Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / staging / lustre / lustre / lov / lov_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 (c) 2008, 2010, Oracle and/or its affiliates. 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/lov/lov_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 "../../include/linux/libcfs/libcfs.h"
48
49 #include "../include/obd.h"
50 #include "lov_internal.h"
51
52 #define pool_tgt(_p, _i) \
53                 _p->pool_lobd->u.lov.lov_tgts[_p->pool_obds.op_array[_i]]
54
55 static void lov_pool_getref(struct pool_desc *pool)
56 {
57         CDEBUG(D_INFO, "pool %p\n", pool);
58         atomic_inc(&pool->pool_refcount);
59 }
60
61 void lov_pool_putref(struct pool_desc *pool)
62 {
63         CDEBUG(D_INFO, "pool %p\n", pool);
64         if (atomic_dec_and_test(&pool->pool_refcount)) {
65                 LASSERT(hlist_unhashed(&pool->pool_hash));
66                 LASSERT(list_empty(&pool->pool_list));
67                 LASSERT(pool->pool_proc_entry == NULL);
68                 lov_ost_pool_free(&(pool->pool_rr.lqr_pool));
69                 lov_ost_pool_free(&(pool->pool_obds));
70                 OBD_FREE_PTR(pool);
71         }
72 }
73
74 static void lov_pool_putref_locked(struct pool_desc *pool)
75 {
76         CDEBUG(D_INFO, "pool %p\n", pool);
77         LASSERT(atomic_read(&pool->pool_refcount) > 1);
78
79         atomic_dec(&pool->pool_refcount);
80 }
81
82 /*
83  * hash function using a Rotating Hash algorithm
84  * Knuth, D. The Art of Computer Programming,
85  * Volume 3: Sorting and Searching,
86  * Chapter 6.4.
87  * Addison Wesley, 1973
88  */
89 static __u32 pool_hashfn(struct cfs_hash *hash_body, const void *key, unsigned mask)
90 {
91         int i;
92         __u32 result;
93         char *poolname;
94
95         result = 0;
96         poolname = (char *)key;
97         for (i = 0; i < LOV_MAXPOOLNAME; i++) {
98                 if (poolname[i] == '\0')
99                         break;
100                 result = (result << 4)^(result >> 28) ^  poolname[i];
101         }
102         return (result % mask);
103 }
104
105 static void *pool_key(struct hlist_node *hnode)
106 {
107         struct pool_desc *pool;
108
109         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
110         return pool->pool_name;
111 }
112
113 static int pool_hashkey_keycmp(const void *key, struct hlist_node *compared_hnode)
114 {
115         char *pool_name;
116         struct pool_desc *pool;
117
118         pool_name = (char *)key;
119         pool = hlist_entry(compared_hnode, struct pool_desc, pool_hash);
120         return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
121 }
122
123 static void *pool_hashobject(struct hlist_node *hnode)
124 {
125         return hlist_entry(hnode, struct pool_desc, pool_hash);
126 }
127
128 static void pool_hashrefcount_get(struct cfs_hash *hs, struct hlist_node *hnode)
129 {
130         struct pool_desc *pool;
131
132         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
133         lov_pool_getref(pool);
134 }
135
136 static void pool_hashrefcount_put_locked(struct cfs_hash *hs,
137                                          struct hlist_node *hnode)
138 {
139         struct pool_desc *pool;
140
141         pool = hlist_entry(hnode, struct pool_desc, pool_hash);
142         lov_pool_putref_locked(pool);
143 }
144
145 cfs_hash_ops_t pool_hash_operations = {
146         .hs_hash        = pool_hashfn,
147         .hs_key  = pool_key,
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,
152
153 };
154
155 #if defined (CONFIG_PROC_FS)
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         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);
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                 lov_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 lov_tgt_desc *tgt;
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         tgt = pool_tgt(iter->pool, iter->idx);
265         up_read(&pool_tgt_rw_sem(iter->pool));
266         if (tgt)
267                 seq_printf(s, "%s\n", obd_uuid2str(&(tgt->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 = PDE_DATA(inode);
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 /* CONFIG_PROC_FS */
298
299 void lov_dump_pool(int level, struct pool_desc *pool)
300 {
301         int i;
302
303         lov_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         lov_pool_putref(pool);
319 }
320
321 #define LOV_POOL_INIT_COUNT 2
322 int lov_ost_pool_init(struct ost_pool *op, unsigned int count)
323 {
324         if (count == 0)
325                 count = LOV_POOL_INIT_COUNT;
326         op->op_array = NULL;
327         op->op_count = 0;
328         init_rwsem(&op->op_rw_sem);
329         op->op_size = count;
330         OBD_ALLOC(op->op_array, op->op_size * sizeof(op->op_array[0]));
331         if (op->op_array == NULL) {
332                 op->op_size = 0;
333                 return -ENOMEM;
334         }
335         return 0;
336 }
337
338 /* Caller must hold write op_rwlock */
339 int lov_ost_pool_extend(struct ost_pool *op, unsigned int min_count)
340 {
341         __u32 *new;
342         int new_size;
343
344         LASSERT(min_count != 0);
345
346         if (op->op_count < op->op_size)
347                 return 0;
348
349         new_size = max(min_count, 2 * op->op_size);
350         OBD_ALLOC(new, new_size * sizeof(op->op_array[0]));
351         if (new == NULL)
352                 return -ENOMEM;
353
354         /* copy old array to new one */
355         memcpy(new, op->op_array, op->op_size * sizeof(op->op_array[0]));
356         OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
357         op->op_array = new;
358         op->op_size = new_size;
359         return 0;
360 }
361
362 int lov_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count)
363 {
364         int rc = 0, i;
365
366         down_write(&op->op_rw_sem);
367
368         rc = lov_ost_pool_extend(op, min_count);
369         if (rc)
370                 goto out;
371
372         /* search ost in pool array */
373         for (i = 0; i < op->op_count; i++) {
374                 if (op->op_array[i] == idx) {
375                         rc = -EEXIST;
376                         goto out;
377                 }
378         }
379         /* ost not found we add it */
380         op->op_array[op->op_count] = idx;
381         op->op_count++;
382 out:
383         up_write(&op->op_rw_sem);
384         return rc;
385 }
386
387 int lov_ost_pool_remove(struct ost_pool *op, __u32 idx)
388 {
389         int i;
390
391         down_write(&op->op_rw_sem);
392
393         for (i = 0; i < op->op_count; i++) {
394                 if (op->op_array[i] == idx) {
395                         memmove(&op->op_array[i], &op->op_array[i + 1],
396                                 (op->op_count - i - 1) * sizeof(op->op_array[0]));
397                         op->op_count--;
398                         up_write(&op->op_rw_sem);
399                         return 0;
400                 }
401         }
402
403         up_write(&op->op_rw_sem);
404         return -EINVAL;
405 }
406
407 int lov_ost_pool_free(struct ost_pool *op)
408 {
409         if (op->op_size == 0)
410                 return 0;
411
412         down_write(&op->op_rw_sem);
413
414         OBD_FREE(op->op_array, op->op_size * sizeof(op->op_array[0]));
415         op->op_array = NULL;
416         op->op_count = 0;
417         op->op_size = 0;
418
419         up_write(&op->op_rw_sem);
420         return 0;
421 }
422
423
424 int lov_pool_new(struct obd_device *obd, char *poolname)
425 {
426         struct lov_obd *lov;
427         struct pool_desc *new_pool;
428         int rc;
429
430         lov = &(obd->u.lov);
431
432         if (strlen(poolname) > LOV_MAXPOOLNAME)
433                 return -ENAMETOOLONG;
434
435         OBD_ALLOC_PTR(new_pool);
436         if (new_pool == NULL)
437                 return -ENOMEM;
438
439         strncpy(new_pool->pool_name, poolname, LOV_MAXPOOLNAME);
440         new_pool->pool_name[LOV_MAXPOOLNAME] = '\0';
441         new_pool->pool_lobd = obd;
442         /* ref count init to 1 because when created a pool is always used
443          * up to deletion
444          */
445         atomic_set(&new_pool->pool_refcount, 1);
446         rc = lov_ost_pool_init(&new_pool->pool_obds, 0);
447         if (rc)
448                 goto out_err;
449
450         memset(&(new_pool->pool_rr), 0, sizeof(struct lov_qos_rr));
451         rc = lov_ost_pool_init(&new_pool->pool_rr.lqr_pool, 0);
452         if (rc)
453                 goto out_free_pool_obds;
454
455         INIT_HLIST_NODE(&new_pool->pool_hash);
456
457 #if defined (CONFIG_PROC_FS)
458         /* we need this assert seq_file is not implemented for liblustre */
459         /* get ref for /proc file */
460         lov_pool_getref(new_pool);
461         new_pool->pool_proc_entry = lprocfs_add_simple(lov->lov_pool_proc_entry,
462                                                        poolname, new_pool,
463                                                        &pool_proc_operations);
464         if (IS_ERR(new_pool->pool_proc_entry)) {
465                 CWARN("Cannot add proc pool entry "LOV_POOLNAMEF"\n", poolname);
466                 new_pool->pool_proc_entry = NULL;
467                 lov_pool_putref(new_pool);
468         }
469         CDEBUG(D_INFO, "pool %p - proc %p\n", new_pool, new_pool->pool_proc_entry);
470 #endif
471
472         spin_lock(&obd->obd_dev_lock);
473         list_add_tail(&new_pool->pool_list, &lov->lov_pool_list);
474         lov->lov_pool_count++;
475         spin_unlock(&obd->obd_dev_lock);
476
477         /* add to find only when it fully ready  */
478         rc = cfs_hash_add_unique(lov->lov_pools_hash_body, poolname,
479                                  &new_pool->pool_hash);
480         if (rc) {
481                 rc = -EEXIST;
482                 goto out_err;
483         }
484
485         CDEBUG(D_CONFIG, LOV_POOLNAMEF" is pool #%d\n",
486                poolname, lov->lov_pool_count);
487
488         return 0;
489
490 out_err:
491         spin_lock(&obd->obd_dev_lock);
492         list_del_init(&new_pool->pool_list);
493         lov->lov_pool_count--;
494         spin_unlock(&obd->obd_dev_lock);
495
496         lprocfs_remove(&new_pool->pool_proc_entry);
497
498         lov_ost_pool_free(&new_pool->pool_rr.lqr_pool);
499 out_free_pool_obds:
500         lov_ost_pool_free(&new_pool->pool_obds);
501         OBD_FREE_PTR(new_pool);
502         return rc;
503 }
504
505 int lov_pool_del(struct obd_device *obd, char *poolname)
506 {
507         struct lov_obd *lov;
508         struct pool_desc *pool;
509
510         lov = &(obd->u.lov);
511
512         /* lookup and kill hash reference */
513         pool = cfs_hash_del_key(lov->lov_pools_hash_body, poolname);
514         if (pool == NULL)
515                 return -ENOENT;
516
517         if (pool->pool_proc_entry != NULL) {
518                 CDEBUG(D_INFO, "proc entry %p\n", pool->pool_proc_entry);
519                 lprocfs_remove(&pool->pool_proc_entry);
520                 lov_pool_putref(pool);
521         }
522
523         spin_lock(&obd->obd_dev_lock);
524         list_del_init(&pool->pool_list);
525         lov->lov_pool_count--;
526         spin_unlock(&obd->obd_dev_lock);
527
528         /* release last reference */
529         lov_pool_putref(pool);
530
531         return 0;
532 }
533
534
535 int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
536 {
537         struct obd_uuid ost_uuid;
538         struct lov_obd *lov;
539         struct pool_desc *pool;
540         unsigned int lov_idx;
541         int rc;
542
543         lov = &(obd->u.lov);
544
545         pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
546         if (pool == NULL)
547                 return -ENOENT;
548
549         obd_str2uuid(&ost_uuid, ostname);
550
551
552         /* search ost in lov array */
553         obd_getref(obd);
554         for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
555                 if (!lov->lov_tgts[lov_idx])
556                         continue;
557                 if (obd_uuid_equals(&ost_uuid,
558                                     &(lov->lov_tgts[lov_idx]->ltd_uuid)))
559                         break;
560         }
561         /* test if ost found in lov */
562         if (lov_idx == lov->desc.ld_tgt_count) {
563                 rc = -EINVAL;
564                 goto out;
565         }
566
567         rc = lov_ost_pool_add(&pool->pool_obds, lov_idx, lov->lov_tgt_size);
568         if (rc)
569                 goto out;
570
571         pool->pool_rr.lqr_dirty = 1;
572
573         CDEBUG(D_CONFIG, "Added %s to "LOV_POOLNAMEF" as member %d\n",
574                ostname, poolname,  pool_tgt_count(pool));
575
576 out:
577         obd_putref(obd);
578         lov_pool_putref(pool);
579         return rc;
580 }
581
582 int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
583 {
584         struct obd_uuid ost_uuid;
585         struct lov_obd *lov;
586         struct pool_desc *pool;
587         unsigned int lov_idx;
588         int rc = 0;
589
590         lov = &(obd->u.lov);
591
592         pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
593         if (pool == NULL)
594                 return -ENOENT;
595
596         obd_str2uuid(&ost_uuid, ostname);
597
598         obd_getref(obd);
599         /* search ost in lov array, to get index */
600         for (lov_idx = 0; lov_idx < lov->desc.ld_tgt_count; lov_idx++) {
601                 if (!lov->lov_tgts[lov_idx])
602                         continue;
603
604                 if (obd_uuid_equals(&ost_uuid,
605                                     &(lov->lov_tgts[lov_idx]->ltd_uuid)))
606                         break;
607         }
608
609         /* test if ost found in lov */
610         if (lov_idx == lov->desc.ld_tgt_count) {
611                 rc = -EINVAL;
612                 goto out;
613         }
614
615         lov_ost_pool_remove(&pool->pool_obds, lov_idx);
616
617         pool->pool_rr.lqr_dirty = 1;
618
619         CDEBUG(D_CONFIG, "%s removed from "LOV_POOLNAMEF"\n", ostname,
620                poolname);
621
622 out:
623         obd_putref(obd);
624         lov_pool_putref(pool);
625         return rc;
626 }
627
628 int lov_check_index_in_pool(__u32 idx, struct pool_desc *pool)
629 {
630         int i, rc;
631
632         /* caller may no have a ref on pool if it got the pool
633          * without calling lov_find_pool() (e.g. go through the lov pool
634          * list)
635          */
636         lov_pool_getref(pool);
637
638         down_read(&pool_tgt_rw_sem(pool));
639
640         for (i = 0; i < pool_tgt_count(pool); i++) {
641                 if (pool_tgt_array(pool)[i] == idx) {
642                         rc = 0;
643                         goto out;
644                 }
645         }
646         rc = -ENOENT;
647 out:
648         up_read(&pool_tgt_rw_sem(pool));
649
650         lov_pool_putref(pool);
651         return rc;
652 }
653
654 struct pool_desc *lov_find_pool(struct lov_obd *lov, char *poolname)
655 {
656         struct pool_desc *pool;
657
658         pool = NULL;
659         if (poolname[0] != '\0') {
660                 pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
661                 if (pool == NULL)
662                         CWARN("Request for an unknown pool ("LOV_POOLNAMEF")\n",
663                               poolname);
664                 if ((pool != NULL) && (pool_tgt_count(pool) == 0)) {
665                         CWARN("Request for an empty pool ("LOV_POOLNAMEF")\n",
666                                poolname);
667                         /* pool is ignored, so we remove ref on it */
668                         lov_pool_putref(pool);
669                         pool = NULL;
670                 }
671         }
672         return pool;
673 }