod->lod_pool_count);
489
490         RETURN(0);
491
492 out_err:
493         spin_lock(&obd->obd_dev_lock);
494         cfs_list_del_init(&new_pool->pool_list);
495         lod->lod_pool_count--;
496         spin_unlock(&obd->obd_dev_lock);
497
498         lprocfs_remove(&new_pool->pool_proc_entry);
499
500         lod_ost_pool_free(&new_pool->pool_rr.lqr_pool);
501 out_free_pool_obds:
502         lod_ost_pool_free(&new_pool->pool_obds);
503         OBD_FREE_PTR(new_pool);
504         return rc;
505 }
506
507 int lod_pool_del(struct obd_device *obd, char *poolname)
508 {
509         struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
510         struct pool_desc  *pool;
511         ENTRY;
512
513         /* lookup and kill hash reference */
514         pool = cfs_hash_del_key(lod->lod_pools_hash_body, poolname);
515         if (pool == NULL)
516                 RETURN(-ENOENT);
517
518         if (pool->pool_proc_entry != NULL) {
519                 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_proc_entry);
520                 lprocfs_remove(&pool->pool_proc_entry);
521                 lod_pool_putref(pool);
522         }
523
524         spin_lock(&obd->obd_dev_lock);
525         cfs_list_del_init(&pool->pool_list);
526         lod->lod_pool_count--;
527         spin_unlock(&obd->obd_dev_lock);
528
529         /* release last reference */
530         lod_pool_putref(pool);
531
532         RETURN(0);
533 }
534
535
536 int lod_pool_add(struct obd_device *obd, char *poolname, char *ostname)
537 {
538         struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
539         struct obd_uuid    ost_uuid;
540         struct pool_desc  *pool;
541         unsigned int       idx;
542         int                rc = -EINVAL;
543         ENTRY;
544
545         pool = cfs_hash_lookup(lod->lod_pools_hash_body, poolname);
546         if (pool == NULL)
547                 RETURN(-ENOENT);
548
549         obd_str2uuid(&ost_uuid, ostname);
550
551         /* search ost in lod array */
552         lod_getref(lod);
553         lod_foreach_ost(lod, idx) {
554                 if (obd_uuid_equals(&ost_uuid, &OST_TGT(lod, idx)->ltd_uuid)) {
555                         rc = 0;
556                         break;
557                 }
558         }
559
560         if (rc)
561                 GOTO(out, rc);
562
563         rc = lod_ost_pool_add(&pool->pool_obds, idx, lod->lod_osts_size);
564         if (rc)
565                 GOTO(out, rc);
566
567         pool->pool_rr.lqr_dirty = 1;
568
569         CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
570                         ostname, poolname,  pool_tgt_count(pool));
571
572         EXIT;
573 out:
574         lod_putref(lod);
575         lod_pool_putref(pool);
576         return rc;
577 }
578
579 int lod_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
580 {
581         struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
582         struct obd_uuid    ost_uuid;
583         struct pool_desc  *pool;
584         unsigned int       idx;
585         int                rc = -EINVAL;
586         ENTRY;
587
588         pool = cfs_hash_lookup(lod->lod_pools_hash_body, poolname);
589         if (pool == NULL)
590                 RETURN(-ENOENT);
591
592         obd_str2uuid(&ost_uuid, ostname);
593
594         lod_getref(lod);
595         /* search ost in lod array, to get index */
596         cfs_foreach_bit(lod->lod_ost_bitmap, idx) {
597                 if (obd_uuid_equals(&ost_uuid, &OST_TGT(lod, idx)->ltd_uuid)) {
598                         rc = 0;
599                         break;
600                 }
601         }
602
603         /* test if ost found in lod array */
604         if (rc)
605                 GOTO(out, rc);
606
607         lod_ost_pool_remove(&pool->pool_obds, idx);
608
609         pool->pool_rr.lqr_dirty = 1;
610
611         CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
612                poolname);
613
614         EXIT;
615 out:
616         lod_putref(lod);
617         lod_pool_putref(pool);
618         return rc;
619 }
620
621 int lod_check_index_in_pool(__u32 idx, struct pool_desc *pool)
622 {
623         int i, rc;
624         ENTRY;
625
626         /* caller may no have a ref on pool if it got the pool
627          * without calling lod_find_pool() (e.g. go through the lod pool
628          * list)
629          */
630         lod_pool_getref(pool);
631
632         down_read(&pool_tgt_rw_sem(pool));
633
634         for (i = 0; i < pool_tgt_count(pool); i++) {
635                 if (pool_tgt_array(pool)[i] == idx)
636                         GOTO(out, rc = 0);
637         }
638         rc = -ENOENT;
639         EXIT;
640 out:
641         up_read(&pool_tgt_rw_sem(pool));
642
643         lod_pool_putref(pool);
644         return rc;
645 }
646
647 struct pool_desc *lod_find_pool(struct lod_device *lod, char *poolname)
648 {
649         struct pool_desc *pool;
650
651         pool = NULL;
652         if (poolname[0] != '\0') {
653                 pool = cfs_hash_lookup(lod->lod_pools_hash_body, poolname);
654                 if (pool == NULL)
655                         CWARN("Request for an unknown pool ("LOV_POOLNAMEF")\n",
656                               poolname);
657                 if ((pool != NULL) && (pool_tgt_count(pool) == 0)) {
658                         CWARN("Request for an empty pool ("LOV_POOLNAMEF")\n",
659                               poolname);
660                         /* pool is ignored, so we remove ref on it */
661                         lod_pool_putref(pool);
662                         pool = NULL;
663                 }
664         }
665         return pool;
666 }
667