Whamcloud - gitweb
LU-147 avoid 8k obd device amount limit
[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 *)value;
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;
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         if (strlen(uuid) >= sizeof(obd->obd_uuid)) {
292                 CERROR("uuid must be < %d bytes long\n",
293                        (int)sizeof(obd->obd_uuid));
294                 RETURN(-EINVAL);
295         }
296
297         CDEBUG(D_IOCTL, "attach type %s name: %s uuid: %s\n",
298                MKSTR(typename), MKSTR(name), MKSTR(uuid));
299
300         obd = class_newdev(typename, name, uuid);
301         if (IS_ERR(obd)) {
302                 /* Already exists or out of obds */
303                 rc = PTR_ERR(obd);
304                 obd = NULL;
305                 CERROR("Cannot create device %s of type %s : %d\n",
306                        name, typename, rc);
307                 GOTO(out, rc);
308         }
309
310         LASSERTF(obd != NULL, "Cannot get obd device %s of type %s\n",
311                  name, typename);
312         LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC,
313                  "obd %p obd_magic %08X != %08X\n",
314                  obd, obd->obd_magic, OBD_DEVICE_MAGIC);
315         LASSERTF(strncmp(obd->obd_name, name, strlen(name)) == 0,
316                  "%p obd_name %s != %s\n", obd, obd->obd_name, name);
317
318         cfs_rwlock_init(&obd->obd_pool_lock);
319         obd->obd_pool_limit = 0;
320         obd->obd_pool_slv = 0;
321
322         CFS_INIT_LIST_HEAD(&obd->obd_exports);
323         CFS_INIT_LIST_HEAD(&obd->obd_unlinked_exports);
324         CFS_INIT_LIST_HEAD(&obd->obd_delayed_exports);
325         CFS_INIT_LIST_HEAD(&obd->obd_exports_timed);
326         CFS_INIT_LIST_HEAD(&obd->obd_nid_stats);
327         cfs_spin_lock_init(&obd->obd_nid_lock);
328         cfs_spin_lock_init(&obd->obd_dev_lock);
329         cfs_sema_init(&obd->obd_dev_sem, 1);
330         cfs_spin_lock_init(&obd->obd_osfs_lock);
331         /* obd->obd_osfs_age must be set to a value in the distant
332          * past to guarantee a fresh statfs is fetched on mount. */
333         obd->obd_osfs_age = cfs_time_shift_64(-1000);
334
335         /* XXX belongs in setup not attach  */
336         cfs_init_rwsem(&obd->obd_observer_link_sem);
337         /* recovery data */
338         cfs_init_timer(&obd->obd_recovery_timer);
339         cfs_spin_lock_init(&obd->obd_recovery_task_lock);
340         cfs_waitq_init(&obd->obd_next_transno_waitq);
341         cfs_waitq_init(&obd->obd_evict_inprogress_waitq);
342         CFS_INIT_LIST_HEAD(&obd->obd_req_replay_queue);
343         CFS_INIT_LIST_HEAD(&obd->obd_lock_replay_queue);
344         CFS_INIT_LIST_HEAD(&obd->obd_final_req_queue);
345         CFS_INIT_LIST_HEAD(&obd->obd_evict_list);
346
347         llog_group_init(&obd->obd_olg, FID_SEQ_LLOG);
348
349         /* do the attach */
350         if (OBP(obd, attach)) {
351                 rc = OBP(obd,attach)(obd, sizeof *lcfg, lcfg);
352                 if (rc)
353                         GOTO(out, rc = -EINVAL);
354         }
355
356         /* Detach drops this */
357         cfs_spin_lock(&obd->obd_dev_lock);
358         cfs_atomic_set(&obd->obd_refcount, 1);
359         cfs_spin_unlock(&obd->obd_dev_lock);
360         lu_ref_init(&obd->obd_reference);
361         lu_ref_add(&obd->obd_reference, "attach", obd);
362
363         obd->obd_attached = 1;
364         CDEBUG(D_IOCTL, "OBD: dev %d attached type %s with refcount %d\n",
365                obd->obd_minor, typename, cfs_atomic_read(&obd->obd_refcount));
366         RETURN(0);
367  out:
368         if (obd != NULL) {
369                 class_release_dev(obd);
370         }
371         return rc;
372 }
373
374 /** Create hashes, self-export, and call type-specific setup.
375  * Setup is effectively the "start this obd" call.
376  */
377 int class_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
378 {
379         int err = 0;
380         struct obd_export *exp;
381         ENTRY;
382
383         LASSERT(obd != NULL);
384         LASSERTF(obd == class_num2obd(obd->obd_minor),
385                  "obd %p != obd_devs[%d] %p\n",
386                  obd, obd->obd_minor, class_num2obd(obd->obd_minor));
387         LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC,
388                  "obd %p obd_magic %08x != %08x\n",
389                  obd, obd->obd_magic, OBD_DEVICE_MAGIC);
390
391         /* have we attached a type to this device? */
392         if (!obd->obd_attached) {
393                 CERROR("Device %d not attached\n", obd->obd_minor);
394                 RETURN(-ENODEV);
395         }
396
397         if (obd->obd_set_up) {
398                 CERROR("Device %d already setup (type %s)\n",
399                        obd->obd_minor, obd->obd_type->typ_name);
400                 RETURN(-EEXIST);
401         }
402
403         /* is someone else setting us up right now? (attach inits spinlock) */
404         cfs_spin_lock(&obd->obd_dev_lock);
405         if (obd->obd_starting) {
406                 cfs_spin_unlock(&obd->obd_dev_lock);
407                 CERROR("Device %d setup in progress (type %s)\n",
408                        obd->obd_minor, obd->obd_type->typ_name);
409                 RETURN(-EEXIST);
410         }
411         /* just leave this on forever.  I can't use obd_set_up here because
412            other fns check that status, and we're not actually set up yet. */
413         obd->obd_starting = 1;
414         obd->obd_uuid_hash = NULL;
415         obd->obd_nid_hash = NULL;
416         obd->obd_nid_stats_hash = NULL;
417         cfs_spin_unlock(&obd->obd_dev_lock);
418
419         /* create an uuid-export lustre hash */
420         obd->obd_uuid_hash = cfs_hash_create("UUID_HASH",
421                                              HASH_UUID_CUR_BITS,
422                                              HASH_UUID_MAX_BITS,
423                                              HASH_UUID_BKT_BITS, 0,
424                                              CFS_HASH_MIN_THETA,
425                                              CFS_HASH_MAX_THETA,
426                                              &uuid_hash_ops, CFS_HASH_DEFAULT);
427         if (!obd->obd_uuid_hash)
428                 GOTO(err_hash, err = -ENOMEM);
429
430         /* create a nid-export lustre hash */
431         obd->obd_nid_hash = cfs_hash_create("NID_HASH",
432                                             HASH_NID_CUR_BITS,
433                                             HASH_NID_MAX_BITS,
434                                             HASH_NID_BKT_BITS, 0,
435                                             CFS_HASH_MIN_THETA,
436                                             CFS_HASH_MAX_THETA,
437                                             &nid_hash_ops, CFS_HASH_DEFAULT);
438         if (!obd->obd_nid_hash)
439                 GOTO(err_hash, err = -ENOMEM);
440
441         /* create a nid-stats lustre hash */
442         obd->obd_nid_stats_hash = cfs_hash_create("NID_STATS",
443                                                   HASH_NID_STATS_CUR_BITS,
444                                                   HASH_NID_STATS_MAX_BITS,
445                                                   HASH_NID_STATS_BKT_BITS, 0,
446                                                   CFS_HASH_MIN_THETA,
447                                                   CFS_HASH_MAX_THETA,
448                                                   &nid_stat_hash_ops, CFS_HASH_DEFAULT);
449         if (!obd->obd_nid_stats_hash)
450                 GOTO(err_hash, err = -ENOMEM);
451
452         exp = class_new_export(obd, &obd->obd_uuid);
453         if (IS_ERR(exp))
454                 GOTO(err_hash, err = PTR_ERR(exp));
455
456         obd->obd_self_export = exp;
457         cfs_list_del_init(&exp->exp_obd_chain_timed);
458         class_export_put(exp);
459
460         err = obd_setup(obd, lcfg);
461         if (err)
462                 GOTO(err_exp, err);
463
464         obd->obd_set_up = 1;
465
466         cfs_spin_lock(&obd->obd_dev_lock);
467         /* cleanup drops this */
468         class_incref(obd, "setup", obd);
469         cfs_spin_unlock(&obd->obd_dev_lock);
470
471         CDEBUG(D_IOCTL, "finished setup of obd %s (uuid %s)\n",
472                obd->obd_name, obd->obd_uuid.uuid);
473
474         RETURN(0);
475 err_exp:
476         if (obd->obd_self_export) {
477                 class_unlink_export(obd->obd_self_export);
478                 obd->obd_self_export = NULL;
479         }
480 err_hash:
481         if (obd->obd_uuid_hash) {
482                 cfs_hash_putref(obd->obd_uuid_hash);
483                 obd->obd_uuid_hash = NULL;
484         }
485         if (obd->obd_nid_hash) {
486                 cfs_hash_putref(obd->obd_nid_hash);
487                 obd->obd_nid_hash = NULL;
488         }
489         if (obd->obd_nid_stats_hash) {
490                 cfs_hash_putref(obd->obd_nid_stats_hash);
491                 obd->obd_nid_stats_hash = NULL;
492         }
493         obd->obd_starting = 0;
494         CERROR("setup %s failed (%d)\n", obd->obd_name, err);
495         return err;
496 }
497
498 /** We have finished using this obd and are ready to destroy it.
499  * There can be no more references to this obd.
500  */
501 int class_detach(struct obd_device *obd, struct lustre_cfg *lcfg)
502 {
503         ENTRY;
504
505         if (obd->obd_set_up) {
506                 CERROR("OBD device %d still set up\n", obd->obd_minor);
507                 RETURN(-EBUSY);
508         }
509
510         cfs_spin_lock(&obd->obd_dev_lock);
511         if (!obd->obd_attached) {
512                 cfs_spin_unlock(&obd->obd_dev_lock);
513                 CERROR("OBD device %d not attached\n", obd->obd_minor);
514                 RETURN(-ENODEV);
515         }
516         obd->obd_attached = 0;
517         cfs_spin_unlock(&obd->obd_dev_lock);
518
519         CDEBUG(D_IOCTL, "detach on obd %s (uuid %s)\n",
520                obd->obd_name, obd->obd_uuid.uuid);
521
522         class_decref(obd, "attach", obd);
523
524         /* not strictly necessary, but cleans up eagerly */
525         obd_zombie_impexp_cull();
526
527         RETURN(0);
528 }
529
530 /** Start shutting down the obd.  There may be in-progess ops when
531  * this is called.  We tell them to start shutting down with a call
532  * to class_disconnect_exports().
533  */
534 int class_cleanup(struct obd_device *obd, struct lustre_cfg *lcfg)
535 {
536         int err = 0;
537         char *flag;
538         ENTRY;
539
540         OBD_RACE(OBD_FAIL_LDLM_RECOV_CLIENTS);
541
542         if (!obd->obd_set_up) {
543                 CERROR("Device %d not setup\n", obd->obd_minor);
544                 RETURN(-ENODEV);
545         }
546
547         cfs_spin_lock(&obd->obd_dev_lock);
548         if (obd->obd_stopping) {
549                 cfs_spin_unlock(&obd->obd_dev_lock);
550                 CERROR("OBD %d already stopping\n", obd->obd_minor);
551                 RETURN(-ENODEV);
552         }
553         /* Leave this on forever */
554         obd->obd_stopping = 1;
555         cfs_spin_unlock(&obd->obd_dev_lock);
556
557         if (lcfg->lcfg_bufcount >= 2 && LUSTRE_CFG_BUFLEN(lcfg, 1) > 0) {
558                 for (flag = lustre_cfg_string(lcfg, 1); *flag != 0; flag++)
559                         switch (*flag) {
560                         case 'F':
561                                 obd->obd_force = 1;
562                                 break;
563                         case 'A':
564                                 LCONSOLE_WARN("Failing over %s\n",
565                                               obd->obd_name);
566                                 obd->obd_fail = 1;
567                                 obd->obd_no_transno = 1;
568                                 obd->obd_no_recov = 1;
569                                 if (OBP(obd, iocontrol)) {
570                                         obd_iocontrol(OBD_IOC_SYNC,
571                                                       obd->obd_self_export,
572                                                       0, NULL, NULL);
573                                 }
574                                 break;
575                         default:
576                                 CERROR("Unrecognised flag '%c'\n", *flag);
577                         }
578         }
579
580         LASSERT(obd->obd_self_export);
581
582         /* The three references that should be remaining are the
583          * obd_self_export and the attach and setup references. */
584         if (cfs_atomic_read(&obd->obd_refcount) > 3) {
585                 /* refcounf - 3 might be the number of real exports
586                    (excluding self export). But class_incref is called
587                    by other things as well, so don't count on it. */
588                 CDEBUG(D_IOCTL, "%s: forcing exports to disconnect: %d\n",
589                        obd->obd_name, cfs_atomic_read(&obd->obd_refcount) - 3);
590                 dump_exports(obd, 0);
591                 class_disconnect_exports(obd);
592         }
593
594         /* destroy an uuid-export hash body */
595         if (obd->obd_uuid_hash) {
596                 cfs_hash_putref(obd->obd_uuid_hash);
597                 obd->obd_uuid_hash = NULL;
598         }
599
600         /* destroy a nid-export hash body */
601         if (obd->obd_nid_hash) {
602                 cfs_hash_putref(obd->obd_nid_hash);
603                 obd->obd_nid_hash = NULL;
604         }
605
606         /* destroy a nid-stats hash body */
607         if (obd->obd_nid_stats_hash) {
608                 cfs_hash_putref(obd->obd_nid_stats_hash);
609                 obd->obd_nid_stats_hash = NULL;
610         }
611
612         /* Precleanup, we must make sure all exports get destroyed. */
613         err = obd_precleanup(obd, OBD_CLEANUP_EXPORTS);
614         if (err)
615                 CERROR("Precleanup %s returned %d\n",
616                        obd->obd_name, err);
617         class_decref(obd, "setup", obd);
618         obd->obd_set_up = 0;
619         RETURN(0);
620 }
621
622 struct obd_device *class_incref(struct obd_device *obd,
623                                 const char *scope, const void *source)
624 {
625         lu_ref_add_atomic(&obd->obd_reference, scope, source);
626         cfs_atomic_inc(&obd->obd_refcount);
627         CDEBUG(D_INFO, "incref %s (%p) now %d\n", obd->obd_name, obd,
628                cfs_atomic_read(&obd->obd_refcount));
629
630         return obd;
631 }
632
633 void class_decref(struct obd_device *obd, const char *scope, const void *source)
634 {
635         int err;
636         int refs;
637
638         refs = cfs_atomic_dec_return(&obd->obd_refcount);
639         lu_ref_del(&obd->obd_reference, scope, source);
640
641         CDEBUG(D_INFO, "Decref %s (%p) now %d\n", obd->obd_name, obd, refs);
642
643         if ((refs == 1) && obd->obd_stopping) {
644                 /* All exports have been destroyed; there should
645                    be no more in-progress ops by this point.*/
646
647                 cfs_spin_lock(&obd->obd_self_export->exp_lock);
648                 obd->obd_self_export->exp_flags |= exp_flags_from_obd(obd);
649                 cfs_spin_unlock(&obd->obd_self_export->exp_lock);
650
651                 /* note that we'll recurse into class_decref again */
652                 class_unlink_export(obd->obd_self_export);
653                 return;
654         }
655
656         if (refs == 0) {
657                 CDEBUG(D_CONFIG, "finishing cleanup of obd %s (%s)\n",
658                        obd->obd_name, obd->obd_uuid.uuid);
659                 LASSERT(!obd->obd_attached);
660                 if (obd->obd_stopping) {
661                         /* If we're not stopping, we were never set up */
662                         err = obd_cleanup(obd);
663                         if (err)
664                                 CERROR("Cleanup %s returned %d\n",
665                                        obd->obd_name, err);
666                 }
667                 if (OBP(obd, detach)) {
668                         err = OBP(obd, detach)(obd);
669                         if (err)
670                                 CERROR("Detach returned %d\n", err);
671                 }
672                 class_release_dev(obd);
673         }
674 }
675
676 /** Add a failover nid location.
677  * Client obd types contact server obd types using this nid list.
678  */
679 int class_add_conn(struct obd_device *obd, struct lustre_cfg *lcfg)
680 {
681         struct obd_import *imp;
682         struct obd_uuid uuid;
683         int rc;
684         ENTRY;
685
686         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1 ||
687             LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(struct obd_uuid)) {
688                 CERROR("invalid conn_uuid\n");
689                 RETURN(-EINVAL);
690         }
691         if (strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) &&
692             strcmp(obd->obd_type->typ_name, LUSTRE_OSC_NAME) &&
693             strcmp(obd->obd_type->typ_name, LUSTRE_MGC_NAME)) {
694                 CERROR("can't add connection on non-client dev\n");
695                 RETURN(-EINVAL);
696         }
697
698         imp = obd->u.cli.cl_import;
699         if (!imp) {
700                 CERROR("try to add conn on immature client dev\n");
701                 RETURN(-EINVAL);
702         }
703
704         obd_str2uuid(&uuid, lustre_cfg_string(lcfg, 1));
705         rc = obd_add_conn(imp, &uuid, lcfg->lcfg_num);
706
707         RETURN(rc);
708 }
709
710 /** Remove a failover nid location.
711  */
712 int class_del_conn(struct obd_device *obd, struct lustre_cfg *lcfg)
713 {
714         struct obd_import *imp;
715         struct obd_uuid uuid;
716         int rc;
717         ENTRY;
718
719         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1 ||
720             LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(struct obd_uuid)) {
721                 CERROR("invalid conn_uuid\n");
722                 RETURN(-EINVAL);
723         }
724         if (strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) &&
725             strcmp(obd->obd_type->typ_name, LUSTRE_OSC_NAME)) {
726                 CERROR("can't del connection on non-client dev\n");
727                 RETURN(-EINVAL);
728         }
729
730         imp = obd->u.cli.cl_import;
731         if (!imp) {
732                 CERROR("try to del conn on immature client dev\n");
733                 RETURN(-EINVAL);
734         }
735
736         obd_str2uuid(&uuid, lustre_cfg_string(lcfg, 1));
737         rc = obd_del_conn(imp, &uuid);
738
739         RETURN(rc);
740 }
741
742 CFS_LIST_HEAD(lustre_profile_list);
743
744 struct lustre_profile *class_get_profile(const char * prof)
745 {
746         struct lustre_profile *lprof;
747
748         ENTRY;
749         cfs_list_for_each_entry(lprof, &lustre_profile_list, lp_list) {
750                 if (!strcmp(lprof->lp_profile, prof)) {
751                         RETURN(lprof);
752                 }
753         }
754         RETURN(NULL);
755 }
756
757 /** Create a named "profile".
758  * This defines the mdc and osc names to use for a client.
759  * This also is used to define the lov to be used by a mdt.
760  */
761 int class_add_profile(int proflen, char *prof, int osclen, char *osc,
762                       int mdclen, char *mdc)
763 {
764         struct lustre_profile *lprof;
765         int err = 0;
766         ENTRY;
767
768         CDEBUG(D_CONFIG, "Add profile %s\n", prof);
769
770         OBD_ALLOC(lprof, sizeof(*lprof));
771         if (lprof == NULL)
772                 RETURN(-ENOMEM);
773         CFS_INIT_LIST_HEAD(&lprof->lp_list);
774
775         LASSERT(proflen == (strlen(prof) + 1));
776         OBD_ALLOC(lprof->lp_profile, proflen);
777         if (lprof->lp_profile == NULL)
778                 GOTO(out, err = -ENOMEM);
779         memcpy(lprof->lp_profile, prof, proflen);
780
781         LASSERT(osclen == (strlen(osc) + 1));
782         OBD_ALLOC(lprof->lp_dt, osclen);
783         if (lprof->lp_dt == NULL)
784                 GOTO(out, err = -ENOMEM);
785         memcpy(lprof->lp_dt, osc, osclen);
786
787         if (mdclen > 0) {
788                 LASSERT(mdclen == (strlen(mdc) + 1));
789                 OBD_ALLOC(lprof->lp_md, mdclen);
790                 if (lprof->lp_md == NULL)
791                         GOTO(out, err = -ENOMEM);
792                 memcpy(lprof->lp_md, mdc, mdclen);
793         }
794
795         cfs_list_add(&lprof->lp_list, &lustre_profile_list);
796         RETURN(err);
797
798 out:
799         if (lprof->lp_md)
800                 OBD_FREE(lprof->lp_md, mdclen);
801         if (lprof->lp_dt)
802                 OBD_FREE(lprof->lp_dt, osclen);
803         if (lprof->lp_profile)
804                 OBD_FREE(lprof->lp_profile, proflen);
805         OBD_FREE(lprof, sizeof(*lprof));
806         RETURN(err);
807 }
808
809 void class_del_profile(const char *prof)
810 {
811         struct lustre_profile *lprof;
812         ENTRY;
813
814         CDEBUG(D_CONFIG, "Del profile %s\n", prof);
815
816         lprof = class_get_profile(prof);
817         if (lprof) {
818                 cfs_list_del(&lprof->lp_list);
819                 OBD_FREE(lprof->lp_profile, strlen(lprof->lp_profile) + 1);
820                 OBD_FREE(lprof->lp_dt, strlen(lprof->lp_dt) + 1);
821                 if (lprof->lp_md)
822                         OBD_FREE(lprof->lp_md, strlen(lprof->lp_md) + 1);
823                 OBD_FREE(lprof, sizeof *lprof);
824         }
825         EXIT;
826 }
827
828 /* COMPAT_146 */
829 void class_del_profiles(void)
830 {
831         struct lustre_profile *lprof, *n;
832         ENTRY;
833
834         cfs_list_for_each_entry_safe(lprof, n, &lustre_profile_list, lp_list) {
835                 cfs_list_del(&lprof->lp_list);
836                 OBD_FREE(lprof->lp_profile, strlen(lprof->lp_profile) + 1);
837                 OBD_FREE(lprof->lp_dt, strlen(lprof->lp_dt) + 1);
838                 if (lprof->lp_md)
839                         OBD_FREE(lprof->lp_md, strlen(lprof->lp_md) + 1);
840                 OBD_FREE(lprof, sizeof *lprof);
841         }
842         EXIT;
843 }
844
845 static int class_set_global(char *ptr, int val) {
846         ENTRY;
847
848         if (class_match_param(ptr, PARAM_AT_MIN, NULL) == 0)
849             at_min = val;
850         else if (class_match_param(ptr, PARAM_AT_MAX, NULL) == 0)
851                 at_max = val;
852         else if (class_match_param(ptr, PARAM_AT_EXTRA, NULL) == 0)
853                 at_extra = val;
854         else if (class_match_param(ptr, PARAM_AT_EARLY_MARGIN, NULL) == 0)
855                 at_early_margin = val;
856         else if (class_match_param(ptr, PARAM_AT_HISTORY, NULL) == 0)
857                 at_history = val;
858         else
859                 RETURN(-EINVAL);
860
861         CDEBUG(D_IOCTL, "global %s = %d\n", ptr, val);
862
863         RETURN(0);
864 }
865
866
867 /* We can't call ll_process_config directly because it lives in a module that
868    must be loaded after this one. */
869 static int (*client_process_config)(struct lustre_cfg *lcfg) = NULL;
870
871 void lustre_register_client_process_config(int (*cpc)(struct lustre_cfg *lcfg))
872 {
873         client_process_config = cpc;
874 }
875 EXPORT_SYMBOL(lustre_register_client_process_config);
876
877 /** Process configuration commands given in lustre_cfg form.
878  * These may come from direct calls (e.g. class_manual_cleanup)
879  * or processing the config llog, or ioctl from lctl.
880  */
881 int class_process_config(struct lustre_cfg *lcfg)
882 {
883         struct obd_device *obd;
884         int err;
885
886         LASSERT(lcfg && !IS_ERR(lcfg));
887         CDEBUG(D_IOCTL, "processing cmd: %x\n", lcfg->lcfg_command);
888
889         /* Commands that don't need a device */
890         switch(lcfg->lcfg_command) {
891         case LCFG_ATTACH: {
892                 err = class_attach(lcfg);
893                 GOTO(out, err);
894         }
895         case LCFG_ADD_UUID: {
896                 CDEBUG(D_IOCTL, "adding mapping from uuid %s to nid "LPX64
897                        " (%s)\n", lustre_cfg_string(lcfg, 1),
898                        lcfg->lcfg_nid, libcfs_nid2str(lcfg->lcfg_nid));
899
900                 err = class_add_uuid(lustre_cfg_string(lcfg, 1), lcfg->lcfg_nid);
901                 GOTO(out, err);
902         }
903         case LCFG_DEL_UUID: {
904                 CDEBUG(D_IOCTL, "removing mappings for uuid %s\n",
905                        (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) == 0)
906                        ? "<all uuids>" : lustre_cfg_string(lcfg, 1));
907
908                 err = class_del_uuid(lustre_cfg_string(lcfg, 1));
909                 GOTO(out, err);
910         }
911         case LCFG_MOUNTOPT: {
912                 CDEBUG(D_IOCTL, "mountopt: profile %s osc %s mdc %s\n",
913                        lustre_cfg_string(lcfg, 1),
914                        lustre_cfg_string(lcfg, 2),
915                        lustre_cfg_string(lcfg, 3));
916                 /* set these mount options somewhere, so ll_fill_super
917                  * can find them. */
918                 err = class_add_profile(LUSTRE_CFG_BUFLEN(lcfg, 1),
919                                         lustre_cfg_string(lcfg, 1),
920                                         LUSTRE_CFG_BUFLEN(lcfg, 2),
921                                         lustre_cfg_string(lcfg, 2),
922                                         LUSTRE_CFG_BUFLEN(lcfg, 3),
923                                         lustre_cfg_string(lcfg, 3));
924                 GOTO(out, err);
925         }
926         case LCFG_DEL_MOUNTOPT: {
927                 CDEBUG(D_IOCTL, "mountopt: profile %s\n",
928                        lustre_cfg_string(lcfg, 1));
929                 class_del_profile(lustre_cfg_string(lcfg, 1));
930                 GOTO(out, err = 0);
931         }
932         case LCFG_SET_TIMEOUT: {
933                 CDEBUG(D_IOCTL, "changing lustre timeout from %d to %d\n",
934                        obd_timeout, lcfg->lcfg_num);
935                 obd_timeout = max(lcfg->lcfg_num, 1U);
936                 GOTO(out, err = 0);
937         }
938         case LCFG_SET_LDLM_TIMEOUT: {
939                 CDEBUG(D_IOCTL, "changing lustre ldlm_timeout from %d to %d\n",
940                        ldlm_timeout, lcfg->lcfg_num);
941                 ldlm_timeout = max(lcfg->lcfg_num, 1U);
942                 if (ldlm_timeout >= obd_timeout)
943                         ldlm_timeout = max(obd_timeout / 3, 1U);
944
945                 GOTO(out, err = 0);
946         }
947         case LCFG_SET_UPCALL: {
948                 LCONSOLE_ERROR_MSG(0x15a, "recovery upcall is deprecated\n");
949                 /* COMPAT_146 Don't fail on old configs */
950                 GOTO(out, err = 0);
951         }
952         case LCFG_MARKER: {
953                 struct cfg_marker *marker;
954                 marker = lustre_cfg_buf(lcfg, 1);
955                 CDEBUG(D_IOCTL, "marker %d (%#x) %.16s %s\n", marker->cm_step,
956                        marker->cm_flags, marker->cm_tgtname, marker->cm_comment);
957                 GOTO(out, err = 0);
958         }
959         case LCFG_PARAM: {
960                 char *tmp;
961                 /* llite has no obd */
962                 if ((class_match_param(lustre_cfg_string(lcfg, 1),
963                                        PARAM_LLITE, 0) == 0) &&
964                     client_process_config) {
965                         err = (*client_process_config)(lcfg);
966                         GOTO(out, err);
967                 } else if ((class_match_param(lustre_cfg_string(lcfg, 1),
968                                               PARAM_SYS, &tmp) == 0)) {
969                         /* Global param settings */
970                         err = class_set_global(tmp, lcfg->lcfg_num);
971                         /* Note that since LCFG_PARAM is LCFG_REQUIRED, new
972                            unknown globals would cause config to fail */
973                         if (err)
974                                 CWARN("Ignoring unknown param %s\n", tmp);
975                         GOTO(out, 0);
976                 }
977
978                 /* Fall through */
979                 break;
980         }
981         }
982
983         /* Commands that require a device */
984         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
985         if (obd == NULL) {
986                 if (!LUSTRE_CFG_BUFLEN(lcfg, 0))
987                         CERROR("this lcfg command requires a device name\n");
988                 else
989                         CERROR("no device for: %s\n",
990                                lustre_cfg_string(lcfg, 0));
991
992                 GOTO(out, err = -EINVAL);
993         }
994
995         switch(lcfg->lcfg_command) {
996         case LCFG_SETUP: {
997                 err = class_setup(obd, lcfg);
998                 GOTO(out, err);
999         }
1000         case LCFG_DETACH: {
1001                 err = class_detach(obd, lcfg);
1002                 GOTO(out, err = 0);
1003         }
1004         case LCFG_CLEANUP: {
1005                 err = class_cleanup(obd, lcfg);
1006                 GOTO(out, err = 0);
1007         }
1008         case LCFG_ADD_CONN: {
1009                 err = class_add_conn(obd, lcfg);
1010                 GOTO(out, err = 0);
1011         }
1012         case LCFG_DEL_CONN: {
1013                 err = class_del_conn(obd, lcfg);
1014                 GOTO(out, err = 0);
1015         }
1016         case LCFG_POOL_NEW: {
1017                 err = obd_pool_new(obd, lustre_cfg_string(lcfg, 2));
1018                 GOTO(out, err = 0);
1019                 break;
1020         }
1021         case LCFG_POOL_ADD: {
1022                 err = obd_pool_add(obd, lustre_cfg_string(lcfg, 2),
1023                                    lustre_cfg_string(lcfg, 3));
1024                 GOTO(out, err = 0);
1025                 break;
1026         }
1027         case LCFG_POOL_REM: {
1028                 err = obd_pool_rem(obd, lustre_cfg_string(lcfg, 2),
1029                                    lustre_cfg_string(lcfg, 3));
1030                 GOTO(out, err = 0);
1031                 break;
1032         }
1033         case LCFG_POOL_DEL: {
1034                 err = obd_pool_del(obd, lustre_cfg_string(lcfg, 2));
1035                 GOTO(out, err = 0);
1036                 break;
1037         }
1038         default: {
1039                 err = obd_process_config(obd, sizeof(*lcfg), lcfg);
1040                 GOTO(out, err);
1041
1042         }
1043         }
1044 out:
1045         if ((err < 0) && !(lcfg->lcfg_command & LCFG_REQUIRED)) {
1046                 CWARN("Ignoring error %d on optional command %#x\n", err,
1047                       lcfg->lcfg_command);
1048                 err = 0;
1049         }
1050         return err;
1051 }
1052
1053 int class_process_proc_param(char *prefix, struct lprocfs_vars *lvars,
1054                              struct lustre_cfg *lcfg, void *data)
1055 {
1056 #ifdef __KERNEL__
1057         struct lprocfs_vars *var;
1058         char *key, *sval;
1059         int i, keylen, vallen;
1060         int matched = 0, j = 0;
1061         int rc = 0;
1062         int skip = 0;
1063         ENTRY;
1064
1065         if (lcfg->lcfg_command != LCFG_PARAM) {
1066                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1067                 RETURN(-EINVAL);
1068         }
1069
1070         /* e.g. tunefs.lustre --param mdt.group_upcall=foo /r/tmp/lustre-mdt
1071            or   lctl conf_param lustre-MDT0000.mdt.group_upcall=bar
1072            or   lctl conf_param lustre-OST0000.osc.max_dirty_mb=36 */
1073         for (i = 1; i < lcfg->lcfg_bufcount; i++) {
1074                 key = lustre_cfg_buf(lcfg, i);
1075                 /* Strip off prefix */
1076                 class_match_param(key, prefix, &key);
1077                 sval = strchr(key, '=');
1078                 if (!sval || (*(sval + 1) == 0)) {
1079                         CERROR("Can't parse param %s (missing '=')\n", key);
1080                         /* rc = -EINVAL;        continue parsing other params */
1081                         continue;
1082                 }
1083                 keylen = sval - key;
1084                 sval++;
1085                 vallen = strlen(sval);
1086                 matched = 0;
1087                 j = 0;
1088                 /* Search proc entries */
1089                 while (lvars[j].name) {
1090                         var = &lvars[j];
1091                         if (class_match_param(key, (char *)var->name, 0) == 0 &&
1092                             keylen == strlen(var->name)) {
1093                                 matched++;
1094                                 rc = -EROFS;
1095                                 if (var->write_fptr) {
1096                                         mm_segment_t oldfs;
1097                                         oldfs = get_fs();
1098                                         set_fs(KERNEL_DS);
1099                                         rc = (var->write_fptr)(NULL, sval,
1100                                                                vallen, data);
1101                                         set_fs(oldfs);
1102                                 }
1103                                 break;
1104                         }
1105                         j++;
1106                 }
1107                 if (!matched) {
1108                         /* If the prefix doesn't match, return error so we
1109                            can pass it down the stack */
1110                         if (strnchr(key, keylen, '.'))
1111                             RETURN(-ENOSYS);
1112                         CERROR("%s: unknown param %s\n",
1113                                (char *)lustre_cfg_string(lcfg, 0), key);
1114                         /* rc = -EINVAL;        continue parsing other params */
1115                         skip++;
1116                 } else if (rc < 0) {
1117                         CERROR("writing proc entry %s err %d\n",
1118                                var->name, rc);
1119                         rc = 0;
1120                 } else {
1121                         LCONSOLE_INFO("%s.%.*s: set parameter %.*s=%s\n",
1122                                       lustre_cfg_string(lcfg, 0),
1123                                       (int)strlen(prefix) - 1, prefix,
1124                                       (int)(sval - key - 1), key, sval);
1125                 }
1126         }
1127
1128         if (rc > 0)
1129                 rc = 0;
1130         if (!rc && skip)
1131                 rc = skip;
1132         RETURN(rc);
1133 #else
1134         CDEBUG(D_CONFIG, "liblustre can't process params.\n");
1135         /* Don't throw config error */
1136         RETURN(0);
1137 #endif
1138 }
1139
1140 int class_config_dump_handler(struct llog_handle * handle,
1141                               struct llog_rec_hdr *rec, void *data);
1142
1143 #ifdef __KERNEL__
1144 extern int lustre_check_exclusion(struct super_block *sb, char *svname);
1145 #else
1146 #define lustre_check_exclusion(a,b)  0
1147 #endif
1148
1149 /** Parse a configuration llog, doing various manipulations on them
1150  * for various reasons, (modifications for compatibility, skip obsolete
1151  * records, change uuids, etc), then class_process_config() resulting
1152  * net records.
1153  */
1154 static int class_config_llog_handler(struct llog_handle * handle,
1155                                      struct llog_rec_hdr *rec, void *data)
1156 {
1157         struct config_llog_instance *clli = data;
1158         int cfg_len = rec->lrh_len;
1159         char *cfg_buf = (char*) (rec + 1);
1160         int rc = 0;
1161         ENTRY;
1162
1163         //class_config_dump_handler(handle, rec, data);
1164
1165         switch (rec->lrh_type) {
1166         case OBD_CFG_REC: {
1167                 struct lustre_cfg *lcfg, *lcfg_new;
1168                 struct lustre_cfg_bufs bufs;
1169                 char *inst_name = NULL;
1170                 int inst_len = 0;
1171                 int inst = 0, swab = 0;
1172
1173                 lcfg = (struct lustre_cfg *)cfg_buf;
1174                 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
1175                         lustre_swab_lustre_cfg(lcfg);
1176                         swab = 1;
1177                 }
1178
1179                 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
1180                 if (rc)
1181                         GOTO(out, rc);
1182
1183                 /* Figure out config state info */
1184                 if (lcfg->lcfg_command == LCFG_MARKER) {
1185                         struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
1186                         lustre_swab_cfg_marker(marker, swab,
1187                                                LUSTRE_CFG_BUFLEN(lcfg, 1));
1188                         CDEBUG(D_CONFIG, "Marker, inst_flg=%#x mark_flg=%#x\n",
1189                                clli->cfg_flags, marker->cm_flags);
1190                         if (marker->cm_flags & CM_START) {
1191                                 /* all previous flags off */
1192                                 clli->cfg_flags = CFG_F_MARKER;
1193                                 if (marker->cm_flags & CM_SKIP) {
1194                                         clli->cfg_flags |= CFG_F_SKIP;
1195                                         CDEBUG(D_CONFIG, "SKIP #%d\n",
1196                                                marker->cm_step);
1197                                 } else if ((marker->cm_flags & CM_EXCLUDE) ||
1198                                            (clli->cfg_sb &&
1199                                             lustre_check_exclusion(clli->cfg_sb,
1200                                                          marker->cm_tgtname))) {
1201                                         clli->cfg_flags |= CFG_F_EXCLUDE;
1202                                         CDEBUG(D_CONFIG, "EXCLUDE %d\n",
1203                                                marker->cm_step);
1204                                 }
1205                         } else if (marker->cm_flags & CM_END) {
1206                                 clli->cfg_flags = 0;
1207                         }
1208                 }
1209                 /* A config command without a start marker before it is
1210                    illegal (post 146) */
1211                 if (!(clli->cfg_flags & CFG_F_COMPAT146) &&
1212                     !(clli->cfg_flags & CFG_F_MARKER) &&
1213                     (lcfg->lcfg_command != LCFG_MARKER)) {
1214                         CWARN("Config not inside markers, ignoring! "
1215                               "(inst: %s, uuid: %s, flags: %#x)\n",
1216                               clli->cfg_instance ? clli->cfg_instance : "<null>",
1217                               clli->cfg_uuid.uuid, clli->cfg_flags);
1218                         clli->cfg_flags |= CFG_F_SKIP;
1219                 }
1220                 if (clli->cfg_flags & CFG_F_SKIP) {
1221                         CDEBUG(D_CONFIG, "skipping %#x\n",
1222                                clli->cfg_flags);
1223                         rc = 0;
1224                         /* No processing! */
1225                         break;
1226                 }
1227
1228                 /*
1229                  * For interoperability between 1.8 and 2.0,
1230                  * rename "mds" obd device type to "mdt".
1231                  */
1232                 {
1233                         char *typename = lustre_cfg_string(lcfg, 1);
1234                         char *index = lustre_cfg_string(lcfg, 2);
1235
1236                         if ((lcfg->lcfg_command == LCFG_ATTACH && typename &&
1237                              strcmp(typename, "mds") == 0)) {
1238                                 CWARN("For 1.8 interoperability, rename obd "
1239                                        "type from mds to mdt\n");
1240                                 typename[2] = 't';
1241                         }
1242                         if ((lcfg->lcfg_command == LCFG_SETUP && index &&
1243                              strcmp(index, "type") == 0)) {
1244                                 CDEBUG(D_INFO, "For 1.8 interoperability, "
1245                                        "set this index to '0'\n");
1246                                 index[0] = '0';
1247                                 index[1] = 0;
1248                         }
1249                 }
1250
1251                 if ((clli->cfg_flags & CFG_F_EXCLUDE) &&
1252                     (lcfg->lcfg_command == LCFG_LOV_ADD_OBD))
1253                         /* Add inactive instead */
1254                         lcfg->lcfg_command = LCFG_LOV_ADD_INA;
1255
1256                 lustre_cfg_bufs_init(&bufs, lcfg);
1257
1258                 if (clli && clli->cfg_instance &&
1259                     LUSTRE_CFG_BUFLEN(lcfg, 0) > 0){
1260                         inst = 1;
1261                         inst_len = LUSTRE_CFG_BUFLEN(lcfg, 0) +
1262                                 strlen(clli->cfg_instance) + 1;
1263                         OBD_ALLOC(inst_name, inst_len);
1264                         if (inst_name == NULL)
1265                                 GOTO(out, rc = -ENOMEM);
1266                         sprintf(inst_name, "%s-%s",
1267                                 lustre_cfg_string(lcfg, 0),
1268                                 clli->cfg_instance);
1269                         lustre_cfg_bufs_set_string(&bufs, 0, inst_name);
1270                         CDEBUG(D_CONFIG, "cmd %x, instance name: %s\n",
1271                                lcfg->lcfg_command, inst_name);
1272                 }
1273
1274                 /* we override the llog's uuid for clients, to insure they
1275                 are unique */
1276                 if (clli && clli->cfg_instance &&
1277                     lcfg->lcfg_command == LCFG_ATTACH) {
1278                         lustre_cfg_bufs_set_string(&bufs, 2,
1279                                                    clli->cfg_uuid.uuid);
1280                 }
1281                 /*
1282                  * sptlrpc config record, we expect 2 data segments:
1283                  *  [0]: fs_name/target_name,
1284                  *  [1]: rule string
1285                  * moving them to index [1] and [2], and insert MGC's
1286                  * obdname at index [0].
1287                  */
1288                 if (clli && clli->cfg_instance == NULL &&
1289                     lcfg->lcfg_command == LCFG_SPTLRPC_CONF) {
1290                         lustre_cfg_bufs_set(&bufs, 2, bufs.lcfg_buf[1],
1291                                             bufs.lcfg_buflen[1]);
1292                         lustre_cfg_bufs_set(&bufs, 1, bufs.lcfg_buf[0],
1293                                             bufs.lcfg_buflen[0]);
1294                         lustre_cfg_bufs_set_string(&bufs, 0,
1295                                                    clli->cfg_obdname);
1296                 }
1297
1298                 lcfg_new = lustre_cfg_new(lcfg->lcfg_command, &bufs);
1299
1300                 lcfg_new->lcfg_num   = lcfg->lcfg_num;
1301                 lcfg_new->lcfg_flags = lcfg->lcfg_flags;
1302
1303                 /* XXX Hack to try to remain binary compatible with
1304                  * pre-newconfig logs */
1305                 if (lcfg->lcfg_nal != 0 &&      /* pre-newconfig log? */
1306                     (lcfg->lcfg_nid >> 32) == 0) {
1307                         __u32 addr = (__u32)(lcfg->lcfg_nid & 0xffffffff);
1308
1309                         lcfg_new->lcfg_nid =
1310                                 LNET_MKNID(LNET_MKNET(lcfg->lcfg_nal, 0), addr);
1311                         CWARN("Converted pre-newconfig NAL %d NID %x to %s\n",
1312                               lcfg->lcfg_nal, addr,
1313                               libcfs_nid2str(lcfg_new->lcfg_nid));
1314                 } else {
1315                         lcfg_new->lcfg_nid = lcfg->lcfg_nid;
1316                 }
1317
1318                 lcfg_new->lcfg_nal = 0; /* illegal value for obsolete field */
1319
1320                 rc = class_process_config(lcfg_new);
1321                 lustre_cfg_free(lcfg_new);
1322
1323                 if (inst)
1324                         OBD_FREE(inst_name, inst_len);
1325                 break;
1326         }
1327         default:
1328                 CERROR("Unknown llog record type %#x encountered\n",
1329                        rec->lrh_type);
1330                 break;
1331         }
1332 out:
1333         if (rc) {
1334                 CERROR("Err %d on cfg command:\n", rc);
1335                 class_config_dump_handler(handle, rec, data);
1336         }
1337         RETURN(rc);
1338 }
1339
1340 int class_config_parse_llog(struct llog_ctxt *ctxt, char *name,
1341                             struct config_llog_instance *cfg)
1342 {
1343         struct llog_process_cat_data cd = {0, 0};
1344         struct llog_handle *llh;
1345         int rc, rc2;
1346         ENTRY;
1347
1348         CDEBUG(D_INFO, "looking up llog %s\n", name);
1349         rc = llog_create(ctxt, &llh, NULL, name);
1350         if (rc)
1351                 RETURN(rc);
1352
1353         rc = llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
1354         if (rc)
1355                 GOTO(parse_out, rc);
1356
1357         /* continue processing from where we last stopped to end-of-log */
1358         if (cfg)
1359                 cd.lpcd_first_idx = cfg->cfg_last_idx;
1360         cd.lpcd_last_idx = 0;
1361
1362         rc = llog_process(llh, class_config_llog_handler, cfg, &cd);
1363
1364         CDEBUG(D_CONFIG, "Processed log %s gen %d-%d (rc=%d)\n", name,
1365                cd.lpcd_first_idx + 1, cd.lpcd_last_idx, rc);
1366
1367         if (cfg)
1368                 cfg->cfg_last_idx = cd.lpcd_last_idx;
1369
1370 parse_out:
1371         rc2 = llog_close(llh);
1372         if (rc == 0)
1373                 rc = rc2;
1374
1375         RETURN(rc);
1376 }
1377
1378 int class_config_dump_handler(struct llog_handle * handle,
1379                               struct llog_rec_hdr *rec, void *data)
1380 {
1381         int cfg_len = rec->lrh_len;
1382         char *cfg_buf = (char*) (rec + 1);
1383         char *outstr, *ptr, *end;
1384         int rc = 0;
1385         ENTRY;
1386
1387         OBD_ALLOC(outstr, 256);
1388         end = outstr + 256;
1389         ptr = outstr;
1390         if (!outstr) {
1391                 RETURN(-ENOMEM);
1392         }
1393         if (rec->lrh_type == OBD_CFG_REC) {
1394                 struct lustre_cfg *lcfg;
1395                 int i;
1396
1397                 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
1398                 if (rc)
1399                         GOTO(out, rc);
1400                 lcfg = (struct lustre_cfg *)cfg_buf;
1401
1402                 ptr += snprintf(ptr, end-ptr, "cmd=%05x ",
1403                                 lcfg->lcfg_command);
1404                 if (lcfg->lcfg_flags) {
1405                         ptr += snprintf(ptr, end-ptr, "flags=%#08x ",
1406                                         lcfg->lcfg_flags);
1407                 }
1408                 if (lcfg->lcfg_num) {
1409                         ptr += snprintf(ptr, end-ptr, "num=%#08x ",
1410                                         lcfg->lcfg_num);
1411                 }
1412                 if (lcfg->lcfg_nid) {
1413                         ptr += snprintf(ptr, end-ptr, "nid=%s("LPX64")\n     ",
1414                                         libcfs_nid2str(lcfg->lcfg_nid),
1415                                         lcfg->lcfg_nid);
1416                 }
1417                 if (lcfg->lcfg_command == LCFG_MARKER) {
1418                         struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
1419                         ptr += snprintf(ptr, end-ptr, "marker=%d(%#x)%s '%s'",
1420                                         marker->cm_step, marker->cm_flags,
1421                                         marker->cm_tgtname, marker->cm_comment);
1422                 } else {
1423                         for (i = 0; i <  lcfg->lcfg_bufcount; i++) {
1424                                 ptr += snprintf(ptr, end-ptr, "%d:%s  ", i,
1425                                                 lustre_cfg_string(lcfg, i));
1426                         }
1427                 }
1428                 LCONSOLE(D_WARNING, "   %s\n", outstr);
1429         } else {
1430                 LCONSOLE(D_WARNING, "unhandled lrh_type: %#x\n", rec->lrh_type);
1431                 rc = -EINVAL;
1432         }
1433 out:
1434         OBD_FREE(outstr, 256);
1435         RETURN(rc);
1436 }
1437
1438 int class_config_dump_llog(struct llog_ctxt *ctxt, char *name,
1439                            struct config_llog_instance *cfg)
1440 {
1441         struct llog_handle *llh;
1442         int rc, rc2;
1443         ENTRY;
1444
1445         LCONSOLE_INFO("Dumping config log %s\n", name);
1446
1447         rc = llog_create(ctxt, &llh, NULL, name);
1448         if (rc)
1449                 RETURN(rc);
1450
1451         rc = llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
1452         if (rc)
1453                 GOTO(parse_out, rc);
1454
1455         rc = llog_process(llh, class_config_dump_handler, cfg, NULL);
1456 parse_out:
1457         rc2 = llog_close(llh);
1458         if (rc == 0)
1459                 rc = rc2;
1460
1461         LCONSOLE_INFO("End config log %s\n", name);
1462         RETURN(rc);
1463
1464 }
1465
1466 /** Call class_cleanup and class_detach.
1467  * "Manual" only in the sense that we're faking lcfg commands.
1468  */
1469 int class_manual_cleanup(struct obd_device *obd)
1470 {
1471         char                    flags[3] = "";
1472         struct lustre_cfg      *lcfg;
1473         struct lustre_cfg_bufs  bufs;
1474         int                     rc;
1475         ENTRY;
1476
1477         if (!obd) {
1478                 CERROR("empty cleanup\n");
1479                 RETURN(-EALREADY);
1480         }
1481
1482         if (obd->obd_force)
1483                 strcat(flags, "F");
1484         if (obd->obd_fail)
1485                 strcat(flags, "A");
1486
1487         CDEBUG(D_CONFIG, "Manual cleanup of %s (flags='%s')\n",
1488                obd->obd_name, flags);
1489
1490         lustre_cfg_bufs_reset(&bufs, obd->obd_name);
1491         lustre_cfg_bufs_set_string(&bufs, 1, flags);
1492         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
1493         if (!lcfg)
1494                 RETURN(-ENOMEM);
1495
1496         rc = class_process_config(lcfg);
1497         if (rc) {
1498                 CERROR("cleanup failed %d: %s\n", rc, obd->obd_name);
1499                 GOTO(out, rc);
1500         }
1501
1502         /* the lcfg is almost the same for both ops */
1503         lcfg->lcfg_command = LCFG_DETACH;
1504         rc = class_process_config(lcfg);
1505         if (rc)
1506                 CERROR("detach failed %d: %s\n", rc, obd->obd_name);
1507 out:
1508         lustre_cfg_free(lcfg);
1509         RETURN(rc);
1510 }
1511
1512 /*
1513  * uuid<->export lustre hash operations
1514  */
1515
1516 static unsigned
1517 uuid_hash(cfs_hash_t *hs, const void *key, unsigned mask)
1518 {
1519         return cfs_hash_djb2_hash(((struct obd_uuid *)key)->uuid,
1520                                   sizeof(((struct obd_uuid *)key)->uuid), mask);
1521 }
1522
1523 static void *
1524 uuid_key(cfs_hlist_node_t *hnode)
1525 {
1526         struct obd_export *exp;
1527
1528         exp = cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1529
1530         return &exp->exp_client_uuid;
1531 }
1532
1533 /*
1534  * NOTE: It is impossible to find an export that is in failed
1535  *       state with this function
1536  */
1537 static int
1538 uuid_keycmp(const void *key, cfs_hlist_node_t *hnode)
1539 {
1540         struct obd_export *exp;
1541
1542         LASSERT(key);
1543         exp = cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1544
1545         return obd_uuid_equals(key, &exp->exp_client_uuid) &&
1546                !exp->exp_failed;
1547 }
1548
1549 static void *
1550 uuid_export_object(cfs_hlist_node_t *hnode)
1551 {
1552         return cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1553 }
1554
1555 static void
1556 uuid_export_get(cfs_hash_t *hs, 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_get(exp);
1562 }
1563
1564 static void
1565 uuid_export_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
1566 {
1567         struct obd_export *exp;
1568
1569         exp = cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1570         class_export_put(exp);
1571 }
1572
1573 static cfs_hash_ops_t uuid_hash_ops = {
1574         .hs_hash        = uuid_hash,
1575         .hs_key         = uuid_key,
1576         .hs_keycmp      = uuid_keycmp,
1577         .hs_object      = uuid_export_object,
1578         .hs_get         = uuid_export_get,
1579         .hs_put_locked  = uuid_export_put_locked,
1580 };
1581
1582
1583 /*
1584  * nid<->export hash operations
1585  */
1586
1587 static unsigned
1588 nid_hash(cfs_hash_t *hs, const void *key, unsigned mask)
1589 {
1590         return cfs_hash_djb2_hash(key, sizeof(lnet_nid_t), mask);
1591 }
1592
1593 static void *
1594 nid_key(cfs_hlist_node_t *hnode)
1595 {
1596         struct obd_export *exp;
1597
1598         exp = cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1599
1600         RETURN(&exp->exp_connection->c_peer.nid);
1601 }
1602
1603 /*
1604  * NOTE: It is impossible to find an export that is in failed
1605  *       state with this function
1606  */
1607 static int
1608 nid_kepcmp(const void *key, cfs_hlist_node_t *hnode)
1609 {
1610         struct obd_export *exp;
1611
1612         LASSERT(key);
1613         exp = cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1614
1615         RETURN(exp->exp_connection->c_peer.nid == *(lnet_nid_t *)key &&
1616                !exp->exp_failed);
1617 }
1618
1619 static void *
1620 nid_export_object(cfs_hlist_node_t *hnode)
1621 {
1622         return cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1623 }
1624
1625 static void
1626 nid_export_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
1627 {
1628         struct obd_export *exp;
1629
1630         exp = cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1631         class_export_get(exp);
1632 }
1633
1634 static void
1635 nid_export_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
1636 {
1637         struct obd_export *exp;
1638
1639         exp = cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1640         class_export_put(exp);
1641 }
1642
1643 static cfs_hash_ops_t nid_hash_ops = {
1644         .hs_hash        = nid_hash,
1645         .hs_key         = nid_key,
1646         .hs_keycmp      = nid_kepcmp,
1647         .hs_object      = nid_export_object,
1648         .hs_get         = nid_export_get,
1649         .hs_put_locked  = nid_export_put_locked,
1650 };
1651
1652
1653 /*
1654  * nid<->nidstats hash operations
1655  */
1656
1657 static void *
1658 nidstats_key(cfs_hlist_node_t *hnode)
1659 {
1660         struct nid_stat *ns;
1661
1662         ns = cfs_hlist_entry(hnode, struct nid_stat, nid_hash);
1663
1664         return &ns->nid;
1665 }
1666
1667 static int
1668 nidstats_keycmp(const void *key, cfs_hlist_node_t *hnode)
1669 {
1670         return *(lnet_nid_t *)nidstats_key(hnode) == *(lnet_nid_t *)key;
1671 }
1672
1673 static void *
1674 nidstats_object(cfs_hlist_node_t *hnode)
1675 {
1676         return cfs_hlist_entry(hnode, struct nid_stat, nid_hash);
1677 }
1678
1679 static void
1680 nidstats_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
1681 {
1682         struct nid_stat *ns;
1683
1684         ns = cfs_hlist_entry(hnode, struct nid_stat, nid_hash);
1685         nidstat_getref(ns);
1686 }
1687
1688 static void
1689 nidstats_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
1690 {
1691         struct nid_stat *ns;
1692
1693         ns = cfs_hlist_entry(hnode, struct nid_stat, nid_hash);
1694         nidstat_putref(ns);
1695 }
1696
1697 static cfs_hash_ops_t nid_stat_hash_ops = {
1698         .hs_hash        = nid_hash,
1699         .hs_key         = nidstats_key,
1700         .hs_keycmp      = nidstats_keycmp,
1701         .hs_object      = nidstats_object,
1702         .hs_get         = nidstats_get,
1703         .hs_put_locked  = nidstats_put_locked,
1704 };