projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[deliverable/linux.git] / drivers / staging / lustre / lustre / mgc / mgc_request.c
1 /*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
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, 2015, 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
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
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 */
95 return mgc_name2resid(fsname, strlen(fsname), res_id, type);
96 }
97 EXPORT_SYMBOL(mgc_fsname2resid);
98
99 static int mgc_logname2resid(char *logname, struct ldlm_res_id *res_id, int type)
100 {
101 char *name_end;
102 int len;
103
104 /* logname consists of "fsname-nodetype".
105 * e.g. "lustre-MDT0001", "SUN-000-client"
106 * there is an exception: llog "params"
107 */
108 name_end = strrchr(logname, '-');
109 if (!name_end)
110 len = strlen(logname);
111 else
112 len = name_end - logname;
113 return mgc_name2resid(logname, len, res_id, type);
114 }
115
116 /********************** config llog list **********************/
117 static LIST_HEAD(config_llog_list);
118 static DEFINE_SPINLOCK(config_list_lock);
119
120 /* Take a reference to a config log */
121 static int config_log_get(struct config_llog_data *cld)
122 {
123 atomic_inc(&cld->cld_refcount);
124 CDEBUG(D_INFO, "log %s refs %d\n", cld->cld_logname,
125 atomic_read(&cld->cld_refcount));
126 return 0;
127 }
128
129 /* Drop a reference to a config log. When no longer referenced,
130 * we can free the config log data
131 */
132 static void config_log_put(struct config_llog_data *cld)
133 {
134 CDEBUG(D_INFO, "log %s refs %d\n", cld->cld_logname,
135 atomic_read(&cld->cld_refcount));
136 LASSERT(atomic_read(&cld->cld_refcount) > 0);
137
138 /* spinlock to make sure no item with 0 refcount in the list */
139 if (atomic_dec_and_lock(&cld->cld_refcount, &config_list_lock)) {
140 list_del(&cld->cld_list_chain);
141 spin_unlock(&config_list_lock);
142
143 CDEBUG(D_MGC, "dropping config log %s\n", cld->cld_logname);
144
145 if (cld->cld_recover)
146 config_log_put(cld->cld_recover);
147 if (cld->cld_sptlrpc)
148 config_log_put(cld->cld_sptlrpc);
149 if (cld->cld_params)
150 config_log_put(cld->cld_params);
151 if (cld_is_sptlrpc(cld))
152 sptlrpc_conf_log_stop(cld->cld_logname);
153
154 class_export_put(cld->cld_mgcexp);
155 kfree(cld);
156 }
157 }
158
159 /* Find a config log by name */
160 static
161 struct config_llog_data *config_log_find(char *logname,
162 struct config_llog_instance *cfg)
163 {
164 struct config_llog_data *cld;
165 struct config_llog_data *found = NULL;
166 void *instance;
167
168 LASSERT(logname);
169
170 instance = cfg ? cfg->cfg_instance : NULL;
171 spin_lock(&config_list_lock);
172 list_for_each_entry(cld, &config_llog_list, cld_list_chain) {
173 /* check if instance equals */
174 if (instance != cld->cld_cfg.cfg_instance)
175 continue;
176
177 /* instance may be NULL, should check name */
178 if (strcmp(logname, cld->cld_logname) == 0) {
179 found = cld;
180 break;
181 }
182 }
183 if (found) {
184 atomic_inc(&found->cld_refcount);
185 LASSERT(found->cld_stopping == 0 || cld_is_sptlrpc(found) == 0);
186 }
187 spin_unlock(&config_list_lock);
188 return found;
189 }
190
191 static
192 struct config_llog_data *do_config_log_add(struct obd_device *obd,
193 char *logname,
194 int type,
195 struct config_llog_instance *cfg,
196 struct super_block *sb)
197 {
198 struct config_llog_data *cld;
199 int rc;
200
201 CDEBUG(D_MGC, "do adding config log %s:%p\n", logname,
202 cfg ? cfg->cfg_instance : NULL);
203
204 cld = kzalloc(sizeof(*cld) + strlen(logname) + 1, GFP_NOFS);
205 if (!cld)
206 return ERR_PTR(-ENOMEM);
207
208 strcpy(cld->cld_logname, logname);
209 if (cfg)
210 cld->cld_cfg = *cfg;
211 else
212 cld->cld_cfg.cfg_callback = class_config_llog_handler;
213 mutex_init(&cld->cld_lock);
214 cld->cld_cfg.cfg_last_idx = 0;
215 cld->cld_cfg.cfg_flags = 0;
216 cld->cld_cfg.cfg_sb = sb;
217 cld->cld_type = type;
218 atomic_set(&cld->cld_refcount, 1);
219
220 /* Keep the mgc around until we are done */
221 cld->cld_mgcexp = class_export_get(obd->obd_self_export);
222
223 if (cld_is_sptlrpc(cld)) {
224 sptlrpc_conf_log_start(logname);
225 cld->cld_cfg.cfg_obdname = obd->obd_name;
226 }
227
228 rc = mgc_logname2resid(logname, &cld->cld_resid, type);
229
230 spin_lock(&config_list_lock);
231 list_add(&cld->cld_list_chain, &config_llog_list);
232 spin_unlock(&config_list_lock);
233
234 if (rc) {
235 config_log_put(cld);
236 return ERR_PTR(rc);
237 }
238
239 if (cld_is_sptlrpc(cld)) {
240 rc = mgc_process_log(obd, cld);
241 if (rc && rc != -ENOENT)
242 CERROR("failed processing sptlrpc log: %d\n", rc);
243 }
244
245 return cld;
246 }
247
248 static struct config_llog_data *
249 config_recover_log_add(struct obd_device *obd, char *fsname,
250 struct config_llog_instance *cfg,
251 struct super_block *sb)
252 {
253 struct config_llog_instance lcfg = *cfg;
254 struct config_llog_data *cld;
255 char logname[32];
256
257 /* we have to use different llog for clients and mdts for cmd
258 * where only clients are notified if one of cmd server restarts
259 */
260 LASSERT(strlen(fsname) < sizeof(logname) / 2);
261 strcpy(logname, fsname);
262 LASSERT(lcfg.cfg_instance);
263 strcat(logname, "-cliir");
264
265 cld = do_config_log_add(obd, logname, CONFIG_T_RECOVER, &lcfg, sb);
266 return cld;
267 }
268
269 static struct config_llog_data *
270 config_params_log_add(struct obd_device *obd,
271 struct config_llog_instance *cfg, struct super_block *sb)
272 {
273 struct config_llog_instance lcfg = *cfg;
274 struct config_llog_data *cld;
275
276 lcfg.cfg_instance = sb;
277
278 cld = do_config_log_add(obd, PARAMS_FILENAME, CONFIG_T_PARAMS,
279 &lcfg, sb);
280
281 return cld;
282 }
283
284 /** Add this log to the list of active logs watched by an MGC.
285 * Active means we're watching for updates.
286 * We have one active log per "mount" - client instance or servername.
287 * Each instance may be at a different point in the log.
288 */
289 static int config_log_add(struct obd_device *obd, char *logname,
290 struct config_llog_instance *cfg,
291 struct super_block *sb)
292 {
293 struct lustre_sb_info *lsi = s2lsi(sb);
294 struct config_llog_data *cld;
295 struct config_llog_data *sptlrpc_cld;
296 struct config_llog_data *params_cld;
297 char seclogname[32];
298 char *ptr;
299 int rc;
300
301 CDEBUG(D_MGC, "adding config log %s:%p\n", logname, cfg->cfg_instance);
302
303 /*
304 * for each regular log, the depended sptlrpc log name is
305 * <fsname>-sptlrpc. multiple regular logs may share one sptlrpc log.
306 */
307 ptr = strrchr(logname, '-');
308 if (!ptr || ptr - logname > 8) {
309 CERROR("logname %s is too long\n", logname);
310 return -EINVAL;
311 }
312
313 memcpy(seclogname, logname, ptr - logname);
314 strcpy(seclogname + (ptr - logname), "-sptlrpc");
315
316 sptlrpc_cld = config_log_find(seclogname, NULL);
317 if (!sptlrpc_cld) {
318 sptlrpc_cld = do_config_log_add(obd, seclogname,
319 CONFIG_T_SPTLRPC, NULL, NULL);
320 if (IS_ERR(sptlrpc_cld)) {
321 CERROR("can't create sptlrpc log: %s\n", seclogname);
322 rc = PTR_ERR(sptlrpc_cld);
323 goto out_err;
324 }
325 }
326 params_cld = config_params_log_add(obd, cfg, sb);
327 if (IS_ERR(params_cld)) {
328 rc = PTR_ERR(params_cld);
329 CERROR("%s: can't create params log: rc = %d\n",
330 obd->obd_name, rc);
331 goto out_err1;
332 }
333
334 cld = do_config_log_add(obd, logname, CONFIG_T_CONFIG, cfg, sb);
335 if (IS_ERR(cld)) {
336 CERROR("can't create log: %s\n", logname);
337 rc = PTR_ERR(cld);
338 goto out_err2;
339 }
340
341 cld->cld_sptlrpc = sptlrpc_cld;
342 cld->cld_params = params_cld;
343
344 LASSERT(lsi->lsi_lmd);
345 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)) {
346 struct config_llog_data *recover_cld;
347
348 ptr = strrchr(seclogname, '-');
349 if (ptr) {
350 *ptr = 0;
351 } else {
352 CERROR("%s: sptlrpc log name not correct, %s: rc = %d\n",
353 obd->obd_name, seclogname, -EINVAL);
354 config_log_put(cld);
355 return -EINVAL;
356 }
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 */
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)
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 int lprocfs_mgc_rd_ir_state(struct seq_file *m, void *data)
452 {
453 struct obd_device *obd = data;
454 struct obd_import *imp;
455 struct obd_connect_data *ocd;
456 struct config_llog_data *cld;
457 int rc;
458
459 rc = lprocfs_climp_check(obd);
460 if (rc)
461 return rc;
462
463 imp = obd->u.cli.cl_import;
464 ocd = &imp->imp_connect_data;
465
466 seq_printf(m, "imperative_recovery: %s\n",
467 OCD_HAS_FLAG(ocd, IMP_RECOV) ? "ENABLED" : "DISABLED");
468 seq_printf(m, "client_state:\n");
469
470 spin_lock(&config_list_lock);
471 list_for_each_entry(cld, &config_llog_list, cld_list_chain) {
472 if (!cld->cld_recover)
473 continue;
474 seq_printf(m, " - { client: %s, nidtbl_version: %u }\n",
475 cld->cld_logname,
476 cld->cld_recover->cld_cfg.cfg_last_idx);
477 }
478 spin_unlock(&config_list_lock);
479
480 up_read(&obd->u.cli.cl_sem);
481 return 0;
482 }
483
484 /* reenqueue any lost locks */
485 #define RQ_RUNNING 0x1
486 #define RQ_NOW 0x2
487 #define RQ_LATER 0x4
488 #define RQ_STOP 0x8
489 #define RQ_PRECLEANUP 0x10
490 static int rq_state;
491 static wait_queue_head_t rq_waitq;
492 static DECLARE_COMPLETION(rq_exit);
493 static DECLARE_COMPLETION(rq_start);
494
495 static void do_requeue(struct config_llog_data *cld)
496 {
497 LASSERT(atomic_read(&cld->cld_refcount) > 0);
498
499 /* Do not run mgc_process_log on a disconnected export or an
500 * export which is being disconnected. Take the client
501 * semaphore to make the check non-racy.
502 */
503 down_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem);
504 if (cld->cld_mgcexp->exp_obd->u.cli.cl_conn_count != 0) {
505 CDEBUG(D_MGC, "updating log %s\n", cld->cld_logname);
506 mgc_process_log(cld->cld_mgcexp->exp_obd, cld);
507 } else {
508 CDEBUG(D_MGC, "disconnecting, won't update log %s\n",
509 cld->cld_logname);
510 }
511 up_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem);
512 }
513
514 /* this timeout represents how many seconds MGC should wait before
515 * requeue config and recover lock to the MGS. We need to randomize this
516 * in order to not flood the MGS.
517 */
518 #define MGC_TIMEOUT_MIN_SECONDS 5
519 #define MGC_TIMEOUT_RAND_CENTISEC 0x1ff /* ~500 */
520
521 static int mgc_requeue_thread(void *data)
522 {
523 bool first = true;
524
525 CDEBUG(D_MGC, "Starting requeue thread\n");
526
527 /* Keep trying failed locks periodically */
528 spin_lock(&config_list_lock);
529 rq_state |= RQ_RUNNING;
530 while (1) {
531 struct l_wait_info lwi;
532 struct config_llog_data *cld, *cld_prev;
533 int rand = cfs_rand() & MGC_TIMEOUT_RAND_CENTISEC;
534 int stopped = !!(rq_state & RQ_STOP);
535 int to;
536
537 /* Any new or requeued lostlocks will change the state */
538 rq_state &= ~(RQ_NOW | RQ_LATER);
539 spin_unlock(&config_list_lock);
540
541 if (first) {
542 first = false;
543 complete(&rq_start);
544 }
545
546 /* Always wait a few seconds to allow the server who
547 * caused the lock revocation to finish its setup, plus some
548 * random so everyone doesn't try to reconnect at once.
549 */
550 to = MGC_TIMEOUT_MIN_SECONDS * HZ;
551 to += rand * HZ / 100; /* rand is centi-seconds */
552 lwi = LWI_TIMEOUT(to, NULL, NULL);
553 l_wait_event(rq_waitq, rq_state & (RQ_STOP | RQ_PRECLEANUP),
554 &lwi);
555
556 /*
557 * iterate & processing through the list. for each cld, process
558 * its depending sptlrpc cld firstly (if any) and then itself.
559 *
560 * it's guaranteed any item in the list must have
561 * reference > 0; and if cld_lostlock is set, at
562 * least one reference is taken by the previous enqueue.
563 */
564 cld_prev = NULL;
565
566 spin_lock(&config_list_lock);
567 rq_state &= ~RQ_PRECLEANUP;
568 list_for_each_entry(cld, &config_llog_list, cld_list_chain) {
569 if (!cld->cld_lostlock)
570 continue;
571
572 spin_unlock(&config_list_lock);
573
574 LASSERT(atomic_read(&cld->cld_refcount) > 0);
575
576 /* Whether we enqueued again or not in mgc_process_log,
577 * we're done with the ref from the old enqueue
578 */
579 if (cld_prev)
580 config_log_put(cld_prev);
581 cld_prev = cld;
582
583 cld->cld_lostlock = 0;
584 if (likely(!stopped))
585 do_requeue(cld);
586
587 spin_lock(&config_list_lock);
588 }
589 spin_unlock(&config_list_lock);
590 if (cld_prev)
591 config_log_put(cld_prev);
592
593 /* break after scanning the list so that we can drop
594 * refcount to losing lock clds
595 */
596 if (unlikely(stopped)) {
597 spin_lock(&config_list_lock);
598 break;
599 }
600
601 /* Wait a bit to see if anyone else needs a requeue */
602 lwi = (struct l_wait_info) { 0 };
603 l_wait_event(rq_waitq, rq_state & (RQ_NOW | RQ_STOP),
604 &lwi);
605 spin_lock(&config_list_lock);
606 }
607 /* spinlock and while guarantee RQ_NOW and RQ_LATER are not set */
608 rq_state &= ~RQ_RUNNING;
609 spin_unlock(&config_list_lock);
610
611 complete(&rq_exit);
612
613 CDEBUG(D_MGC, "Ending requeue thread\n");
614 return 0;
615 }
616
617 /* Add a cld to the list to requeue. Start the requeue thread if needed.
618 * We are responsible for dropping the config log reference from here on out.
619 */
620 static void mgc_requeue_add(struct config_llog_data *cld)
621 {
622 CDEBUG(D_INFO, "log %s: requeue (r=%d sp=%d st=%x)\n",
623 cld->cld_logname, atomic_read(&cld->cld_refcount),
624 cld->cld_stopping, rq_state);
625 LASSERT(atomic_read(&cld->cld_refcount) > 0);
626
627 mutex_lock(&cld->cld_lock);
628 if (cld->cld_stopping || cld->cld_lostlock) {
629 mutex_unlock(&cld->cld_lock);
630 return;
631 }
632 /* this refcount will be released in mgc_requeue_thread. */
633 config_log_get(cld);
634 cld->cld_lostlock = 1;
635 mutex_unlock(&cld->cld_lock);
636
637 /* Hold lock for rq_state */
638 spin_lock(&config_list_lock);
639 if (rq_state & RQ_STOP) {
640 spin_unlock(&config_list_lock);
641 cld->cld_lostlock = 0;
642 config_log_put(cld);
643 } else {
644 rq_state |= RQ_NOW;
645 spin_unlock(&config_list_lock);
646 wake_up(&rq_waitq);
647 }
648 }
649
650 static int mgc_llog_init(const struct lu_env *env, struct obd_device *obd)
651 {
652 struct llog_ctxt *ctxt;
653 int rc;
654
655 /* setup only remote ctxt, the local disk context is switched per each
656 * filesystem during mgc_fs_setup()
657 */
658 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_REPL_CTXT, obd,
659 &llog_client_ops);
660 if (rc)
661 return rc;
662
663 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
664 LASSERT(ctxt);
665
666 llog_initiator_connect(ctxt);
667 llog_ctxt_put(ctxt);
668
669 return 0;
670 }
671
672 static int mgc_llog_fini(const struct lu_env *env, struct obd_device *obd)
673 {
674 struct llog_ctxt *ctxt;
675
676 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
677 if (ctxt)
678 llog_cleanup(env, ctxt);
679
680 return 0;
681 }
682
683 static atomic_t mgc_count = ATOMIC_INIT(0);
684 static int mgc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
685 {
686 int rc = 0;
687 int temp;
688
689 switch (stage) {
690 case OBD_CLEANUP_EARLY:
691 break;
692 case OBD_CLEANUP_EXPORTS:
693 if (atomic_dec_and_test(&mgc_count)) {
694 LASSERT(rq_state & RQ_RUNNING);
695 /* stop requeue thread */
696 temp = RQ_STOP;
697 } else {
698 /* wakeup requeue thread to clean our cld */
699 temp = RQ_NOW | RQ_PRECLEANUP;
700 }
701 spin_lock(&config_list_lock);
702 rq_state |= temp;
703 spin_unlock(&config_list_lock);
704 wake_up(&rq_waitq);
705 if (temp & RQ_STOP)
706 wait_for_completion(&rq_exit);
707 obd_cleanup_client_import(obd);
708 rc = mgc_llog_fini(NULL, obd);
709 if (rc != 0)
710 CERROR("failed to cleanup llogging subsystems\n");
711 break;
712 }
713 return rc;
714 }
715
716 static int mgc_cleanup(struct obd_device *obd)
717 {
718 /* COMPAT_146 - old config logs may have added profiles we don't
719 * know about
720 */
721 if (obd->obd_type->typ_refcnt <= 1)
722 /* Only for the last mgc */
723 class_del_profiles();
724
725 lprocfs_obd_cleanup(obd);
726 ptlrpcd_decref();
727
728 return client_obd_cleanup(obd);
729 }
730
731 static int mgc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
732 {
733 struct lprocfs_static_vars lvars = { NULL };
734 struct task_struct *task;
735 int rc;
736
737 ptlrpcd_addref();
738
739 rc = client_obd_setup(obd, lcfg);
740 if (rc)
741 goto err_decref;
742
743 rc = mgc_llog_init(NULL, obd);
744 if (rc) {
745 CERROR("failed to setup llogging subsystems\n");
746 goto err_cleanup;
747 }
748
749 lprocfs_mgc_init_vars(&lvars);
750 lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
751 sptlrpc_lprocfs_cliobd_attach(obd);
752
753 if (atomic_inc_return(&mgc_count) == 1) {
754 rq_state = 0;
755 init_waitqueue_head(&rq_waitq);
756
757 /* start requeue thread */
758 task = kthread_run(mgc_requeue_thread, NULL, "ll_cfg_requeue");
759 if (IS_ERR(task)) {
760 rc = PTR_ERR(task);
761 CERROR("%s: cannot start requeue thread: rc = %d; no more log updates\n",
762 obd->obd_name, rc);
763 goto err_cleanup;
764 }
765 /* rc is the task_struct pointer of mgc_requeue_thread. */
766 rc = 0;
767 wait_for_completion(&rq_start);
768 }
769
770 return rc;
771
772 err_cleanup:
773 client_obd_cleanup(obd);
774 err_decref:
775 ptlrpcd_decref();
776 return rc;
777 }
778
779 /* based on ll_mdc_blocking_ast */
780 static int mgc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
781 void *data, int flag)
782 {
783 struct lustre_handle lockh;
784 struct config_llog_data *cld = data;
785 int rc = 0;
786
787 switch (flag) {
788 case LDLM_CB_BLOCKING:
789 /* mgs wants the lock, give it up... */
790 LDLM_DEBUG(lock, "MGC blocking CB");
791 ldlm_lock2handle(lock, &lockh);
792 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
793 break;
794 case LDLM_CB_CANCELING:
795 /* We've given up the lock, prepare ourselves to update. */
796 LDLM_DEBUG(lock, "MGC cancel CB");
797
798 CDEBUG(D_MGC, "Lock res "DLDLMRES" (%.8s)\n",
799 PLDLMRES(lock->l_resource),
800 (char *)&lock->l_resource->lr_name.name[0]);
801
802 if (!cld) {
803 CDEBUG(D_INFO, "missing data, won't requeue\n");
804 break;
805 }
806
807 /* held at mgc_process_log(). */
808 LASSERT(atomic_read(&cld->cld_refcount) > 0);
809 /* Are we done with this log? */
810 if (cld->cld_stopping) {
811 CDEBUG(D_MGC, "log %s: stopping, won't requeue\n",
812 cld->cld_logname);
813 config_log_put(cld);
814 break;
815 }
816 /* Make sure not to re-enqueue when the mgc is stopping
817 * (we get called from client_disconnect_export)
818 */
819 if (!lock->l_conn_export ||
820 !lock->l_conn_export->exp_obd->u.cli.cl_conn_count) {
821 CDEBUG(D_MGC, "log %.8s: disconnecting, won't requeue\n",
822 cld->cld_logname);
823 config_log_put(cld);
824 break;
825 }
826
827 /* Re-enqueue now */
828 mgc_requeue_add(cld);
829 config_log_put(cld);
830 break;
831 default:
832 LBUG();
833 }
834
835 return rc;
836 }
837
838 /* Not sure where this should go... */
839 /* This is the timeout value for MGS_CONNECT request plus a ping interval, such
840 * that we can have a chance to try the secondary MGS if any.
841 */
842 #define MGC_ENQUEUE_LIMIT (INITIAL_CONNECT_TIMEOUT + (AT_OFF ? 0 : at_min) \
843 + PING_INTERVAL)
844 #define MGC_TARGET_REG_LIMIT 10
845 #define MGC_SEND_PARAM_LIMIT 10
846
847 /* Send parameter to MGS*/
848 static int mgc_set_mgs_param(struct obd_export *exp,
849 struct mgs_send_param *msp)
850 {
851 struct ptlrpc_request *req;
852 struct mgs_send_param *req_msp, *rep_msp;
853 int rc;
854
855 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
856 &RQF_MGS_SET_INFO, LUSTRE_MGS_VERSION,
857 MGS_SET_INFO);
858 if (!req)
859 return -ENOMEM;
860
861 req_msp = req_capsule_client_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
862 if (!req_msp) {
863 ptlrpc_req_finished(req);
864 return -ENOMEM;
865 }
866
867 memcpy(req_msp, msp, sizeof(*req_msp));
868 ptlrpc_request_set_replen(req);
869
870 /* Limit how long we will wait for the enqueue to complete */
871 req->rq_delay_limit = MGC_SEND_PARAM_LIMIT;
872 rc = ptlrpc_queue_wait(req);
873 if (!rc) {
874 rep_msp = req_capsule_server_get(&req->rq_pill, &RMF_MGS_SEND_PARAM);
875 memcpy(msp, rep_msp, sizeof(*rep_msp));
876 }
877
878 ptlrpc_req_finished(req);
879
880 return rc;
881 }
882
883 /* Take a config lock so we can get cancel notifications */
884 static int mgc_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm,
885 __u32 type, ldlm_policy_data_t *policy, __u32 mode,
886 __u64 *flags, void *bl_cb, void *cp_cb, void *gl_cb,
887 void *data, __u32 lvb_len, void *lvb_swabber,
888 struct lustre_handle *lockh)
889 {
890 struct config_llog_data *cld = data;
891 struct ldlm_enqueue_info einfo = {
892 .ei_type = type,
893 .ei_mode = mode,
894 .ei_cb_bl = mgc_blocking_ast,
895 .ei_cb_cp = ldlm_completion_ast,
896 };
897 struct ptlrpc_request *req;
898 int short_limit = cld_is_sptlrpc(cld);
899 int rc;
900
901 CDEBUG(D_MGC, "Enqueue for %s (res %#llx)\n", cld->cld_logname,
902 cld->cld_resid.name[0]);
903
904 /* We need a callback for every lockholder, so don't try to
905 * ldlm_lock_match (see rev 1.1.2.11.2.47)
906 */
907 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
908 &RQF_LDLM_ENQUEUE, LUSTRE_DLM_VERSION,
909 LDLM_ENQUEUE);
910 if (!req)
911 return -ENOMEM;
912
913 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, 0);
914 ptlrpc_request_set_replen(req);
915
916 /* Limit how long we will wait for the enqueue to complete */
917 req->rq_delay_limit = short_limit ? 5 : MGC_ENQUEUE_LIMIT;
918 rc = ldlm_cli_enqueue(exp, &req, &einfo, &cld->cld_resid, NULL, flags,
919 NULL, 0, LVB_T_NONE, lockh, 0);
920 /* A failed enqueue should still call the mgc_blocking_ast,
921 * where it will be requeued if needed ("grant failed").
922 */
923 ptlrpc_req_finished(req);
924 return rc;
925 }
926
927 static void mgc_notify_active(struct obd_device *unused)
928 {
929 /* wakeup mgc_requeue_thread to requeue mgc lock */
930 spin_lock(&config_list_lock);
931 rq_state |= RQ_NOW;
932 spin_unlock(&config_list_lock);
933 wake_up(&rq_waitq);
934
935 /* TODO: Help the MGS rebuild nidtbl. -jay */
936 }
937
938 /* Send target_reg message to MGS */
939 static int mgc_target_register(struct obd_export *exp,
940 struct mgs_target_info *mti)
941 {
942 struct ptlrpc_request *req;
943 struct mgs_target_info *req_mti, *rep_mti;
944 int rc;
945
946 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
947 &RQF_MGS_TARGET_REG, LUSTRE_MGS_VERSION,
948 MGS_TARGET_REG);
949 if (!req)
950 return -ENOMEM;
951
952 req_mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
953 if (!req_mti) {
954 ptlrpc_req_finished(req);
955 return -ENOMEM;
956 }
957
958 memcpy(req_mti, mti, sizeof(*req_mti));
959 ptlrpc_request_set_replen(req);
960 CDEBUG(D_MGC, "register %s\n", mti->mti_svname);
961 /* Limit how long we will wait for the enqueue to complete */
962 req->rq_delay_limit = MGC_TARGET_REG_LIMIT;
963
964 rc = ptlrpc_queue_wait(req);
965 if (!rc) {
966 rep_mti = req_capsule_server_get(&req->rq_pill,
967 &RMF_MGS_TARGET_INFO);
968 memcpy(mti, rep_mti, sizeof(*rep_mti));
969 CDEBUG(D_MGC, "register %s got index = %d\n",
970 mti->mti_svname, mti->mti_stripe_index);
971 }
972 ptlrpc_req_finished(req);
973
974 return rc;
975 }
976
977 static int mgc_set_info_async(const struct lu_env *env, struct obd_export *exp,
978 u32 keylen, void *key, u32 vallen,
979 void *val, struct ptlrpc_request_set *set)
980 {
981 int rc = -EINVAL;
982
983 /* Turn off initial_recov after we try all backup servers once */
984 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
985 struct obd_import *imp = class_exp2cliimp(exp);
986 int value;
987
988 if (vallen != sizeof(int))
989 return -EINVAL;
990 value = *(int *)val;
991 CDEBUG(D_MGC, "InitRecov %s %d/d%d:i%d:r%d:or%d:%s\n",
992 imp->imp_obd->obd_name, value,
993 imp->imp_deactive, imp->imp_invalid,
994 imp->imp_replayable, imp->imp_obd->obd_replayable,
995 ptlrpc_import_state_name(imp->imp_state));
996 /* Resurrect if we previously died */
997 if ((imp->imp_state != LUSTRE_IMP_FULL &&
998 imp->imp_state != LUSTRE_IMP_NEW) || value > 1)
999 ptlrpc_reconnect_import(imp);
1000 return 0;
1001 }
1002 if (KEY_IS(KEY_SET_INFO)) {
1003 struct mgs_send_param *msp;
1004
1005 msp = val;
1006 rc = mgc_set_mgs_param(exp, msp);
1007 return rc;
1008 }
1009 if (KEY_IS(KEY_MGSSEC)) {
1010 struct client_obd *cli = &exp->exp_obd->u.cli;
1011 struct sptlrpc_flavor flvr;
1012
1013 /*
1014 * empty string means using current flavor, if which haven't
1015 * been set yet, set it as null.
1016 *
1017 * if flavor has been set previously, check the asking flavor
1018 * must match the existing one.
1019 */
1020 if (vallen == 0) {
1021 if (cli->cl_flvr_mgc.sf_rpc != SPTLRPC_FLVR_INVALID)
1022 return 0;
1023 val = "null";
1024 vallen = 4;
1025 }
1026
1027 rc = sptlrpc_parse_flavor(val, &flvr);
1028 if (rc) {
1029 CERROR("invalid sptlrpc flavor %s to MGS\n",
1030 (char *) val);
1031 return rc;
1032 }
1033
1034 /*
1035 * caller already hold a mutex
1036 */
1037 if (cli->cl_flvr_mgc.sf_rpc == SPTLRPC_FLVR_INVALID) {
1038 cli->cl_flvr_mgc = flvr;
1039 } else if (memcmp(&cli->cl_flvr_mgc, &flvr,
1040 sizeof(flvr)) != 0) {
1041 char str[20];
1042
1043 sptlrpc_flavor2name(&cli->cl_flvr_mgc,
1044 str, sizeof(str));
1045 LCONSOLE_ERROR("asking sptlrpc flavor %s to MGS but currently %s is in use\n",
1046 (char *) val, str);
1047 rc = -EPERM;
1048 }
1049 return rc;
1050 }
1051
1052 return rc;
1053 }
1054
1055 static int mgc_get_info(const struct lu_env *env, struct obd_export *exp,
1056 __u32 keylen, void *key, __u32 *vallen, void *val,
1057 struct lov_stripe_md *unused)
1058 {
1059 int rc = -EINVAL;
1060
1061 if (KEY_IS(KEY_CONN_DATA)) {
1062 struct obd_import *imp = class_exp2cliimp(exp);
1063 struct obd_connect_data *data = val;
1064
1065 if (*vallen == sizeof(*data)) {
1066 *data = imp->imp_connect_data;
1067 rc = 0;
1068 }
1069 }
1070
1071 return rc;
1072 }
1073
1074 static int mgc_import_event(struct obd_device *obd,
1075 struct obd_import *imp,
1076 enum obd_import_event event)
1077 {
1078 LASSERT(imp->imp_obd == obd);
1079 CDEBUG(D_MGC, "import event %#x\n", event);
1080
1081 switch (event) {
1082 case IMP_EVENT_DISCON:
1083 /* MGC imports should not wait for recovery */
1084 if (OCD_HAS_FLAG(&imp->imp_connect_data, IMP_RECOV))
1085 ptlrpc_pinger_ir_down();
1086 break;
1087 case IMP_EVENT_INACTIVE:
1088 break;
1089 case IMP_EVENT_INVALIDATE: {
1090 struct ldlm_namespace *ns = obd->obd_namespace;
1091
1092 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1093 break;
1094 }
1095 case IMP_EVENT_ACTIVE:
1096 CDEBUG(D_INFO, "%s: Reactivating import\n", obd->obd_name);
1097 /* Clearing obd_no_recov allows us to continue pinging */
1098 obd->obd_no_recov = 0;
1099 mgc_notify_active(obd);
1100 if (OCD_HAS_FLAG(&imp->imp_connect_data, IMP_RECOV))
1101 ptlrpc_pinger_ir_up();
1102 break;
1103 case IMP_EVENT_OCD:
1104 break;
1105 case IMP_EVENT_DEACTIVATE:
1106 case IMP_EVENT_ACTIVATE:
1107 break;
1108 default:
1109 CERROR("Unknown import event %#x\n", event);
1110 LBUG();
1111 }
1112 return 0;
1113 }
1114
1115 enum {
1116 CONFIG_READ_NRPAGES_INIT = 1 << (20 - PAGE_SHIFT),
1117 CONFIG_READ_NRPAGES = 4
1118 };
1119
1120 static int mgc_apply_recover_logs(struct obd_device *mgc,
1121 struct config_llog_data *cld,
1122 __u64 max_version,
1123 void *data, int datalen, bool mne_swab)
1124 {
1125 struct config_llog_instance *cfg = &cld->cld_cfg;
1126 struct mgs_nidtbl_entry *entry;
1127 struct lustre_cfg *lcfg;
1128 struct lustre_cfg_bufs bufs;
1129 u64 prev_version = 0;
1130 char *inst;
1131 char *buf;
1132 int bufsz;
1133 int pos;
1134 int rc = 0;
1135 int off = 0;
1136
1137 LASSERT(cfg->cfg_instance);
1138 LASSERT(cfg->cfg_sb == cfg->cfg_instance);
1139
1140 inst = kzalloc(PAGE_SIZE, GFP_KERNEL);
1141 if (!inst)
1142 return -ENOMEM;
1143
1144 pos = snprintf(inst, PAGE_SIZE, "%p", cfg->cfg_instance);
1145 if (pos >= PAGE_SIZE) {
1146 kfree(inst);
1147 return -E2BIG;
1148 }
1149
1150 ++pos;
1151 buf = inst + pos;
1152 bufsz = PAGE_SIZE - pos;
1153
1154 while (datalen > 0) {
1155 int entry_len = sizeof(*entry);
1156 int is_ost;
1157 struct obd_device *obd;
1158 char *obdname;
1159 char *cname;
1160 char *params;
1161 char *uuid;
1162
1163 rc = -EINVAL;
1164 if (datalen < sizeof(*entry))
1165 break;
1166
1167 entry = (typeof(entry))(data + off);
1168
1169 /* sanity check */
1170 if (entry->mne_nid_type != 0) /* only support type 0 for ipv4 */
1171 break;
1172 if (entry->mne_nid_count == 0) /* at least one nid entry */
1173 break;
1174 if (entry->mne_nid_size != sizeof(lnet_nid_t))
1175 break;
1176
1177 entry_len += entry->mne_nid_count * entry->mne_nid_size;
1178 if (datalen < entry_len) /* must have entry_len at least */
1179 break;
1180
1181 /* Keep this swab for normal mixed endian handling. LU-1644 */
1182 if (mne_swab)
1183 lustre_swab_mgs_nidtbl_entry(entry);
1184 if (entry->mne_length > PAGE_SIZE) {
1185 CERROR("MNE too large (%u)\n", entry->mne_length);
1186 break;
1187 }
1188
1189 if (entry->mne_length < entry_len)
1190 break;
1191
1192 off += entry->mne_length;
1193 datalen -= entry->mne_length;
1194 if (datalen < 0)
1195 break;
1196
1197 if (entry->mne_version > max_version) {
1198 CERROR("entry index(%lld) is over max_index(%lld)\n",
1199 entry->mne_version, max_version);
1200 break;
1201 }
1202
1203 if (prev_version >= entry->mne_version) {
1204 CERROR("index unsorted, prev %lld, now %lld\n",
1205 prev_version, entry->mne_version);
1206 break;
1207 }
1208 prev_version = entry->mne_version;
1209
1210 /*
1211 * Write a string with format "nid::instance" to
1212 * lustre/<osc|mdc>/<target>-<osc|mdc>-<instance>/import.
1213 */
1214
1215 is_ost = entry->mne_type == LDD_F_SV_TYPE_OST;
1216 memset(buf, 0, bufsz);
1217 obdname = buf;
1218 pos = 0;
1219
1220 /* lustre-OST0001-osc-<instance #> */
1221 strcpy(obdname, cld->cld_logname);
1222 cname = strrchr(obdname, '-');
1223 if (!cname) {
1224 CERROR("mgc %s: invalid logname %s\n",
1225 mgc->obd_name, obdname);
1226 break;
1227 }
1228
1229 pos = cname - obdname;
1230 obdname[pos] = 0;
1231 pos += sprintf(obdname + pos, "-%s%04x",
1232 is_ost ? "OST" : "MDT", entry->mne_index);
1233
1234 cname = is_ost ? "osc" : "mdc";
1235 pos += sprintf(obdname + pos, "-%s-%s", cname, inst);
1236 lustre_cfg_bufs_reset(&bufs, obdname);
1237
1238 /* find the obd by obdname */
1239 obd = class_name2obd(obdname);
1240 if (!obd) {
1241 CDEBUG(D_INFO, "mgc %s: cannot find obdname %s\n",
1242 mgc->obd_name, obdname);
1243 rc = 0;
1244 /* this is a safe race, when the ost is starting up...*/
1245 continue;
1246 }
1247
1248 /* osc.import = "connection=<Conn UUID>::<target instance>" */
1249 ++pos;
1250 params = buf + pos;
1251 pos += sprintf(params, "%s.import=%s", cname, "connection=");
1252 uuid = buf + pos;
1253
1254 down_read(&obd->u.cli.cl_sem);
1255 if (!obd->u.cli.cl_import) {
1256 /* client does not connect to the OST yet */
1257 up_read(&obd->u.cli.cl_sem);
1258 rc = 0;
1259 continue;
1260 }
1261
1262 /* TODO: iterate all nids to find one */
1263 /* find uuid by nid */
1264 rc = client_import_find_conn(obd->u.cli.cl_import,
1265 entry->u.nids[0],
1266 (struct obd_uuid *)uuid);
1267 up_read(&obd->u.cli.cl_sem);
1268 if (rc < 0) {
1269 CERROR("mgc: cannot find uuid by nid %s\n",
1270 libcfs_nid2str(entry->u.nids[0]));
1271 break;
1272 }
1273
1274 CDEBUG(D_INFO, "Find uuid %s by nid %s\n",
1275 uuid, libcfs_nid2str(entry->u.nids[0]));
1276
1277 pos += strlen(uuid);
1278 pos += sprintf(buf + pos, "::%u", entry->mne_instance);
1279 LASSERT(pos < bufsz);
1280
1281 lustre_cfg_bufs_set_string(&bufs, 1, params);
1282
1283 rc = -ENOMEM;
1284 lcfg = lustre_cfg_new(LCFG_PARAM, &bufs);
1285 if (IS_ERR(lcfg)) {
1286 CERROR("mgc: cannot allocate memory\n");
1287 break;
1288 }
1289
1290 CDEBUG(D_INFO, "ir apply logs %lld/%lld for %s -> %s\n",
1291 prev_version, max_version, obdname, params);
1292
1293 rc = class_process_config(lcfg);
1294 lustre_cfg_free(lcfg);
1295 if (rc)
1296 CDEBUG(D_INFO, "process config for %s error %d\n",
1297 obdname, rc);
1298
1299 /* continue, even one with error */
1300 }
1301
1302 kfree(inst);
1303 return rc;
1304 }
1305
1306 /**
1307 * This function is called if this client was notified for target restarting
1308 * by the MGS. A CONFIG_READ RPC is going to send to fetch recovery logs.
1309 */
1310 static int mgc_process_recover_log(struct obd_device *obd,
1311 struct config_llog_data *cld)
1312 {
1313 struct ptlrpc_request *req = NULL;
1314 struct config_llog_instance *cfg = &cld->cld_cfg;
1315 struct mgs_config_body *body;
1316 struct mgs_config_res *res;
1317 struct ptlrpc_bulk_desc *desc;
1318 struct page **pages;
1319 int nrpages;
1320 bool eof = true;
1321 bool mne_swab;
1322 int i;
1323 int ealen;
1324 int rc;
1325
1326 /* allocate buffer for bulk transfer.
1327 * if this is the first time for this mgs to read logs,
1328 * CONFIG_READ_NRPAGES_INIT will be used since it will read all logs
1329 * once; otherwise, it only reads increment of logs, this should be
1330 * small and CONFIG_READ_NRPAGES will be used.
1331 */
1332 nrpages = CONFIG_READ_NRPAGES;
1333 if (cfg->cfg_last_idx == 0) /* the first time */
1334 nrpages = CONFIG_READ_NRPAGES_INIT;
1335
1336 pages = kcalloc(nrpages, sizeof(*pages), GFP_KERNEL);
1337 if (!pages) {
1338 rc = -ENOMEM;
1339 goto out;
1340 }
1341
1342 for (i = 0; i < nrpages; i++) {
1343 pages[i] = alloc_page(GFP_KERNEL);
1344 if (!pages[i]) {
1345 rc = -ENOMEM;
1346 goto out;
1347 }
1348 }
1349
1350 again:
1351 LASSERT(cld_is_recover(cld));
1352 LASSERT(mutex_is_locked(&cld->cld_lock));
1353 req = ptlrpc_request_alloc(class_exp2cliimp(cld->cld_mgcexp),
1354 &RQF_MGS_CONFIG_READ);
1355 if (!req) {
1356 rc = -ENOMEM;
1357 goto out;
1358 }
1359
1360 rc = ptlrpc_request_pack(req, LUSTRE_MGS_VERSION, MGS_CONFIG_READ);
1361 if (rc)
1362 goto out;
1363
1364 /* pack request */
1365 body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
1366 LASSERT(sizeof(body->mcb_name) > strlen(cld->cld_logname));
1367 if (strlcpy(body->mcb_name, cld->cld_logname, sizeof(body->mcb_name))
1368 >= sizeof(body->mcb_name)) {
1369 rc = -E2BIG;
1370 goto out;
1371 }
1372 body->mcb_offset = cfg->cfg_last_idx + 1;
1373 body->mcb_type = cld->cld_type;
1374 body->mcb_bits = PAGE_SHIFT;
1375 body->mcb_units = nrpages;
1376
1377 /* allocate bulk transfer descriptor */
1378 desc = ptlrpc_prep_bulk_imp(req, nrpages, 1, BULK_PUT_SINK,
1379 MGS_BULK_PORTAL);
1380 if (!desc) {
1381 rc = -ENOMEM;
1382 goto out;
1383 }
1384
1385 for (i = 0; i < nrpages; i++)
1386 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_SIZE);
1387
1388 ptlrpc_request_set_replen(req);
1389 rc = ptlrpc_queue_wait(req);
1390 if (rc)
1391 goto out;
1392
1393 res = req_capsule_server_get(&req->rq_pill, &RMF_MGS_CONFIG_RES);
1394 if (res->mcr_size < res->mcr_offset) {
1395 rc = -EINVAL;
1396 goto out;
1397 }
1398
1399 /* always update the index even though it might have errors with
1400 * handling the recover logs
1401 */
1402 cfg->cfg_last_idx = res->mcr_offset;
1403 eof = res->mcr_offset == res->mcr_size;
1404
1405 CDEBUG(D_INFO, "Latest version %lld, more %d.\n",
1406 res->mcr_offset, eof == false);
1407
1408 ealen = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, 0);
1409 if (ealen < 0) {
1410 rc = ealen;
1411 goto out;
1412 }
1413
1414 if (ealen > nrpages << PAGE_SHIFT) {
1415 rc = -EINVAL;
1416 goto out;
1417 }
1418
1419 if (ealen == 0) { /* no logs transferred */
1420 if (!eof)
1421 rc = -EINVAL;
1422 goto out;
1423 }
1424
1425 mne_swab = !!ptlrpc_rep_need_swab(req);
1426 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
1427 /* This import flag means the server did an extra swab of IR MNE
1428 * records (fixed in LU-1252), reverse it here if needed. LU-1644
1429 */
1430 if (unlikely(req->rq_import->imp_need_mne_swab))
1431 mne_swab = !mne_swab;
1432 #else
1433 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
1434 #endif
1435
1436 for (i = 0; i < nrpages && ealen > 0; i++) {
1437 int rc2;
1438 void *ptr;
1439
1440 ptr = kmap(pages[i]);
1441 rc2 = mgc_apply_recover_logs(obd, cld, res->mcr_offset, ptr,
1442 min_t(int, ealen, PAGE_SIZE),
1443 mne_swab);
1444 kunmap(pages[i]);
1445 if (rc2 < 0) {
1446 CWARN("Process recover log %s error %d\n",
1447 cld->cld_logname, rc2);
1448 break;
1449 }
1450
1451 ealen -= PAGE_SIZE;
1452 }
1453
1454 out:
1455 if (req)
1456 ptlrpc_req_finished(req);
1457
1458 if (rc == 0 && !eof)
1459 goto again;
1460
1461 if (pages) {
1462 for (i = 0; i < nrpages; i++) {
1463 if (!pages[i])
1464 break;
1465 __free_page(pages[i]);
1466 }
1467 kfree(pages);
1468 }
1469 return rc;
1470 }
1471
1472 /* local_only means it cannot get remote llogs */
1473 static int mgc_process_cfg_log(struct obd_device *mgc,
1474 struct config_llog_data *cld, int local_only)
1475 {
1476 struct llog_ctxt *ctxt;
1477 struct lustre_sb_info *lsi = NULL;
1478 int rc = 0;
1479 bool sptlrpc_started = false;
1480 struct lu_env *env;
1481
1482 LASSERT(cld);
1483 LASSERT(mutex_is_locked(&cld->cld_lock));
1484
1485 /*
1486 * local copy of sptlrpc log is controlled elsewhere, don't try to
1487 * read it up here.
1488 */
1489 if (cld_is_sptlrpc(cld) && local_only)
1490 return 0;
1491
1492 if (cld->cld_cfg.cfg_sb)
1493 lsi = s2lsi(cld->cld_cfg.cfg_sb);
1494
1495 env = kzalloc(sizeof(*env), GFP_KERNEL);
1496 if (!env)
1497 return -ENOMEM;
1498
1499 rc = lu_env_init(env, LCT_MG_THREAD);
1500 if (rc)
1501 goto out_free;
1502
1503 ctxt = llog_get_context(mgc, LLOG_CONFIG_REPL_CTXT);
1504 LASSERT(ctxt);
1505
1506 if (local_only) /* no local log at client side */ {
1507 rc = -EIO;
1508 goto out_pop;
1509 }
1510
1511 if (cld_is_sptlrpc(cld)) {
1512 sptlrpc_conf_log_update_begin(cld->cld_logname);
1513 sptlrpc_started = true;
1514 }
1515
1516 /* logname and instance info should be the same, so use our
1517 * copy of the instance for the update. The cfg_last_idx will
1518 * be updated here.
1519 */
1520 rc = class_config_parse_llog(env, ctxt, cld->cld_logname,
1521 &cld->cld_cfg);
1522
1523 out_pop:
1524 __llog_ctxt_put(env, ctxt);
1525
1526 /*
1527 * update settings on existing OBDs. doing it inside
1528 * of llog_process_lock so no device is attaching/detaching
1529 * in parallel.
1530 * the logname must be <fsname>-sptlrpc
1531 */
1532 if (sptlrpc_started) {
1533 LASSERT(cld_is_sptlrpc(cld));
1534 sptlrpc_conf_log_update_end(cld->cld_logname);
1535 class_notify_sptlrpc_conf(cld->cld_logname,
1536 strlen(cld->cld_logname) -
1537 strlen("-sptlrpc"));
1538 }
1539
1540 lu_env_fini(env);
1541 out_free:
1542 kfree(env);
1543 return rc;
1544 }
1545
1546 /** Get a config log from the MGS and process it.
1547 * This func is called for both clients and servers.
1548 * Copy the log locally before parsing it if appropriate (non-MGS server)
1549 */
1550 int mgc_process_log(struct obd_device *mgc, struct config_llog_data *cld)
1551 {
1552 struct lustre_handle lockh = { 0 };
1553 __u64 flags = LDLM_FL_NO_LRU;
1554 int rc = 0, rcl;
1555
1556 LASSERT(cld);
1557
1558 /* I don't want multiple processes running process_log at once --
1559 * sounds like badness. It actually might be fine, as long as
1560 * we're not trying to update from the same log
1561 * simultaneously (in which case we should use a per-log sem.)
1562 */
1563 mutex_lock(&cld->cld_lock);
1564 if (cld->cld_stopping) {
1565 mutex_unlock(&cld->cld_lock);
1566 return 0;
1567 }
1568
1569 OBD_FAIL_TIMEOUT(OBD_FAIL_MGC_PAUSE_PROCESS_LOG, 20);
1570
1571 CDEBUG(D_MGC, "Process log %s:%p from %d\n", cld->cld_logname,
1572 cld->cld_cfg.cfg_instance, cld->cld_cfg.cfg_last_idx + 1);
1573
1574 /* Get the cfg lock on the llog */
1575 rcl = mgc_enqueue(mgc->u.cli.cl_mgc_mgsexp, NULL, LDLM_PLAIN, NULL,
1576 LCK_CR, &flags, NULL, NULL, NULL,
1577 cld, 0, NULL, &lockh);
1578 if (rcl == 0) {
1579 /* Get the cld, it will be released in mgc_blocking_ast. */
1580 config_log_get(cld);
1581 rc = ldlm_lock_set_data(&lockh, (void *)cld);
1582 LASSERT(rc == 0);
1583 } else {
1584 CDEBUG(D_MGC, "Can't get cfg lock: %d\n", rcl);
1585
1586 /* mark cld_lostlock so that it will requeue
1587 * after MGC becomes available.
1588 */
1589 cld->cld_lostlock = 1;
1590 /* Get extra reference, it will be put in requeue thread */
1591 config_log_get(cld);
1592 }
1593
1594 if (cld_is_recover(cld)) {
1595 rc = 0; /* this is not a fatal error for recover log */
1596 if (rcl == 0)
1597 rc = mgc_process_recover_log(mgc, cld);
1598 } else {
1599 rc = mgc_process_cfg_log(mgc, cld, rcl != 0);
1600 }
1601
1602 CDEBUG(D_MGC, "%s: configuration from log '%s' %sed (%d).\n",
1603 mgc->obd_name, cld->cld_logname, rc ? "fail" : "succeed", rc);
1604
1605 mutex_unlock(&cld->cld_lock);
1606
1607 /* Now drop the lock so MGS can revoke it */
1608 if (!rcl)
1609 ldlm_lock_decref(&lockh, LCK_CR);
1610
1611 return rc;
1612 }
1613
1614 /** Called from lustre_process_log.
1615 * LCFG_LOG_START gets the config log from the MGS, processes it to start
1616 * any services, and adds it to the list logs to watch (follow).
1617 */
1618 static int mgc_process_config(struct obd_device *obd, u32 len, void *buf)
1619 {
1620 struct lustre_cfg *lcfg = buf;
1621 struct config_llog_instance *cfg = NULL;
1622 char *logname;
1623 int rc = 0;
1624
1625 switch (lcfg->lcfg_command) {
1626 case LCFG_LOV_ADD_OBD: {
1627 /* Overloading this cfg command: register a new target */
1628 struct mgs_target_info *mti;
1629
1630 if (LUSTRE_CFG_BUFLEN(lcfg, 1) !=
1631 sizeof(struct mgs_target_info)) {
1632 rc = -EINVAL;
1633 goto out;
1634 }
1635
1636 mti = (struct mgs_target_info *)lustre_cfg_buf(lcfg, 1);
1637 CDEBUG(D_MGC, "add_target %s %#x\n",
1638 mti->mti_svname, mti->mti_flags);
1639 rc = mgc_target_register(obd->u.cli.cl_mgc_mgsexp, mti);
1640 break;
1641 }
1642 case LCFG_LOV_DEL_OBD:
1643 /* Unregister has no meaning at the moment. */
1644 CERROR("lov_del_obd unimplemented\n");
1645 rc = -ENOSYS;
1646 break;
1647 case LCFG_SPTLRPC_CONF: {
1648 rc = sptlrpc_process_config(lcfg);
1649 break;
1650 }
1651 case LCFG_LOG_START: {
1652 struct config_llog_data *cld;
1653 struct super_block *sb;
1654
1655 logname = lustre_cfg_string(lcfg, 1);
1656 cfg = (struct config_llog_instance *)lustre_cfg_buf(lcfg, 2);
1657 sb = *(struct super_block **)lustre_cfg_buf(lcfg, 3);
1658
1659 CDEBUG(D_MGC, "parse_log %s from %d\n", logname,
1660 cfg->cfg_last_idx);
1661
1662 /* We're only called through here on the initial mount */
1663 rc = config_log_add(obd, logname, cfg, sb);
1664 if (rc)
1665 break;
1666 cld = config_log_find(logname, cfg);
1667 if (!cld) {
1668 rc = -ENOENT;
1669 break;
1670 }
1671
1672 /* COMPAT_146 */
1673 /* FIXME only set this for old logs! Right now this forces
1674 * us to always skip the "inside markers" check
1675 */
1676 cld->cld_cfg.cfg_flags |= CFG_F_COMPAT146;
1677
1678 rc = mgc_process_log(obd, cld);
1679 if (rc == 0 && cld->cld_recover) {
1680 if (OCD_HAS_FLAG(&obd->u.cli.cl_import->
1681 imp_connect_data, IMP_RECOV)) {
1682 rc = mgc_process_log(obd, cld->cld_recover);
1683 } else {
1684 struct config_llog_data *cir = cld->cld_recover;
1685
1686 cld->cld_recover = NULL;
1687 config_log_put(cir);
1688 }
1689 if (rc)
1690 CERROR("Cannot process recover llog %d\n", rc);
1691 }
1692
1693 if (rc == 0 && cld->cld_params) {
1694 rc = mgc_process_log(obd, cld->cld_params);
1695 if (rc == -ENOENT) {
1696 CDEBUG(D_MGC,
1697 "There is no params config file yet\n");
1698 rc = 0;
1699 }
1700 /* params log is optional */
1701 if (rc)
1702 CERROR(
1703 "%s: can't process params llog: rc = %d\n",
1704 obd->obd_name, rc);
1705 }
1706 config_log_put(cld);
1707
1708 break;
1709 }
1710 case LCFG_LOG_END: {
1711 logname = lustre_cfg_string(lcfg, 1);
1712
1713 if (lcfg->lcfg_bufcount >= 2)
1714 cfg = (struct config_llog_instance *)lustre_cfg_buf(
1715 lcfg, 2);
1716 rc = config_log_end(logname, cfg);
1717 break;
1718 }
1719 default: {
1720 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1721 rc = -EINVAL;
1722 goto out;
1723
1724 }
1725 }
1726 out:
1727 return rc;
1728 }
1729
1730 static struct obd_ops mgc_obd_ops = {
1731 .owner = THIS_MODULE,
1732 .setup = mgc_setup,
1733 .precleanup = mgc_precleanup,
1734 .cleanup = mgc_cleanup,
1735 .add_conn = client_import_add_conn,
1736 .del_conn = client_import_del_conn,
1737 .connect = client_connect_import,
1738 .disconnect = client_disconnect_export,
1739 /* .enqueue = mgc_enqueue, */
1740 /* .iocontrol = mgc_iocontrol, */
1741 .set_info_async = mgc_set_info_async,
1742 .get_info = mgc_get_info,
1743 .import_event = mgc_import_event,
1744 .process_config = mgc_process_config,
1745 };
1746
1747 static int __init mgc_init(void)
1748 {
1749 return class_register_type(&mgc_obd_ops, NULL,
1750 LUSTRE_MGC_NAME, NULL);
1751 }
1752
1753 static void /*__exit*/ mgc_exit(void)
1754 {
1755 class_unregister_type(LUSTRE_MGC_NAME);
1756 }
1757
1758 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1759 MODULE_DESCRIPTION("Lustre Management Client");
1760 MODULE_VERSION(LUSTRE_VERSION_STRING);
1761 MODULE_LICENSE("GPL");
1762
1763 module_init(mgc_init);
1764 module_exit(mgc_exit);
Linux kernel - Deliverables
This page took 0.067127 seconds and 5 git commands to generate.