4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
39 * Author: Mike Pershin <tappro@whamcloud.com>
40 * Author: Johann Lombardi <johann@whamcloud.com>
43 #define DEBUG_SUBSYSTEM S_FILTER
45 #include <obd_class.h>
46 #include <lustre_param.h>
48 #include "ofd_internal.h"
50 /* Slab for OFD object allocation */
51 static cfs_mem_cache_t *ofd_object_kmem;
53 static struct lu_kmem_descr ofd_caches[] = {
55 .ckd_cache = &ofd_object_kmem,
56 .ckd_name = "ofd_obj",
57 .ckd_size = sizeof(struct ofd_object)
64 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
65 const char *next, struct obd_export **exp)
67 struct obd_connect_data *data = NULL;
68 struct obd_device *obd;
74 GOTO(out, rc = -ENOMEM);
76 obd = class_name2obd(next);
78 CERROR("%s: can't locate next device: %s\n",
79 m->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name, next);
80 GOTO(out, rc = -ENOTCONN);
83 data->ocd_connect_flags = OBD_CONNECT_VERSION;
84 data->ocd_version = LUSTRE_VERSION_CODE;
86 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
88 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
89 m->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name, next, rc);
93 m->ofd_dt_dev.dd_lu_dev.ld_site =
94 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
95 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
96 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
97 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
105 static int ofd_stack_init(const struct lu_env *env,
106 struct ofd_device *m, struct lustre_cfg *cfg)
108 const char *dev = lustre_cfg_string(cfg, 0);
110 struct ofd_thread_info *info = ofd_info(env);
111 struct lustre_mount_info *lmi;
117 lmi = server_get_mount(dev);
119 CERROR("Cannot get mount info for %s!\n", dev);
123 /* find bottom osd */
124 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
128 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
129 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
130 OBD_FREE(osdname, MTI_NAME_MAXLEN);
134 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
136 m->ofd_osd = lu2dt_dev(d);
138 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
139 "%s-osd", lustre_cfg_string(cfg, 0));
144 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
145 struct lu_device *top)
147 struct obd_device *obd = ofd_obd(m);
148 struct lustre_cfg_bufs bufs;
149 struct lustre_cfg *lcfg;
154 lu_site_purge(env, top->ld_site, ~0);
156 /* process cleanup, pass mdt obd name to get obd umount flags */
157 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
162 lustre_cfg_bufs_set_string(&bufs, 1, flags);
163 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
165 CERROR("Cannot alloc lcfg!\n");
170 top->ld_ops->ldo_process_config(env, top, lcfg);
171 lustre_cfg_free(lcfg);
173 lu_site_purge(env, top->ld_site, ~0);
175 LASSERT(m->ofd_osd_exp);
176 obd_disconnect(m->ofd_osd_exp);
182 /* used by MGS to process specific configurations */
183 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
184 struct lustre_cfg *cfg)
186 struct ofd_device *m = ofd_dev(d);
187 struct dt_device *dt_next = m->ofd_osd;
188 struct lu_device *next = &dt_next->dd_lu_dev;
193 switch (cfg->lcfg_command) {
195 struct lprocfs_static_vars lvars;
197 lprocfs_ofd_init_vars(&lvars);
198 rc = class_process_proc_param(PARAM_OST, lvars.obd_vars, cfg,
200 if (rc > 0 || rc == -ENOSYS)
201 /* we don't understand; pass it on */
202 rc = next->ld_ops->ldo_process_config(env, next, cfg);
205 case LCFG_SPTLRPC_CONF: {
210 /* others are passed further */
211 rc = next->ld_ops->ldo_process_config(env, next, cfg);
217 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
218 const struct lu_object_conf *conf)
220 struct ofd_device *d = ofd_dev(o->lo_dev);
221 struct lu_device *under;
222 struct lu_object *below;
227 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
228 PFID(lu_object_fid(o)));
230 under = &d->ofd_osd->dd_lu_dev;
231 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
233 lu_object_add(o, below);
240 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
242 struct ofd_object *of = ofd_obj(o);
243 struct lu_object_header *h;
248 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
249 PFID(lu_object_fid(o)));
252 lu_object_header_fini(h);
253 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
257 static int ofd_object_print(const struct lu_env *env, void *cookie,
258 lu_printer_t p, const struct lu_object *o)
260 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
263 struct lu_object_operations ofd_obj_ops = {
264 .loo_object_init = ofd_object_init,
265 .loo_object_free = ofd_object_free,
266 .loo_object_print = ofd_object_print
269 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
270 const struct lu_object_header *hdr,
273 struct ofd_object *of;
277 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, CFS_ALLOC_IO);
280 struct lu_object_header *h;
282 o = &of->ofo_obj.do_lu;
284 lu_object_header_init(h);
285 lu_object_init(o, h, d);
286 lu_object_add_top(h, o);
287 o->lo_ops = &ofd_obj_ops;
294 extern int ost_handle(struct ptlrpc_request *req);
296 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
297 struct lu_device *dev)
299 struct ofd_thread_info *info;
300 struct ofd_device *ofd = ofd_dev(dev);
301 struct obd_device *obd = ofd_obd(ofd);
302 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
307 rc = lu_env_refill((struct lu_env *)env);
309 CERROR("Failure to refill session: '%d'\n", rc);
313 info = ofd_info_init(env, NULL);
317 /* initialize lower device */
318 rc = next->ld_ops->ldo_prepare(env, dev, next);
320 target_recovery_init(&ofd->ofd_lut, ost_handle);
322 if (obd->obd_recovering == 0)
323 ofd_postrecov(env, ofd);
328 static int ofd_recovery_complete(const struct lu_env *env,
329 struct lu_device *dev)
331 struct ofd_device *ofd = ofd_dev(dev);
332 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
337 /* Grant space for object precreation on the self export.
338 * This initial reserved space (i.e. 20MB for zfs and 560KB for ldiskfs)
339 * is enough to create 20k objects. It is then adapted based on the
340 * precreate request size (see ofd_grant_create()
342 ofd_grant_connect(env, dev->ld_obd->obd_self_export,
343 OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace);
344 rc = next->ld_ops->ldo_recovery_complete(env, next);
348 static struct lu_device_operations ofd_lu_ops = {
349 .ldo_object_alloc = ofd_object_alloc,
350 .ldo_process_config = ofd_process_config,
351 .ldo_recovery_complete = ofd_recovery_complete,
352 .ldo_prepare = ofd_prepare,
355 static int ofd_procfs_init(struct ofd_device *ofd)
357 struct lprocfs_static_vars lvars;
358 struct obd_device *obd = ofd_obd(ofd);
359 cfs_proc_dir_entry_t *entry;
364 /* lprocfs must be setup before the ofd so state can be safely added
365 * to /proc incrementally as the ofd is setup */
366 lprocfs_ofd_init_vars(&lvars);
367 rc = lprocfs_obd_setup(obd, lvars.obd_vars);
369 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
374 rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_LAST);
376 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
378 GOTO(obd_cleanup, rc);
381 /* Init OFD private stats here */
382 lprocfs_counter_init(obd->obd_stats, LPROC_OFD_READ_BYTES,
383 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
384 lprocfs_counter_init(obd->obd_stats, LPROC_OFD_WRITE_BYTES,
385 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
387 rc = lproc_ofd_attach_seqstat(obd);
389 CERROR("%s: create seqstat failed: %d.\n", obd->obd_name, rc);
390 GOTO(free_obd_stats, rc);
393 entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
396 CERROR("%s: error %d setting up lprocfs for %s\n",
397 obd->obd_name, rc, "exports");
398 GOTO(free_obd_stats, rc);
400 obd->obd_proc_exports_entry = entry;
402 entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
403 lprocfs_nid_stats_clear_read,
404 lprocfs_nid_stats_clear_write, obd, NULL);
407 CERROR("%s: add proc entry 'clear' failed: %d.\n",
409 GOTO(free_obd_stats, rc);
412 rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
413 ofd_stats_counter_init);
415 GOTO(remove_entry_clear, rc);
418 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
420 lprocfs_free_obd_stats(obd);
422 lprocfs_obd_cleanup(obd);
426 static int ofd_procfs_fini(struct ofd_device *ofd)
428 struct obd_device *obd = ofd_obd(ofd);
430 lprocfs_job_stats_fini(obd);
431 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
432 lprocfs_free_per_client_stats(obd);
433 lprocfs_free_obd_stats(obd);
434 lprocfs_obd_cleanup(obd);
438 extern int ost_handle(struct ptlrpc_request *req);
440 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
441 struct lu_device_type *ldt, struct lustre_cfg *cfg)
443 const char *dev = lustre_cfg_string(cfg, 0);
444 struct ofd_thread_info *info = NULL;
445 struct obd_device *obd;
446 struct obd_statfs *osfs;
451 obd = class_name2obd(dev);
453 CERROR("Cannot find obd with name %s\n", dev);
457 rc = lu_env_refill((struct lu_env *)env);
461 obd->u.obt.obt_magic = OBT_MAGIC;
463 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
464 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
466 cfs_spin_lock_init(&m->ofd_flags_lock);
467 m->ofd_raid_degraded = 0;
468 m->ofd_syncjournal = 0;
470 m->ofd_grant_compat_disable = 0;
473 cfs_spin_lock_init(&m->ofd_osfs_lock);
474 m->ofd_osfs_age = cfs_time_shift_64(-1000);
475 m->ofd_osfs_unstable = 0;
476 m->ofd_statfs_inflight = 0;
477 m->ofd_osfs_inflight = 0;
480 cfs_spin_lock_init(&m->ofd_grant_lock);
481 m->ofd_tot_dirty = 0;
482 m->ofd_tot_granted = 0;
483 m->ofd_tot_pending = 0;
484 m->ofd_max_group = 0;
486 cfs_rwlock_init(&obd->u.filter.fo_sptlrpc_lock);
487 sptlrpc_rule_set_init(&obd->u.filter.fo_sptlrpc_rset);
489 obd->u.filter.fo_fl_oss_capa = 0;
490 CFS_INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
491 obd->u.filter.fo_capa_hash = init_capa_hash();
492 if (obd->u.filter.fo_capa_hash == NULL)
495 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
496 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
497 /* set this lu_device to obd, because error handling need it */
498 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
500 rc = ofd_procfs_init(m);
502 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
506 /* No connection accepted until configurations will finish */
507 obd->obd_no_conn = 1;
508 obd->obd_replayable = 1;
509 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
510 char *str = lustre_cfg_string(cfg, 4);
512 if (strchr(str, 'n')) {
513 CWARN("%s: recovery disabled\n", obd->obd_name);
514 obd->obd_replayable = 0;
518 info = ofd_info_init(env, NULL);
522 rc = ofd_stack_init(env, m, cfg);
524 CERROR("Can't init device stack, rc %d\n", rc);
525 GOTO(err_fini_proc, rc);
528 /* populate cached statfs data */
529 osfs = &ofd_info(env)->fti_u.osfs;
530 rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
532 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
533 GOTO(err_fini_stack, rc);
535 if (!IS_PO2(osfs->os_bsize)) {
536 CERROR("%s: blocksize (%d) is not a power of 2\n",
537 obd->obd_name, osfs->os_bsize);
538 GOTO(err_fini_stack, rc = -EPROTO);
540 m->ofd_blockbits = cfs_fls(osfs->os_bsize) - 1;
542 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "filter-%p", m);
543 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
544 LDLM_NAMESPACE_SERVER,
545 LDLM_NAMESPACE_GREEDY,
547 if (m->ofd_namespace == NULL)
548 GOTO(err_fini_stack, rc = -ENOMEM);
549 /* set obd_namespace for compatibility with old code */
550 obd->obd_namespace = m->ofd_namespace;
551 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
552 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
553 m->ofd_namespace->ns_lvbp = m;
555 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
556 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
558 dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
560 /* Allow at most ddp_grant_reserved% of the available filesystem space
561 * to be granted to clients, so that any errors in the grant overhead
562 * calculations do not allow granting more space to clients than can be
563 * written. Assumes that in aggregate the grant overhead calculations do
564 * not have more than ddp_grant_reserved% estimation error in them. */
566 ofd_grant_ratio_conv(m->ofd_dt_conf.ddp_grant_reserved);
568 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd);
570 GOTO(err_free_ns, rc);
572 rc = ofd_fs_setup(env, m, obd);
574 GOTO(err_fini_lut, rc);
578 tgt_fini(env, &m->ofd_lut);
580 ldlm_namespace_free(m->ofd_namespace, 0, obd->obd_force);
581 obd->obd_namespace = m->ofd_namespace = NULL;
583 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
589 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
591 struct obd_device *obd = ofd_obd(m);
592 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
594 target_recovery_fini(obd);
595 obd_exports_barrier(obd);
596 obd_zombie_barrier();
598 tgt_fini(env, &m->ofd_lut);
599 ofd_fs_cleanup(env, m);
601 ofd_free_capa_keys(m);
602 cleanup_capa_hash(obd->u.filter.fo_capa_hash);
604 if (m->ofd_namespace != NULL) {
605 ldlm_namespace_free(m->ofd_namespace, NULL,
606 d->ld_obd->obd_force);
607 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
610 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
612 LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
613 server_put_mount(obd->obd_name, NULL);
617 static struct lu_device *ofd_device_fini(const struct lu_env *env,
621 ofd_fini(env, ofd_dev(d));
625 static struct lu_device *ofd_device_free(const struct lu_env *env,
628 struct ofd_device *m = ofd_dev(d);
630 dt_device_fini(&m->ofd_dt_dev);
635 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
636 struct lu_device_type *t,
637 struct lustre_cfg *cfg)
639 struct ofd_device *m;
645 return ERR_PTR(-ENOMEM);
647 l = &m->ofd_dt_dev.dd_lu_dev;
648 dt_device_init(&m->ofd_dt_dev, t);
649 rc = ofd_init0(env, m, t, cfg);
651 ofd_device_free(env, l);
658 /* thread context key constructor/destructor */
659 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
661 static void ofd_key_exit(const struct lu_context *ctx,
662 struct lu_context_key *key, void *data)
664 struct ofd_thread_info *info = data;
666 info->fti_env = NULL;
667 info->fti_exp = NULL;
670 info->fti_transno = 0;
671 info->fti_pre_version = 0;
672 info->fti_obj = NULL;
673 info->fti_has_trans = 0;
674 info->fti_mult_trans = 0;
677 memset(&info->fti_attr, 0, sizeof info->fti_attr);
680 struct lu_context_key ofd_thread_key = {
681 .lct_tags = LCT_DT_THREAD,
682 .lct_init = ofd_key_init,
683 .lct_fini = ofd_key_fini,
684 .lct_exit = ofd_key_exit
687 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
688 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
690 static struct lu_device_type_operations ofd_device_type_ops = {
691 .ldto_init = ofd_type_init,
692 .ldto_fini = ofd_type_fini,
694 .ldto_start = ofd_type_start,
695 .ldto_stop = ofd_type_stop,
697 .ldto_device_alloc = ofd_device_alloc,
698 .ldto_device_free = ofd_device_free,
699 .ldto_device_fini = ofd_device_fini
702 static struct lu_device_type ofd_device_type = {
703 .ldt_tags = LU_DEVICE_DT,
704 .ldt_name = LUSTRE_OST_NAME,
705 .ldt_ops = &ofd_device_type_ops,
706 .ldt_ctx_tags = LCT_DT_THREAD
709 int __init ofd_init(void)
711 struct lprocfs_static_vars lvars;
714 rc = lu_kmem_init(ofd_caches);
720 lu_kmem_fini(ofd_caches);
724 lprocfs_ofd_init_vars(&lvars);
726 rc = class_register_type(&ofd_obd_ops, NULL, lvars.module_vars,
727 LUSTRE_OST_NAME, &ofd_device_type);
731 void __exit ofd_exit(void)
734 lu_kmem_fini(ofd_caches);
735 class_unregister_type(LUSTRE_OST_NAME);
738 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
739 MODULE_DESCRIPTION("Lustre Object Filtering Device");
740 MODULE_LICENSE("GPL");
742 module_init(ofd_init);
743 module_exit(ofd_exit);