Whamcloud - gitweb
b=22078 Bind ost/mdt to specific nids
[fs/lustre-release.git] / lustre / obdclass / obd_config.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * GPL HEADER START
5  *
6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 only,
10  * as published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope that it will be useful, but
13  * WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * General Public License version 2 for more details (a copy is included
16  * in the LICENSE file that accompanied this code).
17  *
18  * You should have received a copy of the GNU General Public License
19  * version 2 along with this program; If not, see
20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
21  *
22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23  * CA 95054 USA or visit www.sun.com if you need additional information or
24  * have any questions.
25  *
26  * GPL HEADER END
27  */
28 /*
29  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
30  * Use is subject to license terms.
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/obdclass/obd_config.c
37  *
38  * Config API
39  */
40
41 #define DEBUG_SUBSYSTEM S_CLASS
42 #ifdef __KERNEL__
43 #include <obd_class.h>
44 #include <linux/string.h>
45 #else
46 #include <liblustre.h>
47 #include <obd_class.h>
48 #include <obd.h>
49 #endif
50 #include <lustre_log.h>
51 #include <lprocfs_status.h>
52 #include <libcfs/list.h>
53 #include <lustre_param.h>
54
55 static cfs_hash_ops_t uuid_hash_ops;
56 static cfs_hash_ops_t nid_hash_ops;
57 static cfs_hash_ops_t nid_stat_hash_ops;
58
59 /*********** string parsing utils *********/
60
61 /* returns 0 if we find this key in the buffer, else 1 */
62 int class_find_param(char *buf, char *key, char **valp)
63 {
64         char *ptr;
65
66         if (!buf)
67                 return 1;
68
69         if ((ptr = strstr(buf, key)) == NULL)
70                 return 1;
71
72         if (valp)
73                 *valp = ptr + strlen(key);
74
75         return 0;
76 }
77
78 /**
79  * Finds a parameter in \a params and copies it to \a copy.
80  *
81  * Leading spaces are skipped. Next space or end of string is the
82  * parameter terminator with the exception that spaces inside single or double
83  * quotes get included into a parameter. The parameter is copied into \a copy
84  * which has to be allocated big enough by a caller, quotes are stripped in
85  * the copy and the copy is terminated by 0.
86  *
87  * On return \a params is set to next parameter or to NULL if last
88  * parameter is returned.
89  *
90  * \retval 0 if parameter is returned in \a copy
91  * \retval 1 otherwise
92  * \retval -EINVAL if unbalanced quota is found
93  */
94 int class_get_next_param(char **params, char *copy)
95 {
96         char *q1, *q2, *str;
97         int len;
98
99         str = *params;
100         while (*str == ' ')
101                 str++;
102
103         if (*str == '\0') {
104                 *params = NULL;
105                 return 1;
106         }
107
108         while (1) {
109                 q1 = strpbrk(str, " '\"");
110                 if (q1 == NULL) {
111                         len = strlen(str);
112                         memcpy(copy, str, len);
113                         copy[len] = '\0';
114                         *params = NULL;
115                         return 0;
116                 }
117                 len = q1 - str;
118                 if (*q1 == ' ') {
119                         memcpy(copy, str, len);
120                         copy[len] = '\0';
121                         *params = str + len;
122                         return 0;
123                 }
124
125                 memcpy(copy, str, len);
126                 copy += len;
127
128                 /* search for the matching closing quote */
129                 str = q1 + 1;
130                 q2 = strchr(str, *q1);
131                 if (q2 == NULL) {
132                         CERROR("Unbalanced quota in parameters: \"%s\"\n",
133                                *params);
134                         return -EINVAL;
135                 }
136                 len = q2 - str;
137                 memcpy(copy, str, len);
138                 copy += len;
139                 str = q2 + 1;
140         }
141         return 1;
142 }
143
144 /* returns 0 if this is the first key in the buffer, else 1.
145    valp points to first char after key. */
146 int class_match_param(char *buf, char *key, char **valp)
147 {
148         if (!buf)
149                 return 1;
150
151         if (memcmp(buf, key, strlen(key)) != 0)
152                 return 1;
153
154         if (valp)
155                 *valp = buf + strlen(key);
156
157         return 0;
158 }
159
160 static int parse_nid(char *buf, void *value)
161 {
162         lnet_nid_t *nid = (lnet_nid_t *)value;
163
164         *nid = libcfs_str2nid(buf);
165         if (*nid != LNET_NID_ANY)
166                 return 0;
167
168         LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", buf);
169         return -EINVAL;
170 }
171
172 static int parse_net(char *buf, void *value)
173 {
174         __u32 *net = (__u32 *)net;
175
176         *net = libcfs_str2net(buf);
177         CDEBUG(D_INFO, "Net %s\n", libcfs_net2str(*net));
178         return 0;
179 }
180
181 enum {
182         CLASS_PARSE_NID = 1,
183         CLASS_PARSE_NET,
184 };
185
186 /* 0 is good nid,
187    1 not found
188    < 0 error
189    endh is set to next separator */
190 static int class_parse_value(char *buf, int opc, void *value, char **endh)
191 {
192         char *endp;
193         char  tmp;
194         int   rc = 0;
195
196         if (!buf)
197                 return 1;
198         while (*buf == ',' || *buf == ':')
199                 buf++;
200         if (*buf == ' ' || *buf == '/' || *buf == '\0')
201                 return 1;
202
203         /* nid separators or end of nids */
204         endp = strpbrk(buf, ",: /");
205         if (endp == NULL)
206                 endp = buf + strlen(buf);
207
208         tmp = *endp;
209         *endp = '\0';
210         switch (opc) {
211         default:
212                 LBUG();
213         case CLASS_PARSE_NID:
214                 rc = parse_nid(buf, value);
215                 break;
216         case CLASS_PARSE_NET:
217                 rc = parse_net(buf, value);
218                 break;
219         }
220         *endp = tmp;
221         if (rc != 0)
222                 return rc;
223         if (endh)
224                 *endh = endp;
225         return 0;
226 }
227
228 int class_parse_nid(char *buf, lnet_nid_t *nid, char **endh)
229 {
230         return class_parse_value(buf, CLASS_PARSE_NID, (void *)nid, endh);
231 }
232
233 int class_parse_net(char *buf, __u32 *net, char **endh)
234 {
235         return class_parse_value(buf, CLASS_PARSE_NET, (void *)net, endh);
236 }
237
238 int class_match_net(char *buf, lnet_nid_t nid)
239 {
240         __u32 net;
241
242         while (class_find_param(buf, PARAM_NETWORK, &buf) == 0) {
243                 /* please restrict to the nids pertaining to
244                  * the specified networks */
245                 while (class_parse_net(buf, &net, &buf) == 0) {
246                         if (LNET_NIDNET(nid) == net)
247                                 return 1;
248                 }
249         }
250         return 0;
251 }
252
253 EXPORT_SYMBOL(class_find_param);
254 EXPORT_SYMBOL(class_get_next_param);
255 EXPORT_SYMBOL(class_match_param);
256 EXPORT_SYMBOL(class_parse_nid);
257 EXPORT_SYMBOL(class_parse_net);
258 EXPORT_SYMBOL(class_match_net);
259
260 /********************** class fns **********************/
261
262 /**
263  * Create a new obd device and set the type, name and uuid.  If successful,
264  * the new device can be accessed by either name or uuid.
265  */
266 int class_attach(struct lustre_cfg *lcfg)
267 {
268         struct obd_device *obd = NULL;
269         char *typename, *name, *uuid;
270         int rc, len;
271         ENTRY;
272
273         if (!LUSTRE_CFG_BUFLEN(lcfg, 1)) {
274                 CERROR("No type passed!\n");
275                 RETURN(-EINVAL);
276         }
277         typename = lustre_cfg_string(lcfg, 1);
278
279         if (!LUSTRE_CFG_BUFLEN(lcfg, 0)) {
280                 CERROR("No name passed!\n");
281                 RETURN(-EINVAL);
282         }
283         name = lustre_cfg_string(lcfg, 0);
284
285         if (!LUSTRE_CFG_BUFLEN(lcfg, 2)) {
286                 CERROR("No UUID passed!\n");
287                 RETURN(-EINVAL);
288         }
289         uuid = lustre_cfg_string(lcfg, 2);
290
291         CDEBUG(D_IOCTL, "attach type %s name: %s uuid: %s\n",
292                MKSTR(typename), MKSTR(name), MKSTR(uuid));
293
294         obd = class_newdev(typename, name);
295         if (IS_ERR(obd)) {
296                 /* Already exists or out of obds */
297                 rc = PTR_ERR(obd);
298                 obd = NULL;
299                 CERROR("Cannot create device %s of type %s : %d\n",
300                        name, typename, rc);
301                 GOTO(out, rc);
302         }
303         LASSERTF(obd != NULL, "Cannot get obd device %s of type %s\n",
304                  name, typename);
305         LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC,
306                  "obd %p obd_magic %08X != %08X\n",
307                  obd, obd->obd_magic, OBD_DEVICE_MAGIC);
308         LASSERTF(strncmp(obd->obd_name, name, strlen(name)) == 0,
309                  "%p obd_name %s != %s\n", obd, obd->obd_name, name);
310
311         cfs_rwlock_init(&obd->obd_pool_lock);
312         obd->obd_pool_limit = 0;
313         obd->obd_pool_slv = 0;
314
315         CFS_INIT_LIST_HEAD(&obd->obd_exports);
316         CFS_INIT_LIST_HEAD(&obd->obd_unlinked_exports);
317         CFS_INIT_LIST_HEAD(&obd->obd_delayed_exports);
318         CFS_INIT_LIST_HEAD(&obd->obd_exports_timed);
319         CFS_INIT_LIST_HEAD(&obd->obd_nid_stats);
320         cfs_spin_lock_init(&obd->obd_nid_lock);
321         cfs_spin_lock_init(&obd->obd_dev_lock);
322         cfs_sema_init(&obd->obd_dev_sem, 1);
323         cfs_spin_lock_init(&obd->obd_osfs_lock);
324         /* obd->obd_osfs_age must be set to a value in the distant
325          * past to guarantee a fresh statfs is fetched on mount. */
326         obd->obd_osfs_age = cfs_time_shift_64(-1000);
327
328         /* XXX belongs in setup not attach  */
329         cfs_init_rwsem(&obd->obd_observer_link_sem);
330         /* recovery data */
331         cfs_init_timer(&obd->obd_recovery_timer);
332         cfs_spin_lock_init(&obd->obd_processing_task_lock);
333         cfs_waitq_init(&obd->obd_next_transno_waitq);
334         cfs_waitq_init(&obd->obd_evict_inprogress_waitq);
335         CFS_INIT_LIST_HEAD(&obd->obd_req_replay_queue);
336         CFS_INIT_LIST_HEAD(&obd->obd_lock_replay_queue);
337         CFS_INIT_LIST_HEAD(&obd->obd_final_req_queue);
338         CFS_INIT_LIST_HEAD(&obd->obd_evict_list);
339
340         llog_group_init(&obd->obd_olg, FID_SEQ_LLOG);
341
342         len = strlen(uuid);
343         if (len >= sizeof(obd->obd_uuid)) {
344                 CERROR("uuid must be < %d bytes long\n",
345                        (int)sizeof(obd->obd_uuid));
346                 GOTO(out, rc = -EINVAL);
347         }
348         memcpy(obd->obd_uuid.uuid, uuid, len);
349
350         /* do the attach */
351         if (OBP(obd, attach)) {
352                 rc = OBP(obd,attach)(obd, sizeof *lcfg, lcfg);
353                 if (rc)
354                         GOTO(out, rc = -EINVAL);
355         }
356
357         /* Detach drops this */
358         cfs_spin_lock(&obd->obd_dev_lock);
359         cfs_atomic_set(&obd->obd_refcount, 1);
360         cfs_spin_unlock(&obd->obd_dev_lock);
361         lu_ref_init(&obd->obd_reference);
362         lu_ref_add(&obd->obd_reference, "attach", obd);
363
364         obd->obd_attached = 1;
365         CDEBUG(D_IOCTL, "OBD: dev %d attached type %s with refcount %d\n",
366                obd->obd_minor, typename, cfs_atomic_read(&obd->obd_refcount));
367         RETURN(0);
368  out:
369         if (obd != NULL) {
370                 class_release_dev(obd);
371         }
372         return rc;
373 }
374
375 /** Create hashes, self-export, and call type-specific setup.
376  * Setup is effectively the "start this obd" call.
377  */
378 int class_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
379 {
380         int err = 0;
381         struct obd_export *exp;
382         ENTRY;
383
384         LASSERT(obd != NULL);
385         LASSERTF(obd == class_num2obd(obd->obd_minor),
386                  "obd %p != obd_devs[%d] %p\n",
387                  obd, obd->obd_minor, class_num2obd(obd->obd_minor));
388         LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC,
389                  "obd %p obd_magic %08x != %08x\n",
390                  obd, obd->obd_magic, OBD_DEVICE_MAGIC);
391
392         /* have we attached a type to this device? */
393         if (!obd->obd_attached) {
394                 CERROR("Device %d not attached\n", obd->obd_minor);
395                 RETURN(-ENODEV);
396         }
397
398         if (obd->obd_set_up) {
399                 CERROR("Device %d already setup (type %s)\n",
400                        obd->obd_minor, obd->obd_type->typ_name);
401                 RETURN(-EEXIST);
402         }
403
404         /* is someone else setting us up right now? (attach inits spinlock) */
405         cfs_spin_lock(&obd->obd_dev_lock);
406         if (obd->obd_starting) {
407                 cfs_spin_unlock(&obd->obd_dev_lock);
408                 CERROR("Device %d setup in progress (type %s)\n",
409                        obd->obd_minor, obd->obd_type->typ_name);
410                 RETURN(-EEXIST);
411         }
412         /* just leave this on forever.  I can't use obd_set_up here because
413            other fns check that status, and we're not actually set up yet. */
414         obd->obd_starting = 1;
415         obd->obd_uuid_hash = NULL;
416         obd->obd_nid_hash = NULL;
417         obd->obd_nid_stats_hash = NULL;
418         cfs_spin_unlock(&obd->obd_dev_lock);
419
420         /* create an uuid-export lustre hash */
421         obd->obd_uuid_hash = cfs_hash_create("UUID_HASH",
422                                              HASH_UUID_CUR_BITS,
423                                              HASH_UUID_MAX_BITS,
424                                              &uuid_hash_ops, CFS_HASH_REHASH);
425         if (!obd->obd_uuid_hash)
426                 GOTO(err_hash, err = -ENOMEM);
427
428         /* create a nid-export lustre hash */
429         obd->obd_nid_hash = cfs_hash_create("NID_HASH",
430                                             HASH_NID_CUR_BITS,
431                                             HASH_NID_MAX_BITS,
432                                             &nid_hash_ops, CFS_HASH_REHASH);
433         if (!obd->obd_nid_hash)
434                 GOTO(err_hash, err = -ENOMEM);
435
436         /* create a nid-stats lustre hash */
437         obd->obd_nid_stats_hash = cfs_hash_create("NID_STATS",
438                                                   HASH_NID_STATS_CUR_BITS,
439                                                   HASH_NID_STATS_MAX_BITS,
440                                                   &nid_stat_hash_ops, CFS_HASH_REHASH);
441         if (!obd->obd_nid_stats_hash)
442                 GOTO(err_hash, err = -ENOMEM);
443
444         exp = class_new_export(obd, &obd->obd_uuid);
445         if (IS_ERR(exp))
446                 GOTO(err_hash, err = PTR_ERR(exp));
447
448         obd->obd_self_export = exp;
449         cfs_list_del_init(&exp->exp_obd_chain_timed);
450         class_export_put(exp);
451
452         err = obd_setup(obd, lcfg);
453         if (err)
454                 GOTO(err_exp, err);
455
456         obd->obd_set_up = 1;
457
458         cfs_spin_lock(&obd->obd_dev_lock);
459         /* cleanup drops this */
460         class_incref(obd, "setup", obd);
461         cfs_spin_unlock(&obd->obd_dev_lock);
462
463         CDEBUG(D_IOCTL, "finished setup of obd %s (uuid %s)\n",
464                obd->obd_name, obd->obd_uuid.uuid);
465
466         RETURN(0);
467 err_exp:
468         if (obd->obd_self_export) {
469                 class_unlink_export(obd->obd_self_export);
470                 obd->obd_self_export = NULL;
471         }
472 err_hash:
473         if (obd->obd_uuid_hash) {
474                 cfs_hash_putref(obd->obd_uuid_hash);
475                 obd->obd_uuid_hash = NULL;
476         }
477         if (obd->obd_nid_hash) {
478                 cfs_hash_putref(obd->obd_nid_hash);
479                 obd->obd_nid_hash = NULL;
480         }
481         if (obd->obd_nid_stats_hash) {
482                 cfs_hash_putref(obd->obd_nid_stats_hash);
483                 obd->obd_nid_stats_hash = NULL;
484         }
485         obd->obd_starting = 0;
486         CERROR("setup %s failed (%d)\n", obd->obd_name, err);
487         return err;
488 }
489
490 /** We have finished using this obd and are ready to destroy it.
491  * There can be no more references to this obd.
492  */
493 int class_detach(struct obd_device *obd, struct lustre_cfg *lcfg)
494 {
495         ENTRY;
496
497         if (obd->obd_set_up) {
498                 CERROR("OBD device %d still set up\n", obd->obd_minor);
499                 RETURN(-EBUSY);
500         }
501
502         cfs_spin_lock(&obd->obd_dev_lock);
503         if (!obd->obd_attached) {
504                 cfs_spin_unlock(&obd->obd_dev_lock);
505                 CERROR("OBD device %d not attached\n", obd->obd_minor);
506                 RETURN(-ENODEV);
507         }
508         obd->obd_attached = 0;
509         cfs_spin_unlock(&obd->obd_dev_lock);
510
511         CDEBUG(D_IOCTL, "detach on obd %s (uuid %s)\n",
512                obd->obd_name, obd->obd_uuid.uuid);
513
514         class_decref(obd, "attach", obd);
515
516         /* not strictly necessary, but cleans up eagerly */
517         obd_zombie_impexp_cull();
518
519         RETURN(0);
520 }
521
522 /** Start shutting down the obd.  There may be in-progess ops when
523  * this is called.  We tell them to start shutting down with a call
524  * to class_disconnect_exports().
525  */
526 int class_cleanup(struct obd_device *obd, struct lustre_cfg *lcfg)
527 {
528         int err = 0;
529         char *flag;
530         ENTRY;
531
532         OBD_RACE(OBD_FAIL_LDLM_RECOV_CLIENTS);
533
534         if (!obd->obd_set_up) {
535                 CERROR("Device %d not setup\n", obd->obd_minor);
536                 RETURN(-ENODEV);
537         }
538
539         cfs_spin_lock(&obd->obd_dev_lock);
540         if (obd->obd_stopping) {
541                 cfs_spin_unlock(&obd->obd_dev_lock);
542                 CERROR("OBD %d already stopping\n", obd->obd_minor);
543                 RETURN(-ENODEV);
544         }
545         /* Leave this on forever */
546         obd->obd_stopping = 1;
547         cfs_spin_unlock(&obd->obd_dev_lock);
548
549         if (lcfg->lcfg_bufcount >= 2 && LUSTRE_CFG_BUFLEN(lcfg, 1) > 0) {
550                 for (flag = lustre_cfg_string(lcfg, 1); *flag != 0; flag++)
551                         switch (*flag) {
552                         case 'F':
553                                 obd->obd_force = 1;
554                                 break;
555                         case 'A':
556                                 LCONSOLE_WARN("Failing over %s\n",
557                                               obd->obd_name);
558                                 obd->obd_fail = 1;
559                                 obd->obd_no_transno = 1;
560                                 obd->obd_no_recov = 1;
561                                 if (OBP(obd, iocontrol)) {
562                                         obd_iocontrol(OBD_IOC_SYNC,
563                                                       obd->obd_self_export,
564                                                       0, NULL, NULL);
565                                 }
566                                 break;
567                         default:
568                                 CERROR("Unrecognised flag '%c'\n", *flag);
569                         }
570         }
571
572         LASSERT(obd->obd_self_export);
573
574         /* The three references that should be remaining are the
575          * obd_self_export and the attach and setup references. */
576         if (cfs_atomic_read(&obd->obd_refcount) > 3) {
577                 /* refcounf - 3 might be the number of real exports
578                    (excluding self export). But class_incref is called
579                    by other things as well, so don't count on it. */
580                 CDEBUG(D_IOCTL, "%s: forcing exports to disconnect: %d\n",
581                        obd->obd_name, cfs_atomic_read(&obd->obd_refcount) - 3);
582                 dump_exports(obd, 0);
583                 class_disconnect_exports(obd);
584         }
585
586         /* destroy an uuid-export hash body */
587         if (obd->obd_uuid_hash) {
588                 cfs_hash_putref(obd->obd_uuid_hash);
589                 obd->obd_uuid_hash = NULL;
590         }
591
592         /* destroy a nid-export hash body */
593         if (obd->obd_nid_hash) {
594                 cfs_hash_putref(obd->obd_nid_hash);
595                 obd->obd_nid_hash = NULL;
596         }
597
598         /* destroy a nid-stats hash body */
599         if (obd->obd_nid_stats_hash) {
600                 cfs_hash_putref(obd->obd_nid_stats_hash);
601                 obd->obd_nid_stats_hash = NULL;
602         }
603
604         /* Precleanup, we must make sure all exports get destroyed. */
605         err = obd_precleanup(obd, OBD_CLEANUP_EXPORTS);
606         if (err)
607                 CERROR("Precleanup %s returned %d\n",
608                        obd->obd_name, err);
609         class_decref(obd, "setup", obd);
610         obd->obd_set_up = 0;
611         RETURN(0);
612 }
613
614 struct obd_device *class_incref(struct obd_device *obd,
615                                 const char *scope, const void *source)
616 {
617         lu_ref_add_atomic(&obd->obd_reference, scope, source);
618         cfs_atomic_inc(&obd->obd_refcount);
619         CDEBUG(D_INFO, "incref %s (%p) now %d\n", obd->obd_name, obd,
620                cfs_atomic_read(&obd->obd_refcount));
621
622         return obd;
623 }
624
625 void class_decref(struct obd_device *obd, const char *scope, const void *source)
626 {
627         int err;
628         int refs;
629
630         cfs_spin_lock(&obd->obd_dev_lock);
631         cfs_atomic_dec(&obd->obd_refcount);
632         refs = cfs_atomic_read(&obd->obd_refcount);
633         cfs_spin_unlock(&obd->obd_dev_lock);
634         lu_ref_del(&obd->obd_reference, scope, source);
635
636         CDEBUG(D_INFO, "Decref %s (%p) now %d\n", obd->obd_name, obd, refs);
637
638         if ((refs == 1) && obd->obd_stopping) {
639                 /* All exports have been destroyed; there should
640                    be no more in-progress ops by this point.*/
641
642                 cfs_spin_lock(&obd->obd_self_export->exp_lock);
643                 obd->obd_self_export->exp_flags |= exp_flags_from_obd(obd);
644                 cfs_spin_unlock(&obd->obd_self_export->exp_lock);
645
646                 /* note that we'll recurse into class_decref again */
647                 class_unlink_export(obd->obd_self_export);
648                 return;
649         }
650
651         if (refs == 0) {
652                 CDEBUG(D_CONFIG, "finishing cleanup of obd %s (%s)\n",
653                        obd->obd_name, obd->obd_uuid.uuid);
654                 LASSERT(!obd->obd_attached);
655                 if (obd->obd_stopping) {
656                         /* If we're not stopping, we were never set up */
657                         err = obd_cleanup(obd);
658                         if (err)
659                                 CERROR("Cleanup %s returned %d\n",
660                                        obd->obd_name, err);
661                 }
662                 if (OBP(obd, detach)) {
663                         err = OBP(obd, detach)(obd);
664                         if (err)
665                                 CERROR("Detach returned %d\n", err);
666                 }
667                 class_release_dev(obd);
668         }
669 }
670
671 /** Add a failover nid location.
672  * Client obd types contact server obd types using this nid list.
673  */
674 int class_add_conn(struct obd_device *obd, struct lustre_cfg *lcfg)
675 {
676         struct obd_import *imp;
677         struct obd_uuid uuid;
678         int rc;
679         ENTRY;
680
681         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1 ||
682             LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(struct obd_uuid)) {
683                 CERROR("invalid conn_uuid\n");
684                 RETURN(-EINVAL);
685         }
686         if (strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) &&
687             strcmp(obd->obd_type->typ_name, LUSTRE_OSC_NAME) &&
688             strcmp(obd->obd_type->typ_name, LUSTRE_MGC_NAME)) {
689                 CERROR("can't add connection on non-client dev\n");
690                 RETURN(-EINVAL);
691         }
692
693         imp = obd->u.cli.cl_import;
694         if (!imp) {
695                 CERROR("try to add conn on immature client dev\n");
696                 RETURN(-EINVAL);
697         }
698
699         obd_str2uuid(&uuid, lustre_cfg_string(lcfg, 1));
700         rc = obd_add_conn(imp, &uuid, lcfg->lcfg_num);
701
702         RETURN(rc);
703 }
704
705 /** Remove a failover nid location.
706  */
707 int class_del_conn(struct obd_device *obd, struct lustre_cfg *lcfg)
708 {
709         struct obd_import *imp;
710         struct obd_uuid uuid;
711         int rc;
712         ENTRY;
713
714         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1 ||
715             LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(struct obd_uuid)) {
716                 CERROR("invalid conn_uuid\n");
717                 RETURN(-EINVAL);
718         }
719         if (strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) &&
720             strcmp(obd->obd_type->typ_name, LUSTRE_OSC_NAME)) {
721                 CERROR("can't del connection on non-client dev\n");
722                 RETURN(-EINVAL);
723         }
724
725         imp = obd->u.cli.cl_import;
726         if (!imp) {
727                 CERROR("try to del conn on immature client dev\n");
728                 RETURN(-EINVAL);
729         }
730
731         obd_str2uuid(&uuid, lustre_cfg_string(lcfg, 1));
732         rc = obd_del_conn(imp, &uuid);
733
734         RETURN(rc);
735 }
736
737 CFS_LIST_HEAD(lustre_profile_list);
738
739 struct lustre_profile *class_get_profile(const char * prof)
740 {
741         struct lustre_profile *lprof;
742
743         ENTRY;
744         cfs_list_for_each_entry(lprof, &lustre_profile_list, lp_list) {
745                 if (!strcmp(lprof->lp_profile, prof)) {
746                         RETURN(lprof);
747                 }
748         }
749         RETURN(NULL);
750 }
751
752 /** Create a named "profile".
753  * This defines the mdc and osc names to use for a client.
754  * This also is used to define the lov to be used by a mdt.
755  */
756 int class_add_profile(int proflen, char *prof, int osclen, char *osc,
757                       int mdclen, char *mdc)
758 {
759         struct lustre_profile *lprof;
760         int err = 0;
761         ENTRY;
762
763         CDEBUG(D_CONFIG, "Add profile %s\n", prof);
764
765         OBD_ALLOC(lprof, sizeof(*lprof));
766         if (lprof == NULL)
767                 RETURN(-ENOMEM);
768         CFS_INIT_LIST_HEAD(&lprof->lp_list);
769
770         LASSERT(proflen == (strlen(prof) + 1));
771         OBD_ALLOC(lprof->lp_profile, proflen);
772         if (lprof->lp_profile == NULL)
773                 GOTO(out, err = -ENOMEM);
774         memcpy(lprof->lp_profile, prof, proflen);
775
776         LASSERT(osclen == (strlen(osc) + 1));
777         OBD_ALLOC(lprof->lp_dt, osclen);
778         if (lprof->lp_dt == NULL)
779                 GOTO(out, err = -ENOMEM);
780         memcpy(lprof->lp_dt, osc, osclen);
781
782         if (mdclen > 0) {
783                 LASSERT(mdclen == (strlen(mdc) + 1));
784                 OBD_ALLOC(lprof->lp_md, mdclen);
785                 if (lprof->lp_md == NULL)
786                         GOTO(out, err = -ENOMEM);
787                 memcpy(lprof->lp_md, mdc, mdclen);
788         }
789
790         cfs_list_add(&lprof->lp_list, &lustre_profile_list);
791         RETURN(err);
792
793 out:
794         if (lprof->lp_md)
795                 OBD_FREE(lprof->lp_md, mdclen);
796         if (lprof->lp_dt)
797                 OBD_FREE(lprof->lp_dt, osclen);
798         if (lprof->lp_profile)
799                 OBD_FREE(lprof->lp_profile, proflen);
800         OBD_FREE(lprof, sizeof(*lprof));
801         RETURN(err);
802 }
803
804 void class_del_profile(const char *prof)
805 {
806         struct lustre_profile *lprof;
807         ENTRY;
808
809         CDEBUG(D_CONFIG, "Del profile %s\n", prof);
810
811         lprof = class_get_profile(prof);
812         if (lprof) {
813                 cfs_list_del(&lprof->lp_list);
814                 OBD_FREE(lprof->lp_profile, strlen(lprof->lp_profile) + 1);
815                 OBD_FREE(lprof->lp_dt, strlen(lprof->lp_dt) + 1);
816                 if (lprof->lp_md)
817                         OBD_FREE(lprof->lp_md, strlen(lprof->lp_md) + 1);
818                 OBD_FREE(lprof, sizeof *lprof);
819         }
820         EXIT;
821 }
822
823 /* COMPAT_146 */
824 void class_del_profiles(void)
825 {
826         struct lustre_profile *lprof, *n;
827         ENTRY;
828
829         cfs_list_for_each_entry_safe(lprof, n, &lustre_profile_list, lp_list) {
830                 cfs_list_del(&lprof->lp_list);
831                 OBD_FREE(lprof->lp_profile, strlen(lprof->lp_profile) + 1);
832                 OBD_FREE(lprof->lp_dt, strlen(lprof->lp_dt) + 1);
833                 if (lprof->lp_md)
834                         OBD_FREE(lprof->lp_md, strlen(lprof->lp_md) + 1);
835                 OBD_FREE(lprof, sizeof *lprof);
836         }
837         EXIT;
838 }
839
840 static int class_set_global(char *ptr, int val) {
841         ENTRY;
842
843         if (class_match_param(ptr, PARAM_AT_MIN, NULL) == 0)
844             at_min = val;
845         else if (class_match_param(ptr, PARAM_AT_MAX, NULL) == 0)
846                 at_max = val;
847         else if (class_match_param(ptr, PARAM_AT_EXTRA, NULL) == 0)
848                 at_extra = val;
849         else if (class_match_param(ptr, PARAM_AT_EARLY_MARGIN, NULL) == 0)
850                 at_early_margin = val;
851         else if (class_match_param(ptr, PARAM_AT_HISTORY, NULL) == 0)
852                 at_history = val;
853         else
854                 RETURN(-EINVAL);
855
856         CDEBUG(D_IOCTL, "global %s = %d\n", ptr, val);
857
858         RETURN(0);
859 }
860
861
862 /* We can't call ll_process_config directly because it lives in a module that
863    must be loaded after this one. */
864 static int (*client_process_config)(struct lustre_cfg *lcfg) = NULL;
865
866 void lustre_register_client_process_config(int (*cpc)(struct lustre_cfg *lcfg))
867 {
868         client_process_config = cpc;
869 }
870 EXPORT_SYMBOL(lustre_register_client_process_config);
871
872 /** Process configuration commands given in lustre_cfg form.
873  * These may come from direct calls (e.g. class_manual_cleanup)
874  * or processing the config llog, or ioctl from lctl.
875  */
876 int class_process_config(struct lustre_cfg *lcfg)
877 {
878         struct obd_device *obd;
879         int err;
880
881         LASSERT(lcfg && !IS_ERR(lcfg));
882         CDEBUG(D_IOCTL, "processing cmd: %x\n", lcfg->lcfg_command);
883
884         /* Commands that don't need a device */
885         switch(lcfg->lcfg_command) {
886         case LCFG_ATTACH: {
887                 err = class_attach(lcfg);
888                 GOTO(out, err);
889         }
890         case LCFG_ADD_UUID: {
891                 CDEBUG(D_IOCTL, "adding mapping from uuid %s to nid "LPX64
892                        " (%s)\n", lustre_cfg_string(lcfg, 1),
893                        lcfg->lcfg_nid, libcfs_nid2str(lcfg->lcfg_nid));
894
895                 err = class_add_uuid(lustre_cfg_string(lcfg, 1), lcfg->lcfg_nid);
896                 GOTO(out, err);
897         }
898         case LCFG_DEL_UUID: {
899                 CDEBUG(D_IOCTL, "removing mappings for uuid %s\n",
900                        (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) == 0)
901                        ? "<all uuids>" : lustre_cfg_string(lcfg, 1));
902
903                 err = class_del_uuid(lustre_cfg_string(lcfg, 1));
904                 GOTO(out, err);
905         }
906         case LCFG_MOUNTOPT: {
907                 CDEBUG(D_IOCTL, "mountopt: profile %s osc %s mdc %s\n",
908                        lustre_cfg_string(lcfg, 1),
909                        lustre_cfg_string(lcfg, 2),
910                        lustre_cfg_string(lcfg, 3));
911                 /* set these mount options somewhere, so ll_fill_super
912                  * can find them. */
913                 err = class_add_profile(LUSTRE_CFG_BUFLEN(lcfg, 1),
914                                         lustre_cfg_string(lcfg, 1),
915                                         LUSTRE_CFG_BUFLEN(lcfg, 2),
916                                         lustre_cfg_string(lcfg, 2),
917                                         LUSTRE_CFG_BUFLEN(lcfg, 3),
918                                         lustre_cfg_string(lcfg, 3));
919                 GOTO(out, err);
920         }
921         case LCFG_DEL_MOUNTOPT: {
922                 CDEBUG(D_IOCTL, "mountopt: profile %s\n",
923                        lustre_cfg_string(lcfg, 1));
924                 class_del_profile(lustre_cfg_string(lcfg, 1));
925                 GOTO(out, err = 0);
926         }
927         case LCFG_SET_TIMEOUT: {
928                 CDEBUG(D_IOCTL, "changing lustre timeout from %d to %d\n",
929                        obd_timeout, lcfg->lcfg_num);
930                 obd_timeout = max(lcfg->lcfg_num, 1U);
931                 GOTO(out, err = 0);
932         }
933         case LCFG_SET_LDLM_TIMEOUT: {
934                 CDEBUG(D_IOCTL, "changing lustre ldlm_timeout from %d to %d\n",
935                        ldlm_timeout, lcfg->lcfg_num);
936                 ldlm_timeout = max(lcfg->lcfg_num, 1U);
937                 if (ldlm_timeout >= obd_timeout)
938                         ldlm_timeout = max(obd_timeout / 3, 1U);
939
940                 GOTO(out, err = 0);
941         }
942         case LCFG_SET_UPCALL: {
943                 LCONSOLE_ERROR_MSG(0x15a, "recovery upcall is deprecated\n");
944                 /* COMPAT_146 Don't fail on old configs */
945                 GOTO(out, err = 0);
946         }
947         case LCFG_MARKER: {
948                 struct cfg_marker *marker;
949                 marker = lustre_cfg_buf(lcfg, 1);
950                 CDEBUG(D_IOCTL, "marker %d (%#x) %.16s %s\n", marker->cm_step,
951                        marker->cm_flags, marker->cm_tgtname, marker->cm_comment);
952                 GOTO(out, err = 0);
953         }
954         case LCFG_PARAM: {
955                 char *tmp;
956                 /* llite has no obd */
957                 if ((class_match_param(lustre_cfg_string(lcfg, 1),
958                                        PARAM_LLITE, 0) == 0) &&
959                     client_process_config) {
960                         err = (*client_process_config)(lcfg);
961                         GOTO(out, err);
962                 } else if ((class_match_param(lustre_cfg_string(lcfg, 1),
963                                               PARAM_SYS, &tmp) == 0)) {
964                         /* Global param settings */
965                         err = class_set_global(tmp, lcfg->lcfg_num);
966                         /* Note that since LCFG_PARAM is LCFG_REQUIRED, new
967                            unknown globals would cause config to fail */
968                         if (err)
969                                 CWARN("Ignoring unknown param %s\n", tmp);
970                         GOTO(out, 0);
971                 }
972
973                 /* Fall through */
974                 break;
975         }
976         }
977
978         /* Commands that require a device */
979         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
980         if (obd == NULL) {
981                 if (!LUSTRE_CFG_BUFLEN(lcfg, 0))
982                         CERROR("this lcfg command requires a device name\n");
983                 else
984                         CERROR("no device for: %s\n",
985                                lustre_cfg_string(lcfg, 0));
986
987                 GOTO(out, err = -EINVAL);
988         }
989
990         switch(lcfg->lcfg_command) {
991         case LCFG_SETUP: {
992                 err = class_setup(obd, lcfg);
993                 GOTO(out, err);
994         }
995         case LCFG_DETACH: {
996                 err = class_detach(obd, lcfg);
997                 GOTO(out, err = 0);
998         }
999         case LCFG_CLEANUP: {
1000                 err = class_cleanup(obd, lcfg);
1001                 GOTO(out, err = 0);
1002         }
1003         case LCFG_ADD_CONN: {
1004                 err = class_add_conn(obd, lcfg);
1005                 GOTO(out, err = 0);
1006         }
1007         case LCFG_DEL_CONN: {
1008                 err = class_del_conn(obd, lcfg);
1009                 GOTO(out, err = 0);
1010         }
1011         case LCFG_POOL_NEW: {
1012                 err = obd_pool_new(obd, lustre_cfg_string(lcfg, 2));
1013                 GOTO(out, err = 0);
1014                 break;
1015         }
1016         case LCFG_POOL_ADD: {
1017                 err = obd_pool_add(obd, lustre_cfg_string(lcfg, 2),
1018                                    lustre_cfg_string(lcfg, 3));
1019                 GOTO(out, err = 0);
1020                 break;
1021         }
1022         case LCFG_POOL_REM: {
1023                 err = obd_pool_rem(obd, lustre_cfg_string(lcfg, 2),
1024                                    lustre_cfg_string(lcfg, 3));
1025                 GOTO(out, err = 0);
1026                 break;
1027         }
1028         case LCFG_POOL_DEL: {
1029                 err = obd_pool_del(obd, lustre_cfg_string(lcfg, 2));
1030                 GOTO(out, err = 0);
1031                 break;
1032         }
1033         default: {
1034                 err = obd_process_config(obd, sizeof(*lcfg), lcfg);
1035                 GOTO(out, err);
1036
1037         }
1038         }
1039 out:
1040         if ((err < 0) && !(lcfg->lcfg_command & LCFG_REQUIRED)) {
1041                 CWARN("Ignoring error %d on optional command %#x\n", err,
1042                       lcfg->lcfg_command);
1043                 err = 0;
1044         }
1045         return err;
1046 }
1047
1048 int class_process_proc_param(char *prefix, struct lprocfs_vars *lvars,
1049                              struct lustre_cfg *lcfg, void *data)
1050 {
1051 #ifdef __KERNEL__
1052         struct lprocfs_vars *var;
1053         char *key, *sval;
1054         int i, keylen, vallen;
1055         int matched = 0, j = 0;
1056         int rc = 0;
1057         int skip = 0;
1058         ENTRY;
1059
1060         if (lcfg->lcfg_command != LCFG_PARAM) {
1061                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1062                 RETURN(-EINVAL);
1063         }
1064
1065         /* e.g. tunefs.lustre --param mdt.group_upcall=foo /r/tmp/lustre-mdt
1066            or   lctl conf_param lustre-MDT0000.mdt.group_upcall=bar
1067            or   lctl conf_param lustre-OST0000.osc.max_dirty_mb=36 */
1068         for (i = 1; i < lcfg->lcfg_bufcount; i++) {
1069                 key = lustre_cfg_buf(lcfg, i);
1070                 /* Strip off prefix */
1071                 class_match_param(key, prefix, &key);
1072                 sval = strchr(key, '=');
1073                 if (!sval || (*(sval + 1) == 0)) {
1074                         CERROR("Can't parse param %s (missing '=')\n", key);
1075                         /* rc = -EINVAL;        continue parsing other params */
1076                         continue;
1077                 }
1078                 keylen = sval - key;
1079                 sval++;
1080                 vallen = strlen(sval);
1081                 matched = 0;
1082                 j = 0;
1083                 /* Search proc entries */
1084                 while (lvars[j].name) {
1085                         var = &lvars[j];
1086                         if (class_match_param(key, (char *)var->name, 0) == 0 &&
1087                             keylen == strlen(var->name)) {
1088                                 matched++;
1089                                 rc = -EROFS;
1090                                 if (var->write_fptr) {
1091                                         mm_segment_t oldfs;
1092                                         oldfs = get_fs();
1093                                         set_fs(KERNEL_DS);
1094                                         rc = (var->write_fptr)(NULL, sval,
1095                                                                vallen, data);
1096                                         set_fs(oldfs);
1097                                 }
1098                                 break;
1099                         }
1100                         j++;
1101                 }
1102                 if (!matched) {
1103                         /* If the prefix doesn't match, return error so we
1104                            can pass it down the stack */
1105                         if (strnchr(key, keylen, '.'))
1106                             RETURN(-ENOSYS);
1107                         CERROR("%s: unknown param %s\n",
1108                                (char *)lustre_cfg_string(lcfg, 0), key);
1109                         /* rc = -EINVAL;        continue parsing other params */
1110                         skip++;
1111                 } else if (rc < 0) {
1112                         CERROR("writing proc entry %s err %d\n",
1113                                var->name, rc);
1114                         rc = 0;
1115                 } else {
1116                         LCONSOLE_INFO("%s.%.*s: set parameter %.*s=%s\n",
1117                                       lustre_cfg_string(lcfg, 0),
1118                                       (int)strlen(prefix) - 1, prefix,
1119                                       (int)(sval - key - 1), key, sval);
1120                 }
1121         }
1122
1123         if (rc > 0)
1124                 rc = 0;
1125         if (!rc && skip)
1126                 rc = skip;
1127         RETURN(rc);
1128 #else
1129         CDEBUG(D_CONFIG, "liblustre can't process params.\n");
1130         /* Don't throw config error */
1131         RETURN(0);
1132 #endif
1133 }
1134
1135 int class_config_dump_handler(struct llog_handle * handle,
1136                               struct llog_rec_hdr *rec, void *data);
1137
1138 #ifdef __KERNEL__
1139 extern int lustre_check_exclusion(struct super_block *sb, char *svname);
1140 #else
1141 #define lustre_check_exclusion(a,b)  0
1142 #endif
1143
1144 /** Parse a configuration llog, doing various manipulations on them
1145  * for various reasons, (modifications for compatibility, skip obsolete
1146  * records, change uuids, etc), then class_process_config() resulting
1147  * net records.
1148  */
1149 static int class_config_llog_handler(struct llog_handle * handle,
1150                                      struct llog_rec_hdr *rec, void *data)
1151 {
1152         struct config_llog_instance *clli = data;
1153         int cfg_len = rec->lrh_len;
1154         char *cfg_buf = (char*) (rec + 1);
1155         int rc = 0;
1156         ENTRY;
1157
1158         //class_config_dump_handler(handle, rec, data);
1159
1160         switch (rec->lrh_type) {
1161         case OBD_CFG_REC: {
1162                 struct lustre_cfg *lcfg, *lcfg_new;
1163                 struct lustre_cfg_bufs bufs;
1164                 char *inst_name = NULL;
1165                 int inst_len = 0;
1166                 int inst = 0, swab = 0;
1167
1168                 lcfg = (struct lustre_cfg *)cfg_buf;
1169                 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
1170                         lustre_swab_lustre_cfg(lcfg);
1171                         swab = 1;
1172                 }
1173
1174                 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
1175                 if (rc)
1176                         GOTO(out, rc);
1177
1178                 /* Figure out config state info */
1179                 if (lcfg->lcfg_command == LCFG_MARKER) {
1180                         struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
1181                         lustre_swab_cfg_marker(marker, swab,
1182                                                LUSTRE_CFG_BUFLEN(lcfg, 1));
1183                         CDEBUG(D_CONFIG, "Marker, inst_flg=%#x mark_flg=%#x\n",
1184                                clli->cfg_flags, marker->cm_flags);
1185                         if (marker->cm_flags & CM_START) {
1186                                 /* all previous flags off */
1187                                 clli->cfg_flags = CFG_F_MARKER;
1188                                 if (marker->cm_flags & CM_SKIP) {
1189                                         clli->cfg_flags |= CFG_F_SKIP;
1190                                         CDEBUG(D_CONFIG, "SKIP #%d\n",
1191                                                marker->cm_step);
1192                                 } else if ((marker->cm_flags & CM_EXCLUDE) ||
1193                                            (clli->cfg_sb &&
1194                                             lustre_check_exclusion(clli->cfg_sb,
1195                                                          marker->cm_tgtname))) {
1196                                         clli->cfg_flags |= CFG_F_EXCLUDE;
1197                                         CDEBUG(D_CONFIG, "EXCLUDE %d\n",
1198                                                marker->cm_step);
1199                                 }
1200                         } else if (marker->cm_flags & CM_END) {
1201                                 clli->cfg_flags = 0;
1202                         }
1203                 }
1204                 /* A config command without a start marker before it is
1205                    illegal (post 146) */
1206                 if (!(clli->cfg_flags & CFG_F_COMPAT146) &&
1207                     !(clli->cfg_flags & CFG_F_MARKER) &&
1208                     (lcfg->lcfg_command != LCFG_MARKER)) {
1209                         CWARN("Config not inside markers, ignoring! "
1210                               "(inst: %s, uuid: %s, flags: %#x)\n",
1211                               clli->cfg_instance ? clli->cfg_instance : "<null>",
1212                               clli->cfg_uuid.uuid, clli->cfg_flags);
1213                         clli->cfg_flags |= CFG_F_SKIP;
1214                 }
1215                 if (clli->cfg_flags & CFG_F_SKIP) {
1216                         CDEBUG(D_CONFIG, "skipping %#x\n",
1217                                clli->cfg_flags);
1218                         rc = 0;
1219                         /* No processing! */
1220                         break;
1221                 }
1222
1223                 /*
1224                  * For interoperability between 1.8 and 2.0,
1225                  * rename "mds" obd device type to "mdt".
1226                  */
1227                 {
1228                         char *typename = lustre_cfg_string(lcfg, 1);
1229                         char *index = lustre_cfg_string(lcfg, 2);
1230
1231                         if ((lcfg->lcfg_command == LCFG_ATTACH && typename &&
1232                              strcmp(typename, "mds") == 0)) {
1233                                 CWARN("For 1.8 interoperability, rename obd "
1234                                        "type from mds to mdt\n");
1235                                 typename[2] = 't';
1236                         }
1237                         if ((lcfg->lcfg_command == LCFG_SETUP && index &&
1238                              strcmp(index, "type") == 0)) {
1239                                 CDEBUG(D_INFO, "For 1.8 interoperability, "
1240                                        "set this index to '0'\n");
1241                                 index[0] = '0';
1242                                 index[1] = 0;
1243                         }
1244                 }
1245
1246                 if ((clli->cfg_flags & CFG_F_EXCLUDE) &&
1247                     (lcfg->lcfg_command == LCFG_LOV_ADD_OBD))
1248                         /* Add inactive instead */
1249                         lcfg->lcfg_command = LCFG_LOV_ADD_INA;
1250
1251                 lustre_cfg_bufs_init(&bufs, lcfg);
1252
1253                 if (clli && clli->cfg_instance &&
1254                     LUSTRE_CFG_BUFLEN(lcfg, 0) > 0){
1255                         inst = 1;
1256                         inst_len = LUSTRE_CFG_BUFLEN(lcfg, 0) +
1257                                 strlen(clli->cfg_instance) + 1;
1258                         OBD_ALLOC(inst_name, inst_len);
1259                         if (inst_name == NULL)
1260                                 GOTO(out, rc = -ENOMEM);
1261                         sprintf(inst_name, "%s-%s",
1262                                 lustre_cfg_string(lcfg, 0),
1263                                 clli->cfg_instance);
1264                         lustre_cfg_bufs_set_string(&bufs, 0, inst_name);
1265                         CDEBUG(D_CONFIG, "cmd %x, instance name: %s\n",
1266                                lcfg->lcfg_command, inst_name);
1267                 }
1268
1269                 /* we override the llog's uuid for clients, to insure they
1270                 are unique */
1271                 if (clli && clli->cfg_instance &&
1272                     lcfg->lcfg_command == LCFG_ATTACH) {
1273                         lustre_cfg_bufs_set_string(&bufs, 2,
1274                                                    clli->cfg_uuid.uuid);
1275                 }
1276                 /*
1277                  * sptlrpc config record, we expect 2 data segments:
1278                  *  [0]: fs_name/target_name,
1279                  *  [1]: rule string
1280                  * moving them to index [1] and [2], and insert MGC's
1281                  * obdname at index [0].
1282                  */
1283                 if (clli && clli->cfg_instance == NULL &&
1284                     lcfg->lcfg_command == LCFG_SPTLRPC_CONF) {
1285                         lustre_cfg_bufs_set(&bufs, 2, bufs.lcfg_buf[1],
1286                                             bufs.lcfg_buflen[1]);
1287                         lustre_cfg_bufs_set(&bufs, 1, bufs.lcfg_buf[0],
1288                                             bufs.lcfg_buflen[0]);
1289                         lustre_cfg_bufs_set_string(&bufs, 0,
1290                                                    clli->cfg_obdname);
1291                 }
1292
1293                 lcfg_new = lustre_cfg_new(lcfg->lcfg_command, &bufs);
1294
1295                 lcfg_new->lcfg_num   = lcfg->lcfg_num;
1296                 lcfg_new->lcfg_flags = lcfg->lcfg_flags;
1297
1298                 /* XXX Hack to try to remain binary compatible with
1299                  * pre-newconfig logs */
1300                 if (lcfg->lcfg_nal != 0 &&      /* pre-newconfig log? */
1301                     (lcfg->lcfg_nid >> 32) == 0) {
1302                         __u32 addr = (__u32)(lcfg->lcfg_nid & 0xffffffff);
1303
1304                         lcfg_new->lcfg_nid =
1305                                 LNET_MKNID(LNET_MKNET(lcfg->lcfg_nal, 0), addr);
1306                         CWARN("Converted pre-newconfig NAL %d NID %x to %s\n",
1307                               lcfg->lcfg_nal, addr,
1308                               libcfs_nid2str(lcfg_new->lcfg_nid));
1309                 } else {
1310                         lcfg_new->lcfg_nid = lcfg->lcfg_nid;
1311                 }
1312
1313                 lcfg_new->lcfg_nal = 0; /* illegal value for obsolete field */
1314
1315                 rc = class_process_config(lcfg_new);
1316                 lustre_cfg_free(lcfg_new);
1317
1318                 if (inst)
1319                         OBD_FREE(inst_name, inst_len);
1320                 break;
1321         }
1322         default:
1323                 CERROR("Unknown llog record type %#x encountered\n",
1324                        rec->lrh_type);
1325                 break;
1326         }
1327 out:
1328         if (rc) {
1329                 CERROR("Err %d on cfg command:\n", rc);
1330                 class_config_dump_handler(handle, rec, data);
1331         }
1332         RETURN(rc);
1333 }
1334
1335 int class_config_parse_llog(struct llog_ctxt *ctxt, char *name,
1336                             struct config_llog_instance *cfg)
1337 {
1338         struct llog_process_cat_data cd = {0, 0};
1339         struct llog_handle *llh;
1340         int rc, rc2;
1341         ENTRY;
1342
1343         CDEBUG(D_INFO, "looking up llog %s\n", name);
1344         rc = llog_create(ctxt, &llh, NULL, name);
1345         if (rc)
1346                 RETURN(rc);
1347
1348         rc = llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
1349         if (rc)
1350                 GOTO(parse_out, rc);
1351
1352         /* continue processing from where we last stopped to end-of-log */
1353         if (cfg)
1354                 cd.lpcd_first_idx = cfg->cfg_last_idx;
1355         cd.lpcd_last_idx = 0;
1356
1357         rc = llog_process(llh, class_config_llog_handler, cfg, &cd);
1358
1359         CDEBUG(D_CONFIG, "Processed log %s gen %d-%d (rc=%d)\n", name,
1360                cd.lpcd_first_idx + 1, cd.lpcd_last_idx, rc);
1361
1362         if (cfg)
1363                 cfg->cfg_last_idx = cd.lpcd_last_idx;
1364
1365 parse_out:
1366         rc2 = llog_close(llh);
1367         if (rc == 0)
1368                 rc = rc2;
1369
1370         RETURN(rc);
1371 }
1372
1373 int class_config_dump_handler(struct llog_handle * handle,
1374                               struct llog_rec_hdr *rec, void *data)
1375 {
1376         int cfg_len = rec->lrh_len;
1377         char *cfg_buf = (char*) (rec + 1);
1378         char *outstr, *ptr, *end;
1379         int rc = 0;
1380         ENTRY;
1381
1382         OBD_ALLOC(outstr, 256);
1383         end = outstr + 256;
1384         ptr = outstr;
1385         if (!outstr) {
1386                 RETURN(-ENOMEM);
1387         }
1388         if (rec->lrh_type == OBD_CFG_REC) {
1389                 struct lustre_cfg *lcfg;
1390                 int i;
1391
1392                 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
1393                 if (rc)
1394                         GOTO(out, rc);
1395                 lcfg = (struct lustre_cfg *)cfg_buf;
1396
1397                 ptr += snprintf(ptr, end-ptr, "cmd=%05x ",
1398                                 lcfg->lcfg_command);
1399                 if (lcfg->lcfg_flags) {
1400                         ptr += snprintf(ptr, end-ptr, "flags=%#08x ",
1401                                         lcfg->lcfg_flags);
1402                 }
1403                 if (lcfg->lcfg_num) {
1404                         ptr += snprintf(ptr, end-ptr, "num=%#08x ",
1405                                         lcfg->lcfg_num);
1406                 }
1407                 if (lcfg->lcfg_nid) {
1408                         ptr += snprintf(ptr, end-ptr, "nid=%s("LPX64")\n     ",
1409                                         libcfs_nid2str(lcfg->lcfg_nid),
1410                                         lcfg->lcfg_nid);
1411                 }
1412                 if (lcfg->lcfg_command == LCFG_MARKER) {
1413                         struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
1414                         ptr += snprintf(ptr, end-ptr, "marker=%d(%#x)%s '%s'",
1415                                         marker->cm_step, marker->cm_flags,
1416                                         marker->cm_tgtname, marker->cm_comment);
1417                 } else {
1418                         for (i = 0; i <  lcfg->lcfg_bufcount; i++) {
1419                                 ptr += snprintf(ptr, end-ptr, "%d:%s  ", i,
1420                                                 lustre_cfg_string(lcfg, i));
1421                         }
1422                 }
1423                 LCONSOLE(D_WARNING, "   %s\n", outstr);
1424         } else {
1425                 LCONSOLE(D_WARNING, "unhandled lrh_type: %#x\n", rec->lrh_type);
1426                 rc = -EINVAL;
1427         }
1428 out:
1429         OBD_FREE(outstr, 256);
1430         RETURN(rc);
1431 }
1432
1433 int class_config_dump_llog(struct llog_ctxt *ctxt, char *name,
1434                            struct config_llog_instance *cfg)
1435 {
1436         struct llog_handle *llh;
1437         int rc, rc2;
1438         ENTRY;
1439
1440         LCONSOLE_INFO("Dumping config log %s\n", name);
1441
1442         rc = llog_create(ctxt, &llh, NULL, name);
1443         if (rc)
1444                 RETURN(rc);
1445
1446         rc = llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
1447         if (rc)
1448                 GOTO(parse_out, rc);
1449
1450         rc = llog_process(llh, class_config_dump_handler, cfg, NULL);
1451 parse_out:
1452         rc2 = llog_close(llh);
1453         if (rc == 0)
1454                 rc = rc2;
1455
1456         LCONSOLE_INFO("End config log %s\n", name);
1457         RETURN(rc);
1458
1459 }
1460
1461 /** Call class_cleanup and class_detach.
1462  * "Manual" only in the sense that we're faking lcfg commands.
1463  */
1464 int class_manual_cleanup(struct obd_device *obd)
1465 {
1466         char                    flags[3] = "";
1467         struct lustre_cfg      *lcfg;
1468         struct lustre_cfg_bufs  bufs;
1469         int                     rc;
1470         ENTRY;
1471
1472         if (!obd) {
1473                 CERROR("empty cleanup\n");
1474                 RETURN(-EALREADY);
1475         }
1476
1477         if (obd->obd_force)
1478                 strcat(flags, "F");
1479         if (obd->obd_fail)
1480                 strcat(flags, "A");
1481
1482         CDEBUG(D_CONFIG, "Manual cleanup of %s (flags='%s')\n",
1483                obd->obd_name, flags);
1484
1485         lustre_cfg_bufs_reset(&bufs, obd->obd_name);
1486         lustre_cfg_bufs_set_string(&bufs, 1, flags);
1487         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
1488         if (!lcfg)
1489                 RETURN(-ENOMEM);
1490
1491         rc = class_process_config(lcfg);
1492         if (rc) {
1493                 CERROR("cleanup failed %d: %s\n", rc, obd->obd_name);
1494                 GOTO(out, rc);
1495         }
1496
1497         /* the lcfg is almost the same for both ops */
1498         lcfg->lcfg_command = LCFG_DETACH;
1499         rc = class_process_config(lcfg);
1500         if (rc)
1501                 CERROR("detach failed %d: %s\n", rc, obd->obd_name);
1502 out:
1503         lustre_cfg_free(lcfg);
1504         RETURN(rc);
1505 }
1506
1507 /*
1508  * uuid<->export lustre hash operations
1509  */
1510
1511 static unsigned
1512 uuid_hash(cfs_hash_t *hs,  void *key, unsigned mask)
1513 {
1514         return cfs_hash_djb2_hash(((struct obd_uuid *)key)->uuid,
1515                                   sizeof(((struct obd_uuid *)key)->uuid), mask);
1516 }
1517
1518 static void *
1519 uuid_key(cfs_hlist_node_t *hnode)
1520 {
1521         struct obd_export *exp;
1522
1523         exp = cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1524
1525         RETURN(&exp->exp_client_uuid);
1526 }
1527
1528 /*
1529  * NOTE: It is impossible to find an export that is in failed
1530  *       state with this function
1531  */
1532 static int
1533 uuid_compare(void *key, cfs_hlist_node_t *hnode)
1534 {
1535         struct obd_export *exp;
1536
1537         LASSERT(key);
1538         exp = cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1539
1540         RETURN(obd_uuid_equals((struct obd_uuid *)key,&exp->exp_client_uuid) &&
1541                !exp->exp_failed);
1542 }
1543
1544 static void *
1545 uuid_export_get(cfs_hlist_node_t *hnode)
1546 {
1547         struct obd_export *exp;
1548
1549         exp = cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1550         class_export_get(exp);
1551
1552         RETURN(exp);
1553 }
1554
1555 static void *
1556 uuid_export_put(cfs_hlist_node_t *hnode)
1557 {
1558         struct obd_export *exp;
1559
1560         exp = cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1561         class_export_put(exp);
1562
1563         RETURN(exp);
1564 }
1565
1566 static cfs_hash_ops_t uuid_hash_ops = {
1567         .hs_hash    = uuid_hash,
1568         .hs_key     = uuid_key,
1569         .hs_compare = uuid_compare,
1570         .hs_get     = uuid_export_get,
1571         .hs_put     = uuid_export_put,
1572 };
1573
1574
1575 /*
1576  * nid<->export hash operations
1577  */
1578
1579 static unsigned
1580 nid_hash(cfs_hash_t *hs,  void *key, unsigned mask)
1581 {
1582         return cfs_hash_djb2_hash(key, sizeof(lnet_nid_t), mask);
1583 }
1584
1585 static void *
1586 nid_key(cfs_hlist_node_t *hnode)
1587 {
1588         struct obd_export *exp;
1589
1590         exp = cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1591
1592         RETURN(&exp->exp_connection->c_peer.nid);
1593 }
1594
1595 /*
1596  * NOTE: It is impossible to find an export that is in failed
1597  *       state with this function
1598  */
1599 static int
1600 nid_compare(void *key, cfs_hlist_node_t *hnode)
1601 {
1602         struct obd_export *exp;
1603
1604         LASSERT(key);
1605         exp = cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1606
1607         RETURN(exp->exp_connection->c_peer.nid == *(lnet_nid_t *)key &&
1608                !exp->exp_failed);
1609 }
1610
1611 static void *
1612 nid_export_get(cfs_hlist_node_t *hnode)
1613 {
1614         struct obd_export *exp;
1615
1616         exp = cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1617         class_export_get(exp);
1618
1619         RETURN(exp);
1620 }
1621
1622 static void *
1623 nid_export_put(cfs_hlist_node_t *hnode)
1624 {
1625         struct obd_export *exp;
1626
1627         exp = cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1628         class_export_put(exp);
1629
1630         RETURN(exp);
1631 }
1632
1633 static cfs_hash_ops_t nid_hash_ops = {
1634         .hs_hash    = nid_hash,
1635         .hs_key     = nid_key,
1636         .hs_compare = nid_compare,
1637         .hs_get     = nid_export_get,
1638         .hs_put     = nid_export_put,
1639 };
1640
1641
1642 /*
1643  * nid<->nidstats hash operations
1644  */
1645
1646 static void *
1647 nidstats_key(cfs_hlist_node_t *hnode)
1648 {
1649         struct nid_stat *ns;
1650
1651         ns = cfs_hlist_entry(hnode, struct nid_stat, nid_hash);
1652
1653         RETURN(&ns->nid);
1654 }
1655
1656 static int
1657 nidstats_compare(void *key, cfs_hlist_node_t *hnode)
1658 {
1659         RETURN(*(lnet_nid_t *)nidstats_key(hnode) == *(lnet_nid_t *)key);
1660 }
1661
1662 static void *
1663 nidstats_get(cfs_hlist_node_t *hnode)
1664 {
1665         struct nid_stat *ns;
1666
1667         ns = cfs_hlist_entry(hnode, struct nid_stat, nid_hash);
1668         nidstat_getref(ns);
1669
1670         RETURN(ns);
1671 }
1672
1673 static void *
1674 nidstats_put(cfs_hlist_node_t *hnode)
1675 {
1676         struct nid_stat *ns;
1677
1678         ns = cfs_hlist_entry(hnode, struct nid_stat, nid_hash);
1679         nidstat_putref(ns);
1680
1681         RETURN(ns);
1682 }
1683
1684 static cfs_hash_ops_t nid_stat_hash_ops = {
1685         .hs_hash    = nid_hash,
1686         .hs_key     = nidstats_key,
1687         .hs_compare = nidstats_compare,
1688         .hs_get     = nidstats_get,
1689         .hs_put     = nidstats_put,
1690 };