• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2012, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/mgc/mgc_request.c
37  *
38  * Author: Nathan Rutman <nathan@clusterfs.com>
39  */
40 
41 #define DEBUG_SUBSYSTEM S_MGC
42 #define D_MGC D_CONFIG /*|D_WARNING*/
43 
44 #include <linux/module.h>
45 #include "../include/obd_class.h"
46 #include "../include/lustre_dlm.h"
47 #include "../include/lprocfs_status.h"
48 #include "../include/lustre_log.h"
49 #include "../include/lustre_disk.h"
50 
51 #include "mgc_internal.h"
52 
mgc_name2resid(char * name,int len,struct ldlm_res_id * res_id,int type)53 static int mgc_name2resid(char *name, int len, struct ldlm_res_id *res_id,
54 			  int type)
55 {
56 	__u64 resname = 0;
57 
58 	if (len > sizeof(resname)) {
59 		CERROR("name too long: %s\n", name);
60 		return -EINVAL;
61 	}
62 	if (len <= 0) {
63 		CERROR("missing name: %s\n", name);
64 		return -EINVAL;
65 	}
66 	memcpy(&resname, name, len);
67 
68 	/* Always use the same endianness for the resid */
69 	memset(res_id, 0, sizeof(*res_id));
70 	res_id->name[0] = cpu_to_le64(resname);
71 	/* XXX: unfortunately, sptlprc and config llog share one lock */
72 	switch (type) {
73 	case CONFIG_T_CONFIG:
74 	case CONFIG_T_SPTLRPC:
75 		resname = 0;
76 		break;
77 	case CONFIG_T_RECOVER:
78 	case CONFIG_T_PARAMS:
79 		resname = type;
80 		break;
81 	default:
82 		LBUG();
83 	}
84 	res_id->name[1] = cpu_to_le64(resname);
85 	CDEBUG(D_MGC, "log %s to resid %#llx/%#llx (%.8s)\n", name,
86 	       res_id->name[0], res_id->name[1], (char *)&res_id->name[0]);
87 	return 0;
88 }
89 
mgc_fsname2resid(char * fsname,struct ldlm_res_id * res_id,int type)90 int mgc_fsname2resid(char *fsname, struct ldlm_res_id *res_id, int type)
91 {
92 	/* fsname is at most 8 chars long, maybe contain "-".
93 	 * e.g. "lustre", "SUN-000" */
94 	return mgc_name2resid(fsname, strlen(fsname), res_id, type);
95 }
96 EXPORT_SYMBOL(mgc_fsname2resid);
97 
mgc_logname2resid(char * logname,struct ldlm_res_id * res_id,int type)98 int mgc_logname2resid(char *logname, struct ldlm_res_id *res_id, int type)
99 {
100 	char *name_end;
101 	int len;
102 
103 	/* logname consists of "fsname-nodetype".
104 	 * e.g. "lustre-MDT0001", "SUN-000-client"
105 	 * there is an exception: llog "params" */
106 	name_end = strrchr(logname, '-');
107 	if (!name_end)
108 		len = strlen(logname);
109 	else
110 		len = name_end - logname;
111 	return mgc_name2resid(logname, len, res_id, type);
112 }
113 
114 /********************** config llog list **********************/
115 static LIST_HEAD(config_llog_list);
116 static DEFINE_SPINLOCK(config_list_lock);
117 
118 /* Take a reference to a config log */
config_log_get(struct config_llog_data * cld)119 static int config_log_get(struct config_llog_data *cld)
120 {
121 	atomic_inc(&cld->cld_refcount);
122 	CDEBUG(D_INFO, "log %s refs %d\n", cld->cld_logname,
123 	       atomic_read(&cld->cld_refcount));
124 	return 0;
125 }
126 
127 /* Drop a reference to a config log.  When no longer referenced,
128    we can free the config log data */
config_log_put(struct config_llog_data * cld)129 static void config_log_put(struct config_llog_data *cld)
130 {
131 	CDEBUG(D_INFO, "log %s refs %d\n", cld->cld_logname,
132 	       atomic_read(&cld->cld_refcount));
133 	LASSERT(atomic_read(&cld->cld_refcount) > 0);
134 
135 	/* spinlock to make sure no item with 0 refcount in the list */
136 	if (atomic_dec_and_lock(&cld->cld_refcount, &config_list_lock)) {
137 		list_del(&cld->cld_list_chain);
138 		spin_unlock(&config_list_lock);
139 
140 		CDEBUG(D_MGC, "dropping config log %s\n", cld->cld_logname);
141 
142 		if (cld->cld_recover)
143 			config_log_put(cld->cld_recover);
144 		if (cld->cld_sptlrpc)
145 			config_log_put(cld->cld_sptlrpc);
146 		if (cld->cld_params)
147 			config_log_put(cld->cld_params);
148 		if (cld_is_sptlrpc(cld))
149 			sptlrpc_conf_log_stop(cld->cld_logname);
150 
151 		class_export_put(cld->cld_mgcexp);
152 		OBD_FREE(cld, sizeof(*cld) + strlen(cld->cld_logname) + 1);
153 	}
154 }
155 
156 /* Find a config log by name */
157 static
config_log_find(char * logname,struct config_llog_instance * cfg)158 struct config_llog_data *config_log_find(char *logname,
159 					 struct config_llog_instance *cfg)
160 {
161 	struct config_llog_data *cld;
162 	struct config_llog_data *found = NULL;
163 	void *		   instance;
164 
165 	LASSERT(logname != NULL);
166 
167 	instance = cfg ? cfg->cfg_instance : NULL;
168 	spin_lock(&config_list_lock);
169 	list_for_each_entry(cld, &config_llog_list, cld_list_chain) {
170 		/* check if instance equals */
171 		if (instance != cld->cld_cfg.cfg_instance)
172 			continue;
173 
174 		/* instance may be NULL, should check name */
175 		if (strcmp(logname, cld->cld_logname) == 0) {
176 			found = cld;
177 			break;
178 		}
179 	}
180 	if (found) {
181 		atomic_inc(&found->cld_refcount);
182 		LASSERT(found->cld_stopping == 0 || cld_is_sptlrpc(found) == 0);
183 	}
184 	spin_unlock(&config_list_lock);
185 	return found;
186 }
187 
188 static
do_config_log_add(struct obd_device * obd,char * logname,int type,struct config_llog_instance * cfg,struct super_block * sb)189 struct config_llog_data *do_config_log_add(struct obd_device *obd,
190 					   char *logname,
191 					   int type,
192 					   struct config_llog_instance *cfg,
193 					   struct super_block *sb)
194 {
195 	struct config_llog_data *cld;
196 	int		      rc;
197 
198 	CDEBUG(D_MGC, "do adding config log %s:%p\n", logname,
199 	       cfg ? cfg->cfg_instance : NULL);
200 
201 	OBD_ALLOC(cld, sizeof(*cld) + strlen(logname) + 1);
202 	if (!cld)
203 		return ERR_PTR(-ENOMEM);
204 
205 	strcpy(cld->cld_logname, logname);
206 	if (cfg)
207 		cld->cld_cfg = *cfg;
208 	else
209 		cld->cld_cfg.cfg_callback = class_config_llog_handler;
210 	mutex_init(&cld->cld_lock);
211 	cld->cld_cfg.cfg_last_idx = 0;
212 	cld->cld_cfg.cfg_flags = 0;
213 	cld->cld_cfg.cfg_sb = sb;
214 	cld->cld_type = type;
215 	atomic_set(&cld->cld_refcount, 1);
216 
217 	/* Keep the mgc around until we are done */
218 	cld->cld_mgcexp = class_export_get(obd->obd_self_export);
219 
220 	if (cld_is_sptlrpc(cld)) {
221 		sptlrpc_conf_log_start(logname);
222 		cld->cld_cfg.cfg_obdname = obd->obd_name;
223 	}
224 
225 	rc = mgc_logname2resid(logname, &cld->cld_resid, type);
226 
227 	spin_lock(&config_list_lock);
228 	list_add(&cld->cld_list_chain, &config_llog_list);
229 	spin_unlock(&config_list_lock);
230 
231 	if (rc) {
232 		config_log_put(cld);
233 		return ERR_PTR(rc);
234 	}
235 
236 	if (cld_is_sptlrpc(cld)) {
237 		rc = mgc_process_log(obd, cld);
238 		if (rc && rc != -ENOENT)
239 			CERROR("failed processing sptlrpc log: %d\n", rc);
240 	}
241 
242 	return cld;
243 }
244 
config_recover_log_add(struct obd_device * obd,char * fsname,struct config_llog_instance * cfg,struct super_block * sb)245 static struct config_llog_data *config_recover_log_add(struct obd_device *obd,
246 	char *fsname,
247 	struct config_llog_instance *cfg,
248 	struct super_block *sb)
249 {
250 	struct config_llog_instance lcfg = *cfg;
251 	struct lustre_sb_info *lsi = s2lsi(sb);
252 	struct config_llog_data *cld;
253 	char logname[32];
254 
255 	if (IS_OST(lsi))
256 		return NULL;
257 
258 	/* for osp-on-ost, see lustre_start_osp() */
259 	if (IS_MDT(lsi) && lcfg.cfg_instance)
260 		return NULL;
261 
262 	/* we have to use different llog for clients and mdts for cmd
263 	 * where only clients are notified if one of cmd server restarts */
264 	LASSERT(strlen(fsname) < sizeof(logname) / 2);
265 	strcpy(logname, fsname);
266 	if (IS_SERVER(lsi)) { /* mdt */
267 		LASSERT(lcfg.cfg_instance == NULL);
268 		lcfg.cfg_instance = sb;
269 		strcat(logname, "-mdtir");
270 	} else {
271 		LASSERT(lcfg.cfg_instance != NULL);
272 		strcat(logname, "-cliir");
273 	}
274 
275 	cld = do_config_log_add(obd, logname, CONFIG_T_RECOVER, &lcfg, sb);
276 	return cld;
277 }
278 
config_params_log_add(struct obd_device * obd,struct config_llog_instance * cfg,struct super_block * sb)279 static struct config_llog_data *config_params_log_add(struct obd_device *obd,
280 	struct config_llog_instance *cfg, struct super_block *sb)
281 {
282 	struct config_llog_instance	lcfg = *cfg;
283 	struct config_llog_data		*cld;
284 
285 	lcfg.cfg_instance = sb;
286 
287 	cld = do_config_log_add(obd, PARAMS_FILENAME, CONFIG_T_PARAMS,
288 				&lcfg, sb);
289 
290 	return cld;
291 }
292 
293 /** Add this log to the list of active logs watched by an MGC.
294  * Active means we're watching for updates.
295  * We have one active log per "mount" - client instance or servername.
296  * Each instance may be at a different point in the log.
297  */
config_log_add(struct obd_device * obd,char * logname,struct config_llog_instance * cfg,struct super_block * sb)298 static int config_log_add(struct obd_device *obd, char *logname,
299 			  struct config_llog_instance *cfg,
300 			  struct super_block *sb)
301 {
302 	struct lustre_sb_info *lsi = s2lsi(sb);
303 	struct config_llog_data *cld;
304 	struct config_llog_data *sptlrpc_cld;
305 	struct config_llog_data *params_cld;
306 	char			seclogname[32];
307 	char			*ptr;
308 	int			rc;
309 
310 	CDEBUG(D_MGC, "adding config log %s:%p\n", logname, cfg->cfg_instance);
311 
312 	/*
313 	 * for each regular log, the depended sptlrpc log name is
314 	 * <fsname>-sptlrpc. multiple regular logs may share one sptlrpc log.
315 	 */
316 	ptr = strrchr(logname, '-');
317 	if (ptr == NULL || ptr - logname > 8) {
318 		CERROR("logname %s is too long\n", logname);
319 		return -EINVAL;
320 	}
321 
322 	memcpy(seclogname, logname, ptr - logname);
323 	strcpy(seclogname + (ptr - logname), "-sptlrpc");
324 
325 	sptlrpc_cld = config_log_find(seclogname, NULL);
326 	if (sptlrpc_cld == NULL) {
327 		sptlrpc_cld = do_config_log_add(obd, seclogname,
328 						CONFIG_T_SPTLRPC, NULL, NULL);
329 		if (IS_ERR(sptlrpc_cld)) {
330 			CERROR("can't create sptlrpc log: %s\n", seclogname);
331 			rc = PTR_ERR(sptlrpc_cld);
332 			goto out_err;
333 		}
334 	}
335 	params_cld = config_params_log_add(obd, cfg, sb);
336 	if (IS_ERR(params_cld)) {
337 		rc = PTR_ERR(params_cld);
338 		CERROR("%s: can't create params log: rc = %d\n",
339 		       obd->obd_name, rc);
340 		goto out_err1;
341 	}
342 
343 	cld = do_config_log_add(obd, logname, CONFIG_T_CONFIG, cfg, sb);
344 	if (IS_ERR(cld)) {
345 		CERROR("can't create log: %s\n", logname);
346 		rc = PTR_ERR(cld);
347 		goto out_err2;
348 	}
349 
350 	cld->cld_sptlrpc = sptlrpc_cld;
351 	cld->cld_params = params_cld;
352 
353 	LASSERT(lsi->lsi_lmd);
354 	if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)) {
355 		struct config_llog_data *recover_cld;
356 		*strrchr(seclogname, '-') = 0;
357 		recover_cld = config_recover_log_add(obd, seclogname, cfg, sb);
358 		if (IS_ERR(recover_cld)) {
359 			rc = PTR_ERR(recover_cld);
360 			goto out_err3;
361 		}
362 		cld->cld_recover = recover_cld;
363 	}
364 
365 	return 0;
366 
367 out_err3:
368 	config_log_put(cld);
369 
370 out_err2:
371 	config_log_put(params_cld);
372 
373 out_err1:
374 	config_log_put(sptlrpc_cld);
375 
376 out_err:
377 	return rc;
378 }
379 
380 DEFINE_MUTEX(llog_process_lock);
381 
382 /** Stop watching for updates on this log.
383  */
config_log_end(char * logname,struct config_llog_instance * cfg)384 static int config_log_end(char *logname, struct config_llog_instance *cfg)
385 {
386 	struct config_llog_data *cld;
387 	struct config_llog_data *cld_sptlrpc = NULL;
388 	struct config_llog_data *cld_params = NULL;
389 	struct config_llog_data *cld_recover = NULL;
390 	int rc = 0;
391 
392 	cld = config_log_find(logname, cfg);
393 	if (cld == NULL)
394 		return -ENOENT;
395 
396 	mutex_lock(&cld->cld_lock);
397 	/*
398 	 * if cld_stopping is set, it means we didn't start the log thus
399 	 * not owning the start ref. this can happen after previous umount:
400 	 * the cld still hanging there waiting for lock cancel, and we
401 	 * remount again but failed in the middle and call log_end without
402 	 * calling start_log.
403 	 */
404 	if (unlikely(cld->cld_stopping)) {
405 		mutex_unlock(&cld->cld_lock);
406 		/* drop the ref from the find */
407 		config_log_put(cld);
408 		return rc;
409 	}
410 
411 	cld->cld_stopping = 1;
412 
413 	cld_recover = cld->cld_recover;
414 	cld->cld_recover = NULL;
415 	mutex_unlock(&cld->cld_lock);
416 
417 	if (cld_recover) {
418 		mutex_lock(&cld_recover->cld_lock);
419 		cld_recover->cld_stopping = 1;
420 		mutex_unlock(&cld_recover->cld_lock);
421 		config_log_put(cld_recover);
422 	}
423 
424 	spin_lock(&config_list_lock);
425 	cld_sptlrpc = cld->cld_sptlrpc;
426 	cld->cld_sptlrpc = NULL;
427 	cld_params = cld->cld_params;
428 	cld->cld_params = NULL;
429 	spin_unlock(&config_list_lock);
430 
431 	if (cld_sptlrpc)
432 		config_log_put(cld_sptlrpc);
433 
434 	if (cld_params) {
435 		mutex_lock(&cld_params->cld_lock);
436 		cld_params->cld_stopping = 1;
437 		mutex_unlock(&cld_params->cld_lock);
438 		config_log_put(cld_params);
439 	}
440 
441 	/* drop the ref from the find */
442 	config_log_put(cld);
443 	/* drop the start ref */
444 	config_log_put(cld);
445 
446 	CDEBUG(D_MGC, "end config log %s (%d)\n", logname ? logname : "client",
447 	       rc);
448 	return rc;
449 }
450 
451 #if defined (CONFIG_PROC_FS)
lprocfs_mgc_rd_ir_state(struct seq_file * m,void * data)452 int lprocfs_mgc_rd_ir_state(struct seq_file *m, void *data)
453 {
454 	struct obd_device       *obd = data;
455 	struct obd_import       *imp = obd->u.cli.cl_import;
456 	struct obd_connect_data *ocd = &imp->imp_connect_data;
457 	struct config_llog_data *cld;
458 
459 	seq_printf(m, "imperative_recovery: %s\n",
460 		      OCD_HAS_FLAG(ocd, IMP_RECOV) ? "ENABLED" : "DISABLED");
461 	seq_printf(m, "client_state:\n");
462 
463 	spin_lock(&config_list_lock);
464 	list_for_each_entry(cld, &config_llog_list, cld_list_chain) {
465 		if (cld->cld_recover == NULL)
466 			continue;
467 		seq_printf(m,  "    - { client: %s, nidtbl_version: %u }\n",
468 			       cld->cld_logname,
469 			       cld->cld_recover->cld_cfg.cfg_last_idx);
470 	}
471 	spin_unlock(&config_list_lock);
472 
473 	return 0;
474 }
475 #endif
476 
477 /* reenqueue any lost locks */
478 #define RQ_RUNNING 0x1
479 #define RQ_NOW     0x2
480 #define RQ_LATER   0x4
481 #define RQ_STOP    0x8
482 static int		    rq_state = 0;
483 static wait_queue_head_t	    rq_waitq;
484 static DECLARE_COMPLETION(rq_exit);
485 
do_requeue(struct config_llog_data * cld)486 static void do_requeue(struct config_llog_data *cld)
487 {
488 	LASSERT(atomic_read(&cld->cld_refcount) > 0);
489 
490 	/* Do not run mgc_process_log on a disconnected export or an
491 	   export which is being disconnected. Take the client
492 	   semaphore to make the check non-racy. */
493 	down_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem);
494 	if (cld->cld_mgcexp->exp_obd->u.cli.cl_conn_count != 0) {
495 		CDEBUG(D_MGC, "updating log %s\n", cld->cld_logname);
496 		mgc_process_log(cld->cld_mgcexp->exp_obd, cld);
497 	} else {
498 		CDEBUG(D_MGC, "disconnecting, won't update log %s\n",
499 		       cld->cld_logname);
500 	}
501 	up_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem);
502 }
503 
504 /* this timeout represents how many seconds MGC should wait before
505  * requeue config and recover lock to the MGS. We need to randomize this
506  * in order to not flood the MGS.
507  */
508 #define MGC_TIMEOUT_MIN_SECONDS   5
509 #define MGC_TIMEOUT_RAND_CENTISEC 0x1ff /* ~500 */
510 
mgc_requeue_thread(void * data)511 static int mgc_requeue_thread(void *data)
512 {
513 	int rc = 0;
514 
515 	CDEBUG(D_MGC, "Starting requeue thread\n");
516 
517 	/* Keep trying failed locks periodically */
518 	spin_lock(&config_list_lock);
519 	rq_state |= RQ_RUNNING;
520 	while (1) {
521 		struct l_wait_info lwi;
522 		struct config_llog_data *cld, *cld_prev;
523 		int rand = cfs_rand() & MGC_TIMEOUT_RAND_CENTISEC;
524 		int stopped = !!(rq_state & RQ_STOP);
525 		int to;
526 
527 		/* Any new or requeued lostlocks will change the state */
528 		rq_state &= ~(RQ_NOW | RQ_LATER);
529 		spin_unlock(&config_list_lock);
530 
531 		/* Always wait a few seconds to allow the server who
532 		   caused the lock revocation to finish its setup, plus some
533 		   random so everyone doesn't try to reconnect at once. */
534 		to = MGC_TIMEOUT_MIN_SECONDS * HZ;
535 		to += rand * HZ / 100; /* rand is centi-seconds */
536 		lwi = LWI_TIMEOUT(to, NULL, NULL);
537 		l_wait_event(rq_waitq, rq_state & RQ_STOP, &lwi);
538 
539 		/*
540 		 * iterate & processing through the list. for each cld, process
541 		 * its depending sptlrpc cld firstly (if any) and then itself.
542 		 *
543 		 * it's guaranteed any item in the list must have
544 		 * reference > 0; and if cld_lostlock is set, at
545 		 * least one reference is taken by the previous enqueue.
546 		 */
547 		cld_prev = NULL;
548 
549 		spin_lock(&config_list_lock);
550 		list_for_each_entry(cld, &config_llog_list,
551 					cld_list_chain) {
552 			if (!cld->cld_lostlock)
553 				continue;
554 
555 			spin_unlock(&config_list_lock);
556 
557 			LASSERT(atomic_read(&cld->cld_refcount) > 0);
558 
559 			/* Whether we enqueued again or not in mgc_process_log,
560 			 * we're done with the ref from the old enqueue */
561 			if (cld_prev)
562 				config_log_put(cld_prev);
563 			cld_prev = cld;
564 
565 			cld->cld_lostlock = 0;
566 			if (likely(!stopped))
567 				do_requeue(cld);
568 
569 			spin_lock(&config_list_lock);
570 		}
571 		spin_unlock(&config_list_lock);
572 		if (cld_prev)
573 			config_log_put(cld_prev);
574 
575 		/* break after scanning the list so that we can drop
576 		 * refcount to losing lock clds */
577 		if (unlikely(stopped)) {
578 			spin_lock(&config_list_lock);
579 			break;
580 		}
581 
582 		/* Wait a bit to see if anyone else needs a requeue */
583 		lwi = (struct l_wait_info) { 0 };
584 		l_wait_event(rq_waitq, rq_state & (RQ_NOW | RQ_STOP),
585 			     &lwi);
586 		spin_lock(&config_list_lock);
587 	}
588 	/* spinlock and while guarantee RQ_NOW and RQ_LATER are not set */
589 	rq_state &= ~RQ_RUNNING;
590 	spin_unlock(&config_list_lock);
591 
592 	complete(&rq_exit);
593 
594 	CDEBUG(D_MGC, "Ending requeue thread\n");
595 	return rc;
596 }
597 
598 /* Add a cld to the list to requeue.  Start the requeue thread if needed.
599    We are responsible for dropping the config log reference from here on out. */
mgc_requeue_add(struct config_llog_data * cld)600 static void mgc_requeue_add(struct config_llog_data *cld)
601 {
602 	CDEBUG(D_INFO, "log %s: requeue (r=%d sp=%d st=%x)\n",
603 	       cld->cld_logname, atomic_read(&cld->cld_refcount),
604 	       cld->cld_stopping, rq_state);
605 	LASSERT(atomic_read(&cld->cld_refcount) > 0);
606 
607 	mutex_lock(&cld->cld_lock);
608 	if (cld->cld_stopping || cld->cld_lostlock) {
609 		mutex_unlock(&cld->cld_lock);
610 		return;
611 	}
612 	/* this refcount will be released in mgc_requeue_thread. */
613 	config_log_get(cld);
614 	cld->cld_lostlock = 1;
615 	mutex_unlock(&cld->cld_lock);
616 
617 	/* Hold lock for rq_state */
618 	spin_lock(&config_list_lock);
619 	if (rq_state & RQ_STOP) {
620 		spin_unlock(&config_list_lock);
621 		cld->cld_lostlock = 0;
622 		config_log_put(cld);
623 	} else {
624 		rq_state |= RQ_NOW;
625 		spin_unlock(&config_list_lock);
626 		wake_up(&rq_waitq);
627 	}
628 }
629 
mgc_llog_init(const struct lu_env * env,struct obd_device * obd)630 static int mgc_llog_init(const struct lu_env *env, struct obd_device *obd)
631 {
632 	struct llog_ctxt	*ctxt;
633 	int			 rc;
634 
635 	/* setup only remote ctxt, the local disk context is switched per each
636 	 * filesystem during mgc_fs_setup() */
637 	rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_REPL_CTXT, obd,
638 			&llog_client_ops);
639 	if (rc)
640 		return rc;
641 
642 	ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
643 	LASSERT(ctxt);
644 
645 	llog_initiator_connect(ctxt);
646 	llog_ctxt_put(ctxt);
647 
648 	return 0;
649 }
650 
mgc_llog_fini(const struct lu_env * env,struct obd_device * obd)651 static int mgc_llog_fini(const struct lu_env *env, struct obd_device *obd)
652 {
653 	struct llog_ctxt *ctxt;
654 
655 	ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
656 	if (ctxt)
657 		llog_cleanup(env, ctxt);
658 
659 	return 0;
660 }
661 
662 static atomic_t mgc_count = ATOMIC_INIT(0);
mgc_precleanup(struct obd_device * obd,enum obd_cleanup_stage stage)663 static int mgc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
664 {
665 	int rc = 0;
666 
667 	switch (stage) {
668 	case OBD_CLEANUP_EARLY:
669 		break;
670 	case OBD_CLEANUP_EXPORTS:
671 		if (atomic_dec_and_test(&mgc_count)) {
672 			int running;
673 			/* stop requeue thread */
674 			spin_lock(&config_list_lock);
675 			running = rq_state & RQ_RUNNING;
676 			if (running)
677 				rq_state |= RQ_STOP;
678 			spin_unlock(&config_list_lock);
679 			if (running) {
680 				wake_up(&rq_waitq);
681 				wait_for_completion(&rq_exit);
682 			}
683 		}
684 		obd_cleanup_client_import(obd);
685 		rc = mgc_llog_fini(NULL, obd);
686 		if (rc != 0)
687 			CERROR("failed to cleanup llogging subsystems\n");
688 		break;
689 	}
690 	return rc;
691 }
692 
mgc_cleanup(struct obd_device * obd)693 static int mgc_cleanup(struct obd_device *obd)
694 {
695 	int rc;
696 
697 	/* COMPAT_146 - old config logs may have added profiles we don't
698 	   know about */
699 	if (obd->obd_type->typ_refcnt <= 1)
700 		/* Only for the last mgc */
701 		class_del_profiles();
702 
703 	lprocfs_obd_cleanup(obd);
704 	ptlrpcd_decref();
705 
706 	rc = client_obd_cleanup(obd);
707 	return rc;
708 }
709 
mgc_setup(struct obd_device * obd,struct lustre_cfg * lcfg)710 static int mgc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
711 {
712 	struct lprocfs_static_vars lvars;
713 	int rc;
714 
715 	ptlrpcd_addref();
716 
717 	rc = client_obd_setup(obd, lcfg);
718 	if (rc)
719 		goto err_decref;
720 
721 	rc = mgc_llog_init(NULL, obd);
722 	if (rc) {
723 		CERROR("failed to setup llogging subsystems\n");
724 		goto err_cleanup;
725 	}
726 
727 	lprocfs_mgc_init_vars(&lvars);
728 	lprocfs_obd_setup(obd, lvars.obd_vars);
729 	sptlrpc_lprocfs_cliobd_attach(obd);
730 
731 	if (atomic_inc_return(&mgc_count) == 1) {
732 		rq_state = 0;
733 		init_waitqueue_head(&rq_waitq);
734 
735 		/* start requeue thread */
736 		rc = PTR_ERR(kthread_run(mgc_requeue_thread, NULL,
737 					     "ll_cfg_requeue"));
738 		if (IS_ERR_VALUE(rc)) {
739 			CERROR("%s: Cannot start requeue thread (%d),"
740 			       "no more log updates!\n",
741 			       obd->obd_name, rc);
742 			goto err_cleanup;
743 		}
744 		/* rc is the task_struct pointer of mgc_requeue_thread. */
745 		rc = 0;
746 	}
747 
748 	return rc;
749 
750 err_cleanup:
751 	client_obd_cleanup(obd);
752 err_decref:
753 	ptlrpcd_decref();
754 	return rc;
755 }
756 
757 /* based on ll_mdc_blocking_ast */
mgc_blocking_ast(struct ldlm_lock * lock,struct ldlm_lock_desc * desc,void * data,int flag)758 static int mgc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
759 			    void *data, int flag)
760 {
761 	struct lustre_handle lockh;
762 	struct config_llog_data *cld = (struct config_llog_data *)data;
763 	int rc = 0;
764 
765 	switch (flag) {
766 	case LDLM_CB_BLOCKING:
767 		/* mgs wants the lock, give it up... */
768 		LDLM_DEBUG(lock, "MGC blocking CB");
769 		ldlm_lock2handle(lock, &lockh);
770 		rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
771 		break;
772 	case LDLM_CB_CANCELING:
773 		/* We've given up the lock, prepare ourselves to update. */
774 		LDLM_DEBUG(lock, "MGC cancel CB");
775 
776 		CDEBUG(D_MGC, "Lock res "DLDLMRES" (%.8s)\n",
777 		       PLDLMRES(lock->l_resource),
778 		       (char *)&lock->l_resource->lr_name.name[0]);
779 
780 		if (!cld) {
781 			CDEBUG(D_INFO, "missing data, won't requeue\n");
782 			break;
783 		}
784 
785 		/* held at mgc_process_log(). */
786 		LASSERT(atomic_read(&cld->cld_refcount) > 0);
787 		/* Are we done with this log? */
788 		if (cld->cld_stopping) {
789 			CDEBUG(D_MGC, "log %s: stopping, won't requeue\n",
790 			       cld->cld_logname);
791 			config_log_put(cld);
792 			break;
793 		}
794 		/* Make sure not to re-enqueue when the mgc is stopping
795 		   (we get called from client_disconnect_export) */
796 		if (!lock->l_conn_export ||
797 		    !lock->l_conn_export->exp_obd->u.cli.cl_conn_count) {
798 			CDEBUG(D_MGC, "log %.8s: disconnecting, won't requeue\n",
799 			       cld->cld_logname);
800 			config_log_put(cld);
801 			break;
802 		}
803 
804 		/* Re-enqueue now */
805 		mgc_requeue_add(cld);
806 		config_log_put(cld);
807 		break;
808 	default:
809 		LBUG();
810 	}
811 
812 	return rc;
813 }
814 
815 /* Not sure where this should go... */
816 /* This is the timeout value for MGS_CONNECT request plus a ping interval, such
817  * that we can have a chance to try the secondary MGS if any. */
818 #define  MGC_ENQUEUE_LIMIT (INITIAL_CONNECT_TIMEOUT + (AT_OFF ? 0 : at_min) \
819 				+ PING_INTERVAL)
820 #define  MGC_TARGET_REG_LIMIT 10
821 #define  MGC_SEND_PARAM_LIMIT 10
822 
823 /* Send parameter to MGS*/
mgc_set_mgs_param(struct obd_export * exp,struct mgs_send_param * msp)824 static int mgc_set_mgs_param(struct obd_export *exp,
825 			     struct mgs_send_param *msp)
826 {
827 	struct ptlrpc_request *req;
828 	struct mgs_send_param *req_msp, *rep_msp;
829 	int rc;
830 
831 	req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
832 					&RQF_MGS_SET_INFO, LUSTRE_MGS_VERSION,
833 					MGS_SET_INFO);
834 	if (!req)
835 		return -ENOMEM;
836 
837 	req_msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
838 	if (!req_msp) {
839 		ptlrpc_req_finished(req);
840 		return -ENOMEM;
841 	}
842 
843 	memcpy(req_msp, msp, sizeof(*req_msp));
844 	ptlrpc_request_set_replen(req);
845 
846 	/* Limit how long we will wait for the enqueue to complete */
847 	req->rq_delay_limit = MGC_SEND_PARAM_LIMIT;
848 	rc = ptlrpc_queue_wait(req);
849 	if (!rc) {
850 		rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
851 		memcpy(msp, rep_msp, sizeof(*rep_msp));
852 	}
853 
854 	ptlrpc_req_finished(req);
855 
856 	return rc;
857 }
858 
859 /* Take a config lock so we can get cancel notifications */
mgc_enqueue(struct obd_export * exp,struct lov_stripe_md * lsm,__u32 type,ldlm_policy_data_t * policy,__u32 mode,__u64 * flags,void * bl_cb,void * cp_cb,void * gl_cb,void * data,__u32 lvb_len,void * lvb_swabber,struct lustre_handle * lockh)860 static int mgc_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm,
861 		       __u32 type, ldlm_policy_data_t *policy, __u32 mode,
862 		       __u64 *flags, void *bl_cb, void *cp_cb, void *gl_cb,
863 		       void *data, __u32 lvb_len, void *lvb_swabber,
864 		       struct lustre_handle *lockh)
865 {
866 	struct config_llog_data *cld = (struct config_llog_data *)data;
867 	struct ldlm_enqueue_info einfo = {
868 		.ei_type	= type,
869 		.ei_mode	= mode,
870 		.ei_cb_bl	= mgc_blocking_ast,
871 		.ei_cb_cp	= ldlm_completion_ast,
872 	};
873 	struct ptlrpc_request *req;
874 	int short_limit = cld_is_sptlrpc(cld);
875 	int rc;
876 
877 	CDEBUG(D_MGC, "Enqueue for %s (res %#llx)\n", cld->cld_logname,
878 	       cld->cld_resid.name[0]);
879 
880 	/* We need a callback for every lockholder, so don't try to
881 	   ldlm_lock_match (see rev 1.1.2.11.2.47) */
882 	req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
883 					&RQF_LDLM_ENQUEUE, LUSTRE_DLM_VERSION,
884 					LDLM_ENQUEUE);
885 	if (req == NULL)
886 		return -ENOMEM;
887 
888 	req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, 0);
889 	ptlrpc_request_set_replen(req);
890 
891 	/* check if this is server or client */
892 	if (cld->cld_cfg.cfg_sb) {
893 		struct lustre_sb_info *lsi = s2lsi(cld->cld_cfg.cfg_sb);
894 		if (lsi && IS_SERVER(lsi))
895 			short_limit = 1;
896 	}
897 	/* Limit how long we will wait for the enqueue to complete */
898 	req->rq_delay_limit = short_limit ? 5 : MGC_ENQUEUE_LIMIT;
899 	rc = ldlm_cli_enqueue(exp, &req, &einfo, &cld->cld_resid, NULL, flags,
900 			      NULL, 0, LVB_T_NONE, lockh, 0);
901 	/* A failed enqueue should still call the mgc_blocking_ast,
902 	   where it will be requeued if needed ("grant failed"). */
903 	ptlrpc_req_finished(req);
904 	return rc;
905 }
906 
mgc_notify_active(struct obd_device * unused)907 static void mgc_notify_active(struct obd_device *unused)
908 {
909 	/* wakeup mgc_requeue_thread to requeue mgc lock */
910 	spin_lock(&config_list_lock);
911 	rq_state |= RQ_NOW;
912 	spin_unlock(&config_list_lock);
913 	wake_up(&rq_waitq);
914 
915 	/* TODO: Help the MGS rebuild nidtbl. -jay */
916 }
917 
918 /* Send target_reg message to MGS */
mgc_target_register(struct obd_export * exp,struct mgs_target_info * mti)919 static int mgc_target_register(struct obd_export *exp,
920 			       struct mgs_target_info *mti)
921 {
922 	struct ptlrpc_request  *req;
923 	struct mgs_target_info *req_mti, *rep_mti;
924 	int		     rc;
925 
926 	req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
927 					&RQF_MGS_TARGET_REG, LUSTRE_MGS_VERSION,
928 					MGS_TARGET_REG);
929 	if (req == NULL)
930 		return -ENOMEM;
931 
932 	req_mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
933 	if (!req_mti) {
934 		ptlrpc_req_finished(req);
935 		return -ENOMEM;
936 	}
937 
938 	memcpy(req_mti, mti, sizeof(*req_mti));
939 	ptlrpc_request_set_replen(req);
940 	CDEBUG(D_MGC, "register %s\n", mti->mti_svname);
941 	/* Limit how long we will wait for the enqueue to complete */
942 	req->rq_delay_limit = MGC_TARGET_REG_LIMIT;
943 
944 	rc = ptlrpc_queue_wait(req);
945 	if (!rc) {
946 		rep_mti = req_capsule_server_get(&req->rq_pill,
947 						 &RMF_MGS_TARGET_INFO);
948 		memcpy(mti, rep_mti, sizeof(*rep_mti));
949 		CDEBUG(D_MGC, "register %s got index = %d\n",
950 		       mti->mti_svname, mti->mti_stripe_index);
951 	}
952 	ptlrpc_req_finished(req);
953 
954 	return rc;
955 }
956 
mgc_set_info_async(const struct lu_env * env,struct obd_export * exp,u32 keylen,void * key,u32 vallen,void * val,struct ptlrpc_request_set * set)957 int mgc_set_info_async(const struct lu_env *env, struct obd_export *exp,
958 		       u32 keylen, void *key, u32 vallen,
959 		       void *val, struct ptlrpc_request_set *set)
960 {
961 	int rc = -EINVAL;
962 
963 	/* Turn off initial_recov after we try all backup servers once */
964 	if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
965 		struct obd_import *imp = class_exp2cliimp(exp);
966 		int value;
967 		if (vallen != sizeof(int))
968 			return -EINVAL;
969 		value = *(int *)val;
970 		CDEBUG(D_MGC, "InitRecov %s %d/d%d:i%d:r%d:or%d:%s\n",
971 		       imp->imp_obd->obd_name, value,
972 		       imp->imp_deactive, imp->imp_invalid,
973 		       imp->imp_replayable, imp->imp_obd->obd_replayable,
974 		       ptlrpc_import_state_name(imp->imp_state));
975 		/* Resurrect if we previously died */
976 		if ((imp->imp_state != LUSTRE_IMP_FULL &&
977 		     imp->imp_state != LUSTRE_IMP_NEW) || value > 1)
978 			ptlrpc_reconnect_import(imp);
979 		return 0;
980 	}
981 	if (KEY_IS(KEY_SET_INFO)) {
982 		struct mgs_send_param *msp;
983 
984 		msp = (struct mgs_send_param *)val;
985 		rc =  mgc_set_mgs_param(exp, msp);
986 		return rc;
987 	}
988 	if (KEY_IS(KEY_MGSSEC)) {
989 		struct client_obd     *cli = &exp->exp_obd->u.cli;
990 		struct sptlrpc_flavor  flvr;
991 
992 		/*
993 		 * empty string means using current flavor, if which haven't
994 		 * been set yet, set it as null.
995 		 *
996 		 * if flavor has been set previously, check the asking flavor
997 		 * must match the existing one.
998 		 */
999 		if (vallen == 0) {
1000 			if (cli->cl_flvr_mgc.sf_rpc != SPTLRPC_FLVR_INVALID)
1001 				return 0;
1002 			val = "null";
1003 			vallen = 4;
1004 		}
1005 
1006 		rc = sptlrpc_parse_flavor(val, &flvr);
1007 		if (rc) {
1008 			CERROR("invalid sptlrpc flavor %s to MGS\n",
1009 			       (char *) val);
1010 			return rc;
1011 		}
1012 
1013 		/*
1014 		 * caller already hold a mutex
1015 		 */
1016 		if (cli->cl_flvr_mgc.sf_rpc == SPTLRPC_FLVR_INVALID) {
1017 			cli->cl_flvr_mgc = flvr;
1018 		} else if (memcmp(&cli->cl_flvr_mgc, &flvr,
1019 				  sizeof(flvr)) != 0) {
1020 			char    str[20];
1021 
1022 			sptlrpc_flavor2name(&cli->cl_flvr_mgc,
1023 					    str, sizeof(str));
1024 			LCONSOLE_ERROR("asking sptlrpc flavor %s to MGS but "
1025 				       "currently %s is in use\n",
1026 				       (char *) val, str);
1027 			rc = -EPERM;
1028 		}
1029 		return rc;
1030 	}
1031 
1032 	return rc;
1033 }
1034 
mgc_get_info(const struct lu_env * env,struct obd_export * exp,__u32 keylen,void * key,__u32 * vallen,void * val,struct lov_stripe_md * unused)1035 static int mgc_get_info(const struct lu_env *env, struct obd_export *exp,
1036 			__u32 keylen, void *key, __u32 *vallen, void *val,
1037 			struct lov_stripe_md *unused)
1038 {
1039 	int rc = -EINVAL;
1040 
1041 	if (KEY_IS(KEY_CONN_DATA)) {
1042 		struct obd_import *imp = class_exp2cliimp(exp);
1043 		struct obd_connect_data *data = val;
1044 
1045 		if (*vallen == sizeof(*data)) {
1046 			*data = imp->imp_connect_data;
1047 			rc = 0;
1048 		}
1049 	}
1050 
1051 	return rc;
1052 }
1053 
mgc_import_event(struct obd_device * obd,struct obd_import * imp,enum obd_import_event event)1054 static int mgc_import_event(struct obd_device *obd,
1055 			    struct obd_import *imp,
1056 			    enum obd_import_event event)
1057 {
1058 	int rc = 0;
1059 
1060 	LASSERT(imp->imp_obd == obd);
1061 	CDEBUG(D_MGC, "import event %#x\n", event);
1062 
1063 	switch (event) {
1064 	case IMP_EVENT_DISCON:
1065 		/* MGC imports should not wait for recovery */
1066 		if (OCD_HAS_FLAG(&imp->imp_connect_data, IMP_RECOV))
1067 			ptlrpc_pinger_ir_down();
1068 		break;
1069 	case IMP_EVENT_INACTIVE:
1070 		break;
1071 	case IMP_EVENT_INVALIDATE: {
1072 		struct ldlm_namespace *ns = obd->obd_namespace;
1073 		ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1074 		break;
1075 	}
1076 	case IMP_EVENT_ACTIVE:
1077 		CDEBUG(D_INFO, "%s: Reactivating import\n", obd->obd_name);
1078 		/* Clearing obd_no_recov allows us to continue pinging */
1079 		obd->obd_no_recov = 0;
1080 		mgc_notify_active(obd);
1081 		if (OCD_HAS_FLAG(&imp->imp_connect_data, IMP_RECOV))
1082 			ptlrpc_pinger_ir_up();
1083 		break;
1084 	case IMP_EVENT_OCD:
1085 		break;
1086 	case IMP_EVENT_DEACTIVATE:
1087 	case IMP_EVENT_ACTIVATE:
1088 		break;
1089 	default:
1090 		CERROR("Unknown import event %#x\n", event);
1091 		LBUG();
1092 	}
1093 	return rc;
1094 }
1095 
1096 enum {
1097 	CONFIG_READ_NRPAGES_INIT = 1 << (20 - PAGE_CACHE_SHIFT),
1098 	CONFIG_READ_NRPAGES      = 4
1099 };
1100 
mgc_apply_recover_logs(struct obd_device * mgc,struct config_llog_data * cld,__u64 max_version,void * data,int datalen,bool mne_swab)1101 static int mgc_apply_recover_logs(struct obd_device *mgc,
1102 				  struct config_llog_data *cld,
1103 				  __u64 max_version,
1104 				  void *data, int datalen, bool mne_swab)
1105 {
1106 	struct config_llog_instance *cfg = &cld->cld_cfg;
1107 	struct lustre_sb_info       *lsi = s2lsi(cfg->cfg_sb);
1108 	struct mgs_nidtbl_entry *entry;
1109 	struct lustre_cfg       *lcfg;
1110 	struct lustre_cfg_bufs   bufs;
1111 	u64   prev_version = 0;
1112 	char *inst;
1113 	char *buf;
1114 	int   bufsz;
1115 	int   pos;
1116 	int   rc  = 0;
1117 	int   off = 0;
1118 
1119 	LASSERT(cfg->cfg_instance != NULL);
1120 	LASSERT(cfg->cfg_sb == cfg->cfg_instance);
1121 
1122 	OBD_ALLOC(inst, PAGE_CACHE_SIZE);
1123 	if (inst == NULL)
1124 		return -ENOMEM;
1125 
1126 	if (!IS_SERVER(lsi)) {
1127 		pos = snprintf(inst, PAGE_CACHE_SIZE, "%p", cfg->cfg_instance);
1128 		if (pos >= PAGE_CACHE_SIZE) {
1129 			OBD_FREE(inst, PAGE_CACHE_SIZE);
1130 			return -E2BIG;
1131 		}
1132 	} else {
1133 		LASSERT(IS_MDT(lsi));
1134 		rc = server_name2svname(lsi->lsi_svname, inst, NULL,
1135 					PAGE_CACHE_SIZE);
1136 		if (rc) {
1137 			OBD_FREE(inst, PAGE_CACHE_SIZE);
1138 			return -EINVAL;
1139 		}
1140 		pos = strlen(inst);
1141 	}
1142 
1143 	++pos;
1144 	buf   = inst + pos;
1145 	bufsz = PAGE_CACHE_SIZE - pos;
1146 
1147 	while (datalen > 0) {
1148 		int   entry_len = sizeof(*entry);
1149 		int   is_ost;
1150 		struct obd_device *obd;
1151 		char *obdname;
1152 		char *cname;
1153 		char *params;
1154 		char *uuid;
1155 
1156 		rc = -EINVAL;
1157 		if (datalen < sizeof(*entry))
1158 			break;
1159 
1160 		entry = (typeof(entry))(data + off);
1161 
1162 		/* sanity check */
1163 		if (entry->mne_nid_type != 0) /* only support type 0 for ipv4 */
1164 			break;
1165 		if (entry->mne_nid_count == 0) /* at least one nid entry */
1166 			break;
1167 		if (entry->mne_nid_size != sizeof(lnet_nid_t))
1168 			break;
1169 
1170 		entry_len += entry->mne_nid_count * entry->mne_nid_size;
1171 		if (datalen < entry_len) /* must have entry_len at least */
1172 			break;
1173 
1174 		/* Keep this swab for normal mixed endian handling. LU-1644 */
1175 		if (mne_swab)
1176 			lustre_swab_mgs_nidtbl_entry(entry);
1177 		if (entry->mne_length > PAGE_CACHE_SIZE) {
1178 			CERROR("MNE too large (%u)\n", entry->mne_length);
1179 			break;
1180 		}
1181 
1182 		if (entry->mne_length < entry_len)
1183 			break;
1184 
1185 		off     += entry->mne_length;
1186 		datalen -= entry->mne_length;
1187 		if (datalen < 0)
1188 			break;
1189 
1190 		if (entry->mne_version > max_version) {
1191 			CERROR("entry index(%lld) is over max_index(%lld)\n",
1192 			       entry->mne_version, max_version);
1193 			break;
1194 		}
1195 
1196 		if (prev_version >= entry->mne_version) {
1197 			CERROR("index unsorted, prev %lld, now %lld\n",
1198 			       prev_version, entry->mne_version);
1199 			break;
1200 		}
1201 		prev_version = entry->mne_version;
1202 
1203 		/*
1204 		 * Write a string with format "nid::instance" to
1205 		 * lustre/<osc|mdc>/<target>-<osc|mdc>-<instance>/import.
1206 		 */
1207 
1208 		is_ost = entry->mne_type == LDD_F_SV_TYPE_OST;
1209 		memset(buf, 0, bufsz);
1210 		obdname = buf;
1211 		pos = 0;
1212 
1213 		/* lustre-OST0001-osc-<instance #> */
1214 		strcpy(obdname, cld->cld_logname);
1215 		cname = strrchr(obdname, '-');
1216 		if (cname == NULL) {
1217 			CERROR("mgc %s: invalid logname %s\n",
1218 			       mgc->obd_name, obdname);
1219 			break;
1220 		}
1221 
1222 		pos = cname - obdname;
1223 		obdname[pos] = 0;
1224 		pos += sprintf(obdname + pos, "-%s%04x",
1225 				  is_ost ? "OST" : "MDT", entry->mne_index);
1226 
1227 		cname = is_ost ? "osc" : "mdc",
1228 		pos += sprintf(obdname + pos, "-%s-%s", cname, inst);
1229 		lustre_cfg_bufs_reset(&bufs, obdname);
1230 
1231 		/* find the obd by obdname */
1232 		obd = class_name2obd(obdname);
1233 		if (obd == NULL) {
1234 			CDEBUG(D_INFO, "mgc %s: cannot find obdname %s\n",
1235 			       mgc->obd_name, obdname);
1236 			rc = 0;
1237 			/* this is a safe race, when the ost is starting up...*/
1238 			continue;
1239 		}
1240 
1241 		/* osc.import = "connection=<Conn UUID>::<target instance>" */
1242 		++pos;
1243 		params = buf + pos;
1244 		pos += sprintf(params, "%s.import=%s", cname, "connection=");
1245 		uuid = buf + pos;
1246 
1247 		down_read(&obd->u.cli.cl_sem);
1248 		if (obd->u.cli.cl_import == NULL) {
1249 			/* client does not connect to the OST yet */
1250 			up_read(&obd->u.cli.cl_sem);
1251 			rc = 0;
1252 			continue;
1253 		}
1254 
1255 		/* TODO: iterate all nids to find one */
1256 		/* find uuid by nid */
1257 		rc = client_import_find_conn(obd->u.cli.cl_import,
1258 					     entry->u.nids[0],
1259 					     (struct obd_uuid *)uuid);
1260 		up_read(&obd->u.cli.cl_sem);
1261 		if (rc < 0) {
1262 			CERROR("mgc: cannot find uuid by nid %s\n",
1263 			       libcfs_nid2str(entry->u.nids[0]));
1264 			break;
1265 		}
1266 
1267 		CDEBUG(D_INFO, "Find uuid %s by nid %s\n",
1268 		       uuid, libcfs_nid2str(entry->u.nids[0]));
1269 
1270 		pos += strlen(uuid);
1271 		pos += sprintf(buf + pos, "::%u", entry->mne_instance);
1272 		LASSERT(pos < bufsz);
1273 
1274 		lustre_cfg_bufs_set_string(&bufs, 1, params);
1275 
1276 		rc = -ENOMEM;
1277 		lcfg = lustre_cfg_new(LCFG_PARAM, &bufs);
1278 		if (lcfg == NULL) {
1279 			CERROR("mgc: cannot allocate memory\n");
1280 			break;
1281 		}
1282 
1283 		CDEBUG(D_INFO, "ir apply logs %lld/%lld for %s -> %s\n",
1284 		       prev_version, max_version, obdname, params);
1285 
1286 		rc = class_process_config(lcfg);
1287 		lustre_cfg_free(lcfg);
1288 		if (rc)
1289 			CDEBUG(D_INFO, "process config for %s error %d\n",
1290 			       obdname, rc);
1291 
1292 		/* continue, even one with error */
1293 	}
1294 
1295 	OBD_FREE(inst, PAGE_CACHE_SIZE);
1296 	return rc;
1297 }
1298 
1299 /**
1300  * This function is called if this client was notified for target restarting
1301  * by the MGS. A CONFIG_READ RPC is going to send to fetch recovery logs.
1302  */
mgc_process_recover_log(struct obd_device * obd,struct config_llog_data * cld)1303 static int mgc_process_recover_log(struct obd_device *obd,
1304 				   struct config_llog_data *cld)
1305 {
1306 	struct ptlrpc_request *req = NULL;
1307 	struct config_llog_instance *cfg = &cld->cld_cfg;
1308 	struct mgs_config_body *body;
1309 	struct mgs_config_res  *res;
1310 	struct ptlrpc_bulk_desc *desc;
1311 	struct page **pages;
1312 	int nrpages;
1313 	bool eof = true;
1314 	bool mne_swab = false;
1315 	int i;
1316 	int ealen;
1317 	int rc;
1318 
1319 	/* allocate buffer for bulk transfer.
1320 	 * if this is the first time for this mgs to read logs,
1321 	 * CONFIG_READ_NRPAGES_INIT will be used since it will read all logs
1322 	 * once; otherwise, it only reads increment of logs, this should be
1323 	 * small and CONFIG_READ_NRPAGES will be used.
1324 	 */
1325 	nrpages = CONFIG_READ_NRPAGES;
1326 	if (cfg->cfg_last_idx == 0) /* the first time */
1327 		nrpages = CONFIG_READ_NRPAGES_INIT;
1328 
1329 	OBD_ALLOC(pages, sizeof(*pages) * nrpages);
1330 	if (pages == NULL) {
1331 		rc = -ENOMEM;
1332 		goto out;
1333 	}
1334 
1335 	for (i = 0; i < nrpages; i++) {
1336 		pages[i] = alloc_page(GFP_IOFS);
1337 		if (pages[i] == NULL) {
1338 			rc = -ENOMEM;
1339 			goto out;
1340 		}
1341 	}
1342 
1343 again:
1344 	LASSERT(cld_is_recover(cld));
1345 	LASSERT(mutex_is_locked(&cld->cld_lock));
1346 	req = ptlrpc_request_alloc(class_exp2cliimp(cld->cld_mgcexp),
1347 				   &RQF_MGS_CONFIG_READ);
1348 	if (req == NULL) {
1349 		rc = -ENOMEM;
1350 		goto out;
1351 	}
1352 
1353 	rc = ptlrpc_request_pack(req, LUSTRE_MGS_VERSION, MGS_CONFIG_READ);
1354 	if (rc)
1355 		goto out;
1356 
1357 	/* pack request */
1358 	body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
1359 	LASSERT(body != NULL);
1360 	LASSERT(sizeof(body->mcb_name) > strlen(cld->cld_logname));
1361 	if (strlcpy(body->mcb_name, cld->cld_logname, sizeof(body->mcb_name))
1362 	    >= sizeof(body->mcb_name)) {
1363 		rc = -E2BIG;
1364 		goto out;
1365 	}
1366 	body->mcb_offset = cfg->cfg_last_idx + 1;
1367 	body->mcb_type   = cld->cld_type;
1368 	body->mcb_bits   = PAGE_CACHE_SHIFT;
1369 	body->mcb_units  = nrpages;
1370 
1371 	/* allocate bulk transfer descriptor */
1372 	desc = ptlrpc_prep_bulk_imp(req, nrpages, 1, BULK_PUT_SINK,
1373 				    MGS_BULK_PORTAL);
1374 	if (desc == NULL) {
1375 		rc = -ENOMEM;
1376 		goto out;
1377 	}
1378 
1379 	for (i = 0; i < nrpages; i++)
1380 		ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1381 
1382 	ptlrpc_request_set_replen(req);
1383 	rc = ptlrpc_queue_wait(req);
1384 	if (rc)
1385 		goto out;
1386 
1387 	res = req_capsule_server_get(&req->rq_pill, &RMF_MGS_CONFIG_RES);
1388 	if (res->mcr_size < res->mcr_offset) {
1389 		rc = -EINVAL;
1390 		goto out;
1391 	}
1392 
1393 	/* always update the index even though it might have errors with
1394 	 * handling the recover logs */
1395 	cfg->cfg_last_idx = res->mcr_offset;
1396 	eof = res->mcr_offset == res->mcr_size;
1397 
1398 	CDEBUG(D_INFO, "Latest version %lld, more %d.\n",
1399 	       res->mcr_offset, eof == false);
1400 
1401 	ealen = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, 0);
1402 	if (ealen < 0) {
1403 		rc = ealen;
1404 		goto out;
1405 	}
1406 
1407 	if (ealen > nrpages << PAGE_CACHE_SHIFT) {
1408 		rc = -EINVAL;
1409 		goto out;
1410 	}
1411 
1412 	if (ealen == 0) { /* no logs transferred */
1413 		if (!eof)
1414 			rc = -EINVAL;
1415 		goto out;
1416 	}
1417 
1418 	mne_swab = !!ptlrpc_rep_need_swab(req);
1419 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
1420 	/* This import flag means the server did an extra swab of IR MNE
1421 	 * records (fixed in LU-1252), reverse it here if needed. LU-1644 */
1422 	if (unlikely(req->rq_import->imp_need_mne_swab))
1423 		mne_swab = !mne_swab;
1424 #else
1425 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
1426 #endif
1427 
1428 	for (i = 0; i < nrpages && ealen > 0; i++) {
1429 		int rc2;
1430 		void *ptr;
1431 
1432 		ptr = kmap(pages[i]);
1433 		rc2 = mgc_apply_recover_logs(obd, cld, res->mcr_offset, ptr,
1434 					     min_t(int, ealen, PAGE_CACHE_SIZE),
1435 					     mne_swab);
1436 		kunmap(pages[i]);
1437 		if (rc2 < 0) {
1438 			CWARN("Process recover log %s error %d\n",
1439 			      cld->cld_logname, rc2);
1440 			break;
1441 		}
1442 
1443 		ealen -= PAGE_CACHE_SIZE;
1444 	}
1445 
1446 out:
1447 	if (req)
1448 		ptlrpc_req_finished(req);
1449 
1450 	if (rc == 0 && !eof)
1451 		goto again;
1452 
1453 	if (pages) {
1454 		for (i = 0; i < nrpages; i++) {
1455 			if (pages[i] == NULL)
1456 				break;
1457 			__free_page(pages[i]);
1458 		}
1459 		OBD_FREE(pages, sizeof(*pages) * nrpages);
1460 	}
1461 	return rc;
1462 }
1463 
1464 /* local_only means it cannot get remote llogs */
mgc_process_cfg_log(struct obd_device * mgc,struct config_llog_data * cld,int local_only)1465 static int mgc_process_cfg_log(struct obd_device *mgc,
1466 			       struct config_llog_data *cld, int local_only)
1467 {
1468 	struct llog_ctxt	*ctxt;
1469 	struct lustre_sb_info	*lsi = NULL;
1470 	int			 rc = 0;
1471 	bool			 sptlrpc_started = false;
1472 	struct lu_env		*env;
1473 
1474 	LASSERT(cld);
1475 	LASSERT(mutex_is_locked(&cld->cld_lock));
1476 
1477 	/*
1478 	 * local copy of sptlrpc log is controlled elsewhere, don't try to
1479 	 * read it up here.
1480 	 */
1481 	if (cld_is_sptlrpc(cld) && local_only)
1482 		return 0;
1483 
1484 	if (cld->cld_cfg.cfg_sb)
1485 		lsi = s2lsi(cld->cld_cfg.cfg_sb);
1486 
1487 	OBD_ALLOC_PTR(env);
1488 	if (env == NULL)
1489 		return -ENOMEM;
1490 
1491 	rc = lu_env_init(env, LCT_MG_THREAD);
1492 	if (rc)
1493 		goto out_free;
1494 
1495 	ctxt = llog_get_context(mgc, LLOG_CONFIG_REPL_CTXT);
1496 	LASSERT(ctxt);
1497 
1498 	if (local_only) /* no local log at client side */ {
1499 		rc = -EIO;
1500 		goto out_pop;
1501 	}
1502 
1503 	if (cld_is_sptlrpc(cld)) {
1504 		sptlrpc_conf_log_update_begin(cld->cld_logname);
1505 		sptlrpc_started = true;
1506 	}
1507 
1508 	/* logname and instance info should be the same, so use our
1509 	 * copy of the instance for the update.  The cfg_last_idx will
1510 	 * be updated here. */
1511 	rc = class_config_parse_llog(env, ctxt, cld->cld_logname,
1512 				     &cld->cld_cfg);
1513 
1514 out_pop:
1515 	__llog_ctxt_put(env, ctxt);
1516 
1517 	/*
1518 	 * update settings on existing OBDs. doing it inside
1519 	 * of llog_process_lock so no device is attaching/detaching
1520 	 * in parallel.
1521 	 * the logname must be <fsname>-sptlrpc
1522 	 */
1523 	if (sptlrpc_started) {
1524 		LASSERT(cld_is_sptlrpc(cld));
1525 		sptlrpc_conf_log_update_end(cld->cld_logname);
1526 		class_notify_sptlrpc_conf(cld->cld_logname,
1527 					  strlen(cld->cld_logname) -
1528 					  strlen("-sptlrpc"));
1529 	}
1530 
1531 	lu_env_fini(env);
1532 out_free:
1533 	OBD_FREE_PTR(env);
1534 	return rc;
1535 }
1536 
1537 /** Get a config log from the MGS and process it.
1538  * This func is called for both clients and servers.
1539  * Copy the log locally before parsing it if appropriate (non-MGS server)
1540  */
mgc_process_log(struct obd_device * mgc,struct config_llog_data * cld)1541 int mgc_process_log(struct obd_device *mgc, struct config_llog_data *cld)
1542 {
1543 	struct lustre_handle lockh = { 0 };
1544 	__u64 flags = LDLM_FL_NO_LRU;
1545 	int rc = 0, rcl;
1546 
1547 	LASSERT(cld);
1548 
1549 	/* I don't want multiple processes running process_log at once --
1550 	   sounds like badness.  It actually might be fine, as long as
1551 	   we're not trying to update from the same log
1552 	   simultaneously (in which case we should use a per-log sem.) */
1553 	mutex_lock(&cld->cld_lock);
1554 	if (cld->cld_stopping) {
1555 		mutex_unlock(&cld->cld_lock);
1556 		return 0;
1557 	}
1558 
1559 	OBD_FAIL_TIMEOUT(OBD_FAIL_MGC_PAUSE_PROCESS_LOG, 20);
1560 
1561 	CDEBUG(D_MGC, "Process log %s:%p from %d\n", cld->cld_logname,
1562 	       cld->cld_cfg.cfg_instance, cld->cld_cfg.cfg_last_idx + 1);
1563 
1564 	/* Get the cfg lock on the llog */
1565 	rcl = mgc_enqueue(mgc->u.cli.cl_mgc_mgsexp, NULL, LDLM_PLAIN, NULL,
1566 			  LCK_CR, &flags, NULL, NULL, NULL,
1567 			  cld, 0, NULL, &lockh);
1568 	if (rcl == 0) {
1569 		/* Get the cld, it will be released in mgc_blocking_ast. */
1570 		config_log_get(cld);
1571 		rc = ldlm_lock_set_data(&lockh, (void *)cld);
1572 		LASSERT(rc == 0);
1573 	} else {
1574 		CDEBUG(D_MGC, "Can't get cfg lock: %d\n", rcl);
1575 
1576 		/* mark cld_lostlock so that it will requeue
1577 		 * after MGC becomes available. */
1578 		cld->cld_lostlock = 1;
1579 		/* Get extra reference, it will be put in requeue thread */
1580 		config_log_get(cld);
1581 	}
1582 
1583 
1584 	if (cld_is_recover(cld)) {
1585 		rc = 0; /* this is not a fatal error for recover log */
1586 		if (rcl == 0)
1587 			rc = mgc_process_recover_log(mgc, cld);
1588 	} else {
1589 		rc = mgc_process_cfg_log(mgc, cld, rcl != 0);
1590 	}
1591 
1592 	CDEBUG(D_MGC, "%s: configuration from log '%s' %sed (%d).\n",
1593 	       mgc->obd_name, cld->cld_logname, rc ? "fail" : "succeed", rc);
1594 
1595 	mutex_unlock(&cld->cld_lock);
1596 
1597 	/* Now drop the lock so MGS can revoke it */
1598 	if (!rcl)
1599 		ldlm_lock_decref(&lockh, LCK_CR);
1600 
1601 	return rc;
1602 }
1603 
1604 
1605 /** Called from lustre_process_log.
1606  * LCFG_LOG_START gets the config log from the MGS, processes it to start
1607  * any services, and adds it to the list logs to watch (follow).
1608  */
mgc_process_config(struct obd_device * obd,u32 len,void * buf)1609 static int mgc_process_config(struct obd_device *obd, u32 len, void *buf)
1610 {
1611 	struct lustre_cfg *lcfg = buf;
1612 	struct config_llog_instance *cfg = NULL;
1613 	char *logname;
1614 	int rc = 0;
1615 
1616 	switch (lcfg->lcfg_command) {
1617 	case LCFG_LOV_ADD_OBD: {
1618 		/* Overloading this cfg command: register a new target */
1619 		struct mgs_target_info *mti;
1620 
1621 		if (LUSTRE_CFG_BUFLEN(lcfg, 1) !=
1622 		    sizeof(struct mgs_target_info)) {
1623 			rc = -EINVAL;
1624 			goto out;
1625 		}
1626 
1627 		mti = (struct mgs_target_info *)lustre_cfg_buf(lcfg, 1);
1628 		CDEBUG(D_MGC, "add_target %s %#x\n",
1629 		       mti->mti_svname, mti->mti_flags);
1630 		rc = mgc_target_register(obd->u.cli.cl_mgc_mgsexp, mti);
1631 		break;
1632 	}
1633 	case LCFG_LOV_DEL_OBD:
1634 		/* Unregister has no meaning at the moment. */
1635 		CERROR("lov_del_obd unimplemented\n");
1636 		rc = -ENOSYS;
1637 		break;
1638 	case LCFG_SPTLRPC_CONF: {
1639 		rc = sptlrpc_process_config(lcfg);
1640 		break;
1641 	}
1642 	case LCFG_LOG_START: {
1643 		struct config_llog_data *cld;
1644 		struct super_block *sb;
1645 
1646 		logname = lustre_cfg_string(lcfg, 1);
1647 		cfg = (struct config_llog_instance *)lustre_cfg_buf(lcfg, 2);
1648 		sb = *(struct super_block **)lustre_cfg_buf(lcfg, 3);
1649 
1650 		CDEBUG(D_MGC, "parse_log %s from %d\n", logname,
1651 		       cfg->cfg_last_idx);
1652 
1653 		/* We're only called through here on the initial mount */
1654 		rc = config_log_add(obd, logname, cfg, sb);
1655 		if (rc)
1656 			break;
1657 		cld = config_log_find(logname, cfg);
1658 		if (cld == NULL) {
1659 			rc = -ENOENT;
1660 			break;
1661 		}
1662 
1663 		/* COMPAT_146 */
1664 		/* FIXME only set this for old logs!  Right now this forces
1665 		   us to always skip the "inside markers" check */
1666 		cld->cld_cfg.cfg_flags |= CFG_F_COMPAT146;
1667 
1668 		rc = mgc_process_log(obd, cld);
1669 		if (rc == 0 && cld->cld_recover != NULL) {
1670 			if (OCD_HAS_FLAG(&obd->u.cli.cl_import->
1671 					 imp_connect_data, IMP_RECOV)) {
1672 				rc = mgc_process_log(obd, cld->cld_recover);
1673 			} else {
1674 				struct config_llog_data *cir = cld->cld_recover;
1675 				cld->cld_recover = NULL;
1676 				config_log_put(cir);
1677 			}
1678 			if (rc)
1679 				CERROR("Cannot process recover llog %d\n", rc);
1680 		}
1681 
1682 		if (rc == 0 && cld->cld_params != NULL) {
1683 			rc = mgc_process_log(obd, cld->cld_params);
1684 			if (rc == -ENOENT) {
1685 				CDEBUG(D_MGC,
1686 				       "There is no params config file yet\n");
1687 				rc = 0;
1688 			}
1689 			/* params log is optional */
1690 			if (rc)
1691 				CERROR(
1692 				       "%s: can't process params llog: rc = %d\n",
1693 				       obd->obd_name, rc);
1694 		}
1695 		config_log_put(cld);
1696 
1697 		break;
1698 	}
1699 	case LCFG_LOG_END: {
1700 		logname = lustre_cfg_string(lcfg, 1);
1701 
1702 		if (lcfg->lcfg_bufcount >= 2)
1703 			cfg = (struct config_llog_instance *)lustre_cfg_buf(
1704 				lcfg, 2);
1705 		rc = config_log_end(logname, cfg);
1706 		break;
1707 	}
1708 	default: {
1709 		CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1710 		rc = -EINVAL;
1711 		goto out;
1712 
1713 	}
1714 	}
1715 out:
1716 	return rc;
1717 }
1718 
1719 struct obd_ops mgc_obd_ops = {
1720 	.o_owner	= THIS_MODULE,
1721 	.o_setup	= mgc_setup,
1722 	.o_precleanup   = mgc_precleanup,
1723 	.o_cleanup      = mgc_cleanup,
1724 	.o_add_conn     = client_import_add_conn,
1725 	.o_del_conn     = client_import_del_conn,
1726 	.o_connect      = client_connect_import,
1727 	.o_disconnect   = client_disconnect_export,
1728 	/* .o_enqueue      = mgc_enqueue, */
1729 	/* .o_iocontrol    = mgc_iocontrol, */
1730 	.o_set_info_async = mgc_set_info_async,
1731 	.o_get_info       = mgc_get_info,
1732 	.o_import_event = mgc_import_event,
1733 	.o_process_config = mgc_process_config,
1734 };
1735 
mgc_init(void)1736 int __init mgc_init(void)
1737 {
1738 	return class_register_type(&mgc_obd_ops, NULL, NULL,
1739 				   LUSTRE_MGC_NAME, NULL);
1740 }
1741 
mgc_exit(void)1742 static void /*__exit*/ mgc_exit(void)
1743 {
1744 	class_unregister_type(LUSTRE_MGC_NAME);
1745 }
1746 
1747 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1748 MODULE_DESCRIPTION("Lustre Management Client");
1749 MODULE_LICENSE("GPL");
1750 
1751 module_init(mgc_init);
1752 module_exit(mgc_exit);
1753