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_stack_init(const struct lu_env *env,
65 struct ofd_device *m, struct lustre_cfg *cfg)
67 struct lu_device *ofd_lu = &m->ofd_dt_dev.dd_lu_dev;
68 const char *dev = lustre_cfg_string(cfg, 0);
69 struct obd_type *type;
70 struct lu_device_type *ldt;
72 struct ofd_thread_info *info = ofd_info(env);
77 /* XXX: we should be able to use different OSDs here */
78 type = class_get_type(LUSTRE_OSD_NAME);
80 CERROR("Unknown type: '%s'\n", LUSTRE_OSD_NAME);
84 rc = lu_env_refill((struct lu_env *)env);
86 CERROR("Failure to refill session: '%d'\n", rc);
92 CERROR("type: '%s'\n", LUSTRE_OSD_NAME);
93 GOTO(out_type, rc = -EINVAL);
96 ldt->ldt_obd_type = type;
97 d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
99 CERROR("Cannot allocate device: '%s'\n", LUSTRE_OSD_NAME);
100 GOTO(out_type, rc = -ENODEV);
103 LASSERT(ofd_lu->ld_site);
104 d->ld_site = ofd_lu->ld_site;
106 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
107 "%s-osd", lustre_cfg_string(cfg, 0));
110 rc = ldt->ldt_ops->ldto_device_init(env, d, dev, NULL);
112 CERROR("can't init device '%s', rc %d\n", LUSTRE_OSD_NAME, rc);
116 lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
118 m->ofd_osd = lu2dt_dev(d);
120 /* process setup config */
121 rc = d->ld_ops->ldo_process_config(env, d, cfg);
128 ldt->ldt_ops->ldto_device_fini(env, d);
131 ldt->ldt_ops->ldto_device_free(env, d);
133 class_put_type(type);
137 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
138 struct lu_device *top)
140 struct obd_device *obd = ofd_obd(m);
141 struct lustre_cfg_bufs bufs;
142 struct lustre_cfg *lcfg;
147 lu_site_purge(env, top->ld_site, ~0);
149 /* process cleanup, pass mdt obd name to get obd umount flags */
150 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
155 lustre_cfg_bufs_set_string(&bufs, 1, flags);
156 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
158 CERROR("Cannot alloc lcfg!\n");
163 top->ld_ops->ldo_process_config(env, top, lcfg);
164 lustre_cfg_free(lcfg);
166 lu_stack_fini(env, &m->ofd_osd->dd_lu_dev);
172 /* used by MGS to process specific configurations */
173 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
174 struct lustre_cfg *cfg)
176 struct ofd_device *m = ofd_dev(d);
177 struct dt_device *dt_next = m->ofd_osd;
178 struct lu_device *next = &dt_next->dd_lu_dev;
183 switch (cfg->lcfg_command) {
185 struct lprocfs_static_vars lvars;
187 lprocfs_ofd_init_vars(&lvars);
188 rc = class_process_proc_param(PARAM_OST, lvars.obd_vars, cfg,
190 if (rc > 0 || rc == -ENOSYS)
191 /* we don't understand; pass it on */
192 rc = next->ld_ops->ldo_process_config(env, next, cfg);
195 case LCFG_SPTLRPC_CONF: {
200 /* others are passed further */
201 rc = next->ld_ops->ldo_process_config(env, next, cfg);
207 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
208 const struct lu_object_conf *conf)
210 struct ofd_device *d = ofd_dev(o->lo_dev);
211 struct lu_device *under;
212 struct lu_object *below;
217 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
218 PFID(lu_object_fid(o)));
220 under = &d->ofd_osd->dd_lu_dev;
221 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
223 lu_object_add(o, below);
230 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
232 struct ofd_object *of = ofd_obj(o);
233 struct lu_object_header *h;
238 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
239 PFID(lu_object_fid(o)));
242 lu_object_header_fini(h);
243 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
247 static int ofd_object_print(const struct lu_env *env, void *cookie,
248 lu_printer_t p, const struct lu_object *o)
250 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
253 struct lu_object_operations ofd_obj_ops = {
254 .loo_object_init = ofd_object_init,
255 .loo_object_free = ofd_object_free,
256 .loo_object_print = ofd_object_print
259 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
260 const struct lu_object_header *hdr,
263 struct ofd_object *of;
267 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, CFS_ALLOC_IO);
270 struct lu_object_header *h;
272 o = &of->ofo_obj.do_lu;
274 lu_object_header_init(h);
275 lu_object_init(o, h, d);
276 lu_object_add_top(h, o);
277 o->lo_ops = &ofd_obj_ops;
284 extern int ost_handle(struct ptlrpc_request *req);
286 static int ofd_start(const struct lu_env *env, struct lu_device *dev)
288 struct ofd_device *ofd = ofd_dev(dev);
289 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
294 /* initialize lower device */
295 rc = next->ld_ops->ldo_prepare(env, dev, next);
300 static int ofd_recovery_complete(const struct lu_env *env,
301 struct lu_device *dev)
303 struct ofd_device *ofd = ofd_dev(dev);
304 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
309 /* Grant space for object precreation on the self export.
310 * This initial reserved space (i.e. 20MB for zfs and 560KB for ldiskfs)
311 * is enough to create 20k objects. It is then adapted based on the
312 * precreate request size (see ofd_grant_create()
314 ofd_grant_connect(env, dev->ld_obd->obd_self_export,
315 OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace);
316 rc = next->ld_ops->ldo_recovery_complete(env, next);
320 static struct lu_device_operations ofd_lu_ops = {
321 .ldo_object_alloc = ofd_object_alloc,
322 .ldo_process_config = ofd_process_config,
323 .ldo_recovery_complete = ofd_recovery_complete,
326 static int ofd_procfs_init(struct ofd_device *ofd)
328 struct lprocfs_static_vars lvars;
329 struct obd_device *obd = ofd_obd(ofd);
330 cfs_proc_dir_entry_t *entry;
335 /* lprocfs must be setup before the ofd so state can be safely added
336 * to /proc incrementally as the ofd is setup */
337 lprocfs_ofd_init_vars(&lvars);
338 rc = lprocfs_obd_setup(obd, lvars.obd_vars);
340 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
345 rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_LAST);
347 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
349 GOTO(obd_cleanup, rc);
352 /* Init OFD private stats here */
353 lprocfs_counter_init(obd->obd_stats, LPROC_OFD_READ_BYTES,
354 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
355 lprocfs_counter_init(obd->obd_stats, LPROC_OFD_WRITE_BYTES,
356 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
358 rc = lproc_ofd_attach_seqstat(obd);
360 CERROR("%s: create seqstat failed: %d.\n", obd->obd_name, rc);
361 GOTO(free_obd_stats, rc);
364 entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
367 CERROR("%s: error %d setting up lprocfs for %s\n",
368 obd->obd_name, rc, "exports");
369 GOTO(free_obd_stats, rc);
371 obd->obd_proc_exports_entry = entry;
373 entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
374 lprocfs_nid_stats_clear_read,
375 lprocfs_nid_stats_clear_write, obd, NULL);
378 CERROR("%s: add proc entry 'clear' failed: %d.\n",
380 GOTO(free_obd_stats, rc);
385 lprocfs_free_obd_stats(obd);
387 lprocfs_obd_cleanup(obd);
391 static int ofd_procfs_fini(struct ofd_device *ofd)
393 struct obd_device *obd = ofd_obd(ofd);
395 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
396 lprocfs_free_per_client_stats(obd);
397 lprocfs_free_obd_stats(obd);
398 lprocfs_obd_cleanup(obd);
402 extern int ost_handle(struct ptlrpc_request *req);
404 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
405 struct lu_device_type *ldt, struct lustre_cfg *cfg)
407 const char *dev = lustre_cfg_string(cfg, 0);
408 struct ofd_thread_info *info = NULL;
409 struct obd_device *obd;
410 struct obd_statfs *osfs;
415 obd = class_name2obd(dev);
417 CERROR("Cannot find obd with name %s\n", dev);
421 rc = lu_env_refill((struct lu_env *)env);
425 obd->u.obt.obt_magic = OBT_MAGIC;
427 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
428 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
430 cfs_spin_lock_init(&m->ofd_flags_lock);
431 m->ofd_raid_degraded = 0;
432 m->ofd_syncjournal = 0;
434 m->ofd_grant_compat_disable = 0;
437 cfs_spin_lock_init(&m->ofd_osfs_lock);
438 m->ofd_osfs_age = cfs_time_shift_64(-1000);
439 m->ofd_osfs_unstable = 0;
440 m->ofd_statfs_inflight = 0;
441 m->ofd_osfs_inflight = 0;
444 cfs_spin_lock_init(&m->ofd_grant_lock);
445 m->ofd_tot_dirty = 0;
446 m->ofd_tot_granted = 0;
447 m->ofd_tot_pending = 0;
448 m->ofd_max_group = 0;
450 cfs_rwlock_init(&obd->u.filter.fo_sptlrpc_lock);
451 sptlrpc_rule_set_init(&obd->u.filter.fo_sptlrpc_rset);
453 obd->u.filter.fo_fl_oss_capa = 0;
454 CFS_INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
455 obd->u.filter.fo_capa_hash = init_capa_hash();
456 if (obd->u.filter.fo_capa_hash == NULL)
459 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
460 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
461 /* set this lu_device to obd, because error handling need it */
462 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
464 rc = ofd_procfs_init(m);
466 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
470 /* No connection accepted until configurations will finish */
471 obd->obd_no_conn = 1;
472 obd->obd_replayable = 1;
473 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
474 char *str = lustre_cfg_string(cfg, 4);
476 if (strchr(str, 'n')) {
477 CWARN("%s: recovery disabled\n", obd->obd_name);
478 obd->obd_replayable = 0;
482 info = ofd_info_init(env, NULL);
486 rc = lu_site_init(&m->ofd_site, &m->ofd_dt_dev.dd_lu_dev);
488 GOTO(err_fini_proc, rc);
489 m->ofd_site.ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
491 rc = ofd_stack_init(env, m, cfg);
493 CERROR("Can't init device stack, rc %d\n", rc);
494 GOTO(err_lu_site, rc);
497 /* populate cached statfs data */
498 osfs = &ofd_info(env)->fti_u.osfs;
499 rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
501 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
502 GOTO(err_fini_stack, rc);
504 if (!IS_PO2(osfs->os_bsize)) {
505 CERROR("%s: blocksize (%d) is not a power of 2\n",
506 obd->obd_name, osfs->os_bsize);
507 GOTO(err_fini_stack, rc = -EPROTO);
509 m->ofd_blockbits = cfs_fls(osfs->os_bsize) - 1;
511 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "filter-%p", m);
512 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
513 LDLM_NAMESPACE_SERVER,
514 LDLM_NAMESPACE_GREEDY,
516 if (m->ofd_namespace == NULL)
517 GOTO(err_fini_stack, rc = -ENOMEM);
518 /* set obd_namespace for compatibility with old code */
519 obd->obd_namespace = m->ofd_namespace;
521 dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
523 /* Allow at most ddp_grant_reserved% of the available filesystem space
524 * to be granted to clients, so that any errors in the grant overhead
525 * calculations do not allow granting more space to clients than can be
526 * written. Assumes that in aggregate the grant overhead calculations do
527 * not have more than ddp_grant_reserved% estimation error in them. */
529 ofd_grant_ratio_conv(m->ofd_dt_conf.ddp_grant_reserved);
531 rc = ofd_start(env, &m->ofd_dt_dev.dd_lu_dev);
533 GOTO(err_fini_stack, rc);
535 rc = lut_init(env, &m->ofd_lut, obd, m->ofd_osd);
537 GOTO(err_free_ns, rc);
539 rc = ofd_fs_setup(env, m, obd);
541 GOTO(err_fini_lut, rc);
543 target_recovery_init(&m->ofd_lut, ost_handle);
545 rc = lu_site_init_finish(&m->ofd_site);
547 GOTO(err_fs_cleanup, rc);
551 target_recovery_fini(obd);
552 ofd_fs_cleanup(env, m);
554 lut_fini(env, &m->ofd_lut);
556 ldlm_namespace_free(m->ofd_namespace, 0, obd->obd_force);
557 obd->obd_namespace = m->ofd_namespace = NULL;
559 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
561 lu_site_fini(&m->ofd_site);
567 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
569 struct obd_device *obd = ofd_obd(m);
570 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
572 target_recovery_fini(obd);
573 obd_exports_barrier(obd);
574 obd_zombie_barrier();
576 lut_fini(env, &m->ofd_lut);
577 ofd_fs_cleanup(env, m);
579 ofd_free_capa_keys(m);
580 cleanup_capa_hash(obd->u.filter.fo_capa_hash);
582 if (m->ofd_namespace != NULL) {
583 ldlm_namespace_free(m->ofd_namespace, NULL,
584 d->ld_obd->obd_force);
585 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
588 ofd_stack_fini(env, m, m->ofd_site.ls_top_dev);
589 lu_site_fini(&m->ofd_site);
591 LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
595 static struct lu_device *ofd_device_fini(const struct lu_env *env,
599 ofd_fini(env, ofd_dev(d));
603 static struct lu_device *ofd_device_free(const struct lu_env *env,
606 struct ofd_device *m = ofd_dev(d);
608 dt_device_fini(&m->ofd_dt_dev);
613 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
614 struct lu_device_type *t,
615 struct lustre_cfg *cfg)
617 struct ofd_device *m;
623 return ERR_PTR(-ENOMEM);
625 l = &m->ofd_dt_dev.dd_lu_dev;
626 dt_device_init(&m->ofd_dt_dev, t);
627 rc = ofd_init0(env, m, t, cfg);
629 ofd_device_free(env, l);
636 /* thread context key constructor/destructor */
637 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
639 static void ofd_key_exit(const struct lu_context *ctx,
640 struct lu_context_key *key, void *data)
642 struct ofd_thread_info *info = data;
644 info->fti_env = NULL;
645 info->fti_exp = NULL;
648 struct lu_context_key ofd_thread_key = {
649 .lct_tags = LCT_DT_THREAD,
650 .lct_init = ofd_key_init,
651 .lct_fini = ofd_key_fini,
652 .lct_exit = ofd_key_exit
655 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
656 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
658 static struct lu_device_type_operations ofd_device_type_ops = {
659 .ldto_init = ofd_type_init,
660 .ldto_fini = ofd_type_fini,
662 .ldto_start = ofd_type_start,
663 .ldto_stop = ofd_type_stop,
665 .ldto_device_alloc = ofd_device_alloc,
666 .ldto_device_free = ofd_device_free,
667 .ldto_device_fini = ofd_device_fini
670 static struct lu_device_type ofd_device_type = {
671 .ldt_tags = LU_DEVICE_DT,
672 .ldt_name = LUSTRE_OST_NAME,
673 .ldt_ops = &ofd_device_type_ops,
674 .ldt_ctx_tags = LCT_DT_THREAD
677 int __init ofd_init(void)
679 struct lprocfs_static_vars lvars;
682 rc = lu_kmem_init(ofd_caches);
688 lu_kmem_fini(ofd_caches);
692 lprocfs_ofd_init_vars(&lvars);
694 rc = class_register_type(&ofd_obd_ops, NULL, lvars.module_vars,
695 LUSTRE_OST_NAME, &ofd_device_type);
699 void __exit ofd_exit(void)
702 lu_kmem_fini(ofd_caches);
703 class_unregister_type(LUSTRE_OST_NAME);
706 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
707 MODULE_DESCRIPTION("Lustre Object Filtering Device");
708 MODULE_LICENSE("GPL");
710 module_init(ofd_init);
711 module_exit(ofd_exit);