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) 2012, 2013, Intel Corporation.
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>
47 #include <lustre_fid.h>
48 #include <lustre_lfsck.h>
49 #include <lustre/lustre_idl.h>
50 #include <lustre_dlm.h>
51 #include <lustre_quota.h>
53 #include "ofd_internal.h"
55 /* Slab for OFD object allocation */
56 static struct kmem_cache *ofd_object_kmem;
58 static struct lu_kmem_descr ofd_caches[] = {
60 .ckd_cache = &ofd_object_kmem,
61 .ckd_name = "ofd_obj",
62 .ckd_size = sizeof(struct ofd_object)
69 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
70 const char *next, struct obd_export **exp)
72 struct obd_connect_data *data = NULL;
73 struct obd_device *obd;
79 GOTO(out, rc = -ENOMEM);
81 obd = class_name2obd(next);
83 CERROR("%s: can't locate next device: %s\n",
84 m->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name, next);
85 GOTO(out, rc = -ENOTCONN);
88 data->ocd_connect_flags = OBD_CONNECT_VERSION;
89 data->ocd_version = LUSTRE_VERSION_CODE;
91 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
93 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
94 m->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name, next, rc);
98 m->ofd_dt_dev.dd_lu_dev.ld_site =
99 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
100 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
101 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
102 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
110 static int ofd_stack_init(const struct lu_env *env,
111 struct ofd_device *m, struct lustre_cfg *cfg)
113 const char *dev = lustre_cfg_string(cfg, 0);
115 struct ofd_thread_info *info = ofd_info(env);
116 struct lustre_mount_info *lmi;
122 lmi = server_get_mount(dev);
124 CERROR("Cannot get mount info for %s!\n", dev);
128 /* find bottom osd */
129 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
133 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
134 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
135 OBD_FREE(osdname, MTI_NAME_MAXLEN);
139 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
141 m->ofd_osd = lu2dt_dev(d);
143 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
144 "%s-osd", lustre_cfg_string(cfg, 0));
149 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
150 struct lu_device *top)
152 struct obd_device *obd = ofd_obd(m);
153 struct lustre_cfg_bufs bufs;
154 struct lustre_cfg *lcfg;
159 lu_site_purge(env, top->ld_site, ~0);
160 /* process cleanup, pass mdt obd name to get obd umount flags */
161 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
166 lustre_cfg_bufs_set_string(&bufs, 1, flags);
167 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
169 CERROR("Cannot alloc lcfg!\n");
174 top->ld_ops->ldo_process_config(env, top, lcfg);
175 lustre_cfg_free(lcfg);
177 lu_site_purge(env, top->ld_site, ~0);
178 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
179 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
180 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
183 LASSERT(m->ofd_osd_exp);
184 obd_disconnect(m->ofd_osd_exp);
190 /* For interoperability, see mdt_interop_param[]. */
191 static struct cfg_interop_param ofd_interop_param[] = {
192 { "ost.quota_type", NULL },
196 /* Some parameters were moved from ofd to osd and only their
197 * symlinks were kept in ofd by LU-3106. They are:
198 * -writehthrough_cache_enable
199 * -readcache_max_filese
202 * Since they are not included by the static lprocfs var list,
203 * a pre-check is added for them to avoid "unknown param" error
204 * message confuses the customer. If they are matched in this
205 * check, they will be passed to the osd directly.
207 static bool match_symlink_param(char *param)
212 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
213 sval = strchr(param, '=');
215 paramlen = sval - param;
216 if (strncmp(param, "writethrough_cache_enable",
218 strncmp(param, "readcache_max_filesize",
220 strncmp(param, "read_cache_enable",
222 strncmp(param, "brw_stats", paramlen) == 0)
230 /* used by MGS to process specific configurations */
231 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
232 struct lustre_cfg *cfg)
234 struct ofd_device *m = ofd_dev(d);
235 struct dt_device *dt_next = m->ofd_osd;
236 struct lu_device *next = &dt_next->dd_lu_dev;
241 switch (cfg->lcfg_command) {
243 struct lprocfs_static_vars lvars;
245 /* For interoperability */
246 struct cfg_interop_param *ptr = NULL;
247 struct lustre_cfg *old_cfg = NULL;
250 param = lustre_cfg_string(cfg, 1);
252 CERROR("param is empty\n");
257 ptr = class_find_old_param(param, ofd_interop_param);
259 if (ptr->new_param == NULL) {
261 CWARN("For interoperability, skip this %s."
262 " It is obsolete.\n", ptr->old_param);
266 CWARN("Found old param %s, changed it to %s.\n",
267 ptr->old_param, ptr->new_param);
270 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
277 if (match_symlink_param(param)) {
278 rc = next->ld_ops->ldo_process_config(env, next, cfg);
282 lprocfs_ofd_init_vars(&lvars);
283 rc = class_process_proc_param(PARAM_OST, lvars.obd_vars, cfg,
285 if (rc > 0 || rc == -ENOSYS) {
286 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
288 /* we don't understand; pass it on */
289 rc = next->ld_ops->ldo_process_config(env, next, cfg);
293 case LCFG_SPTLRPC_CONF: {
298 /* others are passed further */
299 rc = next->ld_ops->ldo_process_config(env, next, cfg);
305 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
306 const struct lu_object_conf *conf)
308 struct ofd_device *d = ofd_dev(o->lo_dev);
309 struct lu_device *under;
310 struct lu_object *below;
315 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
316 PFID(lu_object_fid(o)));
318 under = &d->ofd_osd->dd_lu_dev;
319 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
321 lu_object_add(o, below);
328 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
330 struct ofd_object *of = ofd_obj(o);
331 struct lu_object_header *h;
336 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
337 PFID(lu_object_fid(o)));
340 lu_object_header_fini(h);
341 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
345 static int ofd_object_print(const struct lu_env *env, void *cookie,
346 lu_printer_t p, const struct lu_object *o)
348 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
351 struct lu_object_operations ofd_obj_ops = {
352 .loo_object_init = ofd_object_init,
353 .loo_object_free = ofd_object_free,
354 .loo_object_print = ofd_object_print
357 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
358 const struct lu_object_header *hdr,
361 struct ofd_object *of;
365 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, __GFP_IO);
368 struct lu_object_header *h;
370 o = &of->ofo_obj.do_lu;
372 lu_object_header_init(h);
373 lu_object_init(o, h, d);
374 lu_object_add_top(h, o);
375 o->lo_ops = &ofd_obj_ops;
382 extern int ost_handle(struct ptlrpc_request *req);
384 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
385 struct lu_device *dev)
387 struct ofd_thread_info *info;
388 struct ofd_device *ofd = ofd_dev(dev);
389 struct obd_device *obd = ofd_obd(ofd);
390 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
391 struct lfsck_start_param lsp;
396 rc = lu_env_refill((struct lu_env *)env);
398 CERROR("Failure to refill session: '%d'\n", rc);
402 info = ofd_info_init(env, NULL);
406 /* initialize lower device */
407 rc = next->ld_ops->ldo_prepare(env, dev, next);
411 rc = lfsck_register(env, ofd->ofd_osd, &ofd->ofd_dt_dev, false);
413 CERROR("%s: failed to initialize lfsck: rc = %d\n",
418 lsp.lsp_start = NULL;
419 lsp.lsp_namespace = ofd->ofd_namespace;
420 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
422 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
427 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
428 LASSERT(obd->obd_no_conn);
429 spin_lock(&obd->obd_dev_lock);
430 obd->obd_no_conn = 0;
431 spin_unlock(&obd->obd_dev_lock);
433 if (obd->obd_recovering == 0)
434 ofd_postrecov(env, ofd);
439 static int ofd_recovery_complete(const struct lu_env *env,
440 struct lu_device *dev)
442 struct ofd_device *ofd = ofd_dev(dev);
443 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
444 int rc = 0, max_precreate;
448 /* Grant space for object precreation on the self export.
449 * This initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
450 * is enough to create 10k objects. More space is then acquired for
451 * precreation in ofd_grant_create().
453 max_precreate = OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace / 2;
454 ofd_grant_connect(env, dev->ld_obd->obd_self_export, max_precreate,
456 rc = next->ld_ops->ldo_recovery_complete(env, next);
460 static struct lu_device_operations ofd_lu_ops = {
461 .ldo_object_alloc = ofd_object_alloc,
462 .ldo_process_config = ofd_process_config,
463 .ldo_recovery_complete = ofd_recovery_complete,
464 .ldo_prepare = ofd_prepare,
467 static int ofd_procfs_init(struct ofd_device *ofd)
469 struct lprocfs_static_vars lvars;
470 struct obd_device *obd = ofd_obd(ofd);
471 cfs_proc_dir_entry_t *entry;
476 /* lprocfs must be setup before the ofd so state can be safely added
477 * to /proc incrementally as the ofd is setup */
478 lprocfs_ofd_init_vars(&lvars);
479 rc = lprocfs_obd_setup(obd, lvars.obd_vars);
481 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
486 rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
488 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
490 GOTO(obd_cleanup, rc);
493 obd->obd_uses_nid_stats = 1;
495 entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
498 CERROR("%s: error %d setting up lprocfs for %s\n",
499 obd->obd_name, rc, "exports");
500 GOTO(obd_cleanup, rc);
502 obd->obd_proc_exports_entry = entry;
504 entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
505 lprocfs_nid_stats_clear_read,
506 lprocfs_nid_stats_clear_write, obd, NULL);
509 CERROR("%s: add proc entry 'clear' failed: %d.\n",
511 GOTO(obd_cleanup, rc);
514 ofd_stats_counter_init(obd->obd_stats);
516 rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
517 ofd_stats_counter_init);
519 GOTO(remove_entry_clear, rc);
522 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
524 lprocfs_obd_cleanup(obd);
525 lprocfs_free_obd_stats(obd);
530 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
532 struct obd_device *obd = ofd_obd(ofd);
533 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
534 cfs_proc_dir_entry_t *osd_root = osd_obd->obd_type->typ_procroot;
535 cfs_proc_dir_entry_t *osd_dir;
537 osd_dir = lprocfs_srch(osd_root, obd->obd_name);
541 if (lprocfs_srch(osd_dir, "brw_stats") != NULL)
542 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
543 "../../%s/%s/brw_stats",
544 osd_root->name, osd_dir->name);
546 if (lprocfs_srch(osd_dir, "read_cache_enable") != NULL)
547 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
548 "../../%s/%s/read_cache_enable",
549 osd_root->name, osd_dir->name);
551 if (lprocfs_srch(osd_dir, "readcache_max_filesize") != NULL)
552 lprocfs_add_symlink("readcache_max_filesize",
554 "../../%s/%s/readcache_max_filesize",
555 osd_root->name, osd_dir->name);
557 if (lprocfs_srch(osd_dir, "writethrough_cache_enable") != NULL)
558 lprocfs_add_symlink("writethrough_cache_enable",
560 "../../%s/%s/writethrough_cache_enable",
561 osd_root->name, osd_dir->name);
564 static void ofd_procfs_fini(struct ofd_device *ofd)
566 struct obd_device *obd = ofd_obd(ofd);
568 lprocfs_remove_proc_entry("writethrough_cache_enable",
569 obd->obd_proc_entry);
570 lprocfs_remove_proc_entry("readcache_max_filesize",
571 obd->obd_proc_entry);
572 lprocfs_remove_proc_entry("read_cache_enable", obd->obd_proc_entry);
573 lprocfs_remove_proc_entry("brw_stats", obd->obd_proc_entry);
574 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
575 lprocfs_free_per_client_stats(obd);
576 lprocfs_obd_cleanup(obd);
577 lprocfs_free_obd_stats(obd);
578 lprocfs_job_stats_fini(obd);
581 extern int ost_handle(struct ptlrpc_request *req);
583 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
585 return seq_site_fini(env, &ofd->ofd_seq_site);
588 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
590 struct seq_server_site *ss = &ofd->ofd_seq_site;
591 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
592 char *obd_name = ofd_name(ofd);
596 ss = &ofd->ofd_seq_site;
597 lu->ld_site->ld_seq_site = ss;
598 ss->ss_lu = lu->ld_site;
599 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
601 OBD_ALLOC_PTR(ss->ss_server_seq);
602 if (ss->ss_server_seq == NULL)
603 GOTO(out_free, rc = -ENOMEM);
605 OBD_ALLOC(name, strlen(obd_name) + 10);
607 OBD_FREE_PTR(ss->ss_server_seq);
608 ss->ss_server_seq = NULL;
609 GOTO(out_free, rc = -ENOMEM);
612 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
613 LUSTRE_SEQ_SERVER, ss);
615 CERROR("%s : seq server init error %d\n", obd_name, rc);
618 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
620 OBD_ALLOC_PTR(ss->ss_client_seq);
621 if (ss->ss_client_seq == NULL)
622 GOTO(out_free, rc = -ENOMEM);
624 snprintf(name, strlen(obd_name) + 6, "%p-super", obd_name);
625 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
628 CERROR("%s : seq client init error %d\n", obd_name, rc);
631 OBD_FREE(name, strlen(obd_name) + 10);
634 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
638 if (ss->ss_server_seq) {
639 seq_server_fini(ss->ss_server_seq, env);
640 OBD_FREE_PTR(ss->ss_server_seq);
641 ss->ss_server_seq = NULL;
644 if (ss->ss_client_seq) {
645 seq_client_fini(ss->ss_client_seq);
646 OBD_FREE_PTR(ss->ss_client_seq);
647 ss->ss_client_seq = NULL;
651 OBD_FREE(name, strlen(obd_name) + 10);
659 int ofd_set_info_hdl(struct tgt_session_info *tsi)
661 struct ptlrpc_request *req = tgt_ses_req(tsi);
662 struct ost_body *body = NULL, *repbody;
663 void *key, *val = NULL;
664 int keylen, vallen, rc = 0;
665 bool is_grant_shrink;
666 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
670 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
672 DEBUG_REQ(D_HA, req, "no set_info key");
673 RETURN(err_serious(-EFAULT));
675 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
678 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
680 DEBUG_REQ(D_HA, req, "no set_info val");
681 RETURN(err_serious(-EFAULT));
683 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
686 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
688 /* In this case the value is actually an RMF_OST_BODY, so we
689 * transmutate the type of this PTLRPC */
690 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
692 rc = req_capsule_server_pack(tsi->tsi_pill);
696 if (is_grant_shrink) {
697 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
699 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
702 /** handle grant shrink, similar to a read request */
703 ofd_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
705 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
707 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
709 } else if (KEY_IS(KEY_CAPA_KEY)) {
710 rc = ofd_update_capa_key(ofd, val);
711 } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
712 rc = tgt_adapt_sptlrpc_conf(tsi->tsi_tgt, 0);
714 CERROR("%s: Unsupported key %s\n",
715 tgt_name(tsi->tsi_tgt), (char *)key);
718 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
724 static int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
725 struct lu_fid *fid, struct ll_user_fiemap *fiemap)
727 struct ofd_object *fo;
730 fo = ofd_object_find(env, ofd, fid);
732 CERROR("%s: error finding object "DFID"\n",
733 ofd_name(ofd), PFID(fid));
737 ofd_read_lock(env, fo);
738 if (ofd_object_exists(fo))
739 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
742 ofd_read_unlock(env, fo);
743 ofd_object_put(env, fo);
747 struct locked_region {
749 struct lustre_handle lh;
752 static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
753 unsigned long long begin, unsigned long long end,
756 struct locked_region *region = NULL;
760 LASSERT(begin <= end);
761 OBD_ALLOC_PTR(region);
765 rc = tgt_extent_lock(ns, res_id, begin, end, ®ion->lh,
770 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
772 cfs_list_add(®ion->list, locked);
777 static int lock_zero_regions(struct ldlm_namespace *ns,
778 struct ldlm_res_id *res_id,
779 struct ll_user_fiemap *fiemap,
782 __u64 begin = fiemap->fm_start;
785 struct ll_fiemap_extent *fiemap_start = fiemap->fm_extents;
789 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
790 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
791 if (fiemap_start[i].fe_logical > begin) {
792 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
793 begin, fiemap_start[i].fe_logical);
794 rc = lock_region(ns, res_id, begin,
795 fiemap_start[i].fe_logical, locked);
800 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
803 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
804 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
805 begin, fiemap->fm_start + fiemap->fm_length);
806 rc = lock_region(ns, res_id, begin,
807 fiemap->fm_start + fiemap->fm_length, locked);
813 static void unlock_zero_regions(struct ldlm_namespace *ns, cfs_list_t *locked)
815 struct locked_region *entry, *temp;
817 cfs_list_for_each_entry_safe(entry, temp, locked, list) {
818 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
819 tgt_extent_unlock(&entry->lh, LCK_PR);
820 cfs_list_del(&entry->list);
825 int ofd_get_info_hdl(struct tgt_session_info *tsi)
827 struct obd_export *exp = tsi->tsi_exp;
828 struct ofd_device *ofd = ofd_exp(exp);
829 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
832 int replylen, rc = 0;
836 /* this common part for get_info rpc */
837 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
839 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
840 RETURN(err_serious(-EPROTO));
842 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
845 if (KEY_IS(KEY_LAST_ID)) {
847 struct ofd_seq *oseq;
849 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
850 rc = req_capsule_server_pack(tsi->tsi_pill);
852 RETURN(err_serious(rc));
854 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
856 oseq = ofd_seq_load(tsi->tsi_env, ofd,
857 (obd_seq)exp->exp_filter_data.fed_group);
861 *last_id = ofd_seq_last_oid(oseq);
862 ofd_seq_put(tsi->tsi_env, oseq);
863 } else if (KEY_IS(KEY_FIEMAP)) {
864 struct ll_fiemap_info_key *fm_key;
865 struct ll_user_fiemap *fiemap;
866 struct lu_fid *fid = &fti->fti_fid;
868 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
870 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
871 rc = tgt_validate_obdo(tsi, &fm_key->oa);
873 RETURN(err_serious(rc));
875 *fid = fm_key->oa.o_oi.oi_fid;
877 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
879 replylen = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
880 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
881 RCL_SERVER, replylen);
883 rc = req_capsule_server_pack(tsi->tsi_pill);
885 RETURN(err_serious(rc));
887 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
891 *fiemap = fm_key->fiemap;
892 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
894 /* LU-3219: Lock the sparse areas to make sure dirty
895 * flushed back from client, then call fiemap again. */
896 if (fm_key->oa.o_valid & OBD_MD_FLFLAGS &&
897 fm_key->oa.o_flags & OBD_FL_SRVLOCK) {
898 cfs_list_t locked = CFS_LIST_HEAD_INIT(locked);
900 ost_fid_build_resid(fid, &fti->fti_resid);
901 rc = lock_zero_regions(ofd->ofd_namespace,
902 &fti->fti_resid, fiemap,
904 if (rc == 0 && !cfs_list_empty(&locked)) {
905 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
907 unlock_zero_regions(ofd->ofd_namespace,
911 } else if (KEY_IS(KEY_LAST_FID)) {
912 struct ofd_device *ofd = ofd_exp(exp);
913 struct ofd_seq *oseq;
917 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
918 rc = req_capsule_server_pack(tsi->tsi_pill);
920 RETURN(err_serious(rc));
922 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
924 RETURN(err_serious(-EPROTO));
926 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
928 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
932 oseq = ofd_seq_load(tsi->tsi_env, ofd,
933 ostid_seq(&fti->fti_ostid));
935 RETURN(PTR_ERR(oseq));
937 rc = ostid_to_fid(fid, &oseq->os_oi,
938 ofd->ofd_lut.lut_lsd.lsd_osd_index);
942 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
945 ofd_seq_put(tsi->tsi_env, oseq);
947 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
951 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
957 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
959 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
960 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
961 struct ost_body *repbody;
962 struct lustre_handle lh = { 0 };
963 struct ofd_object *fo;
965 ldlm_mode_t lock_mode = LCK_PR;
970 LASSERT(tsi->tsi_ost_body != NULL);
972 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
976 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
977 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
979 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
980 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
983 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
986 rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace,
987 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
993 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
995 GOTO(out, rc = PTR_ERR(fo));
997 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1001 obdo_from_la(&repbody->oa, &fti->fti_attr,
1002 OFD_VALID_FLAGS | LA_UID | LA_GID);
1003 tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1005 /* Store object version in reply */
1006 curr_version = dt_version_get(tsi->tsi_env,
1007 ofd_object_child(fo));
1008 if ((__s64)curr_version != -EOPNOTSUPP) {
1009 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1010 repbody->oa.o_data_version = curr_version;
1014 ofd_object_put(tsi->tsi_env, fo);
1017 tgt_extent_unlock(&lh, lock_mode);
1019 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1022 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1023 repbody->oa.o_flags = OBD_FL_FLUSH;
1028 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1030 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1031 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1032 struct ost_body *body = tsi->tsi_ost_body;
1033 struct ost_body *repbody;
1034 struct ldlm_resource *res;
1035 struct ofd_object *fo;
1036 struct filter_fid *ff = NULL;
1041 LASSERT(body != NULL);
1043 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1044 if (repbody == NULL)
1047 repbody->oa.o_oi = body->oa.o_oi;
1048 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1050 /* This would be very bad - accidentally truncating a file when
1051 * changing the time or similar - bug 12203. */
1052 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1053 body->oa.o_size != OBD_OBJECT_EOF) {
1054 static char mdsinum[48];
1056 if (body->oa.o_valid & OBD_MD_FLFID)
1057 snprintf(mdsinum, sizeof(mdsinum) - 1,
1058 "of parent "DFID, body->oa.o_parent_seq,
1059 body->oa.o_parent_oid, 0);
1063 CERROR("%s: setattr from %s is trying to truncate object "DFID
1064 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1065 PFID(&tsi->tsi_fid), mdsinum);
1069 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1071 GOTO(out, rc = PTR_ERR(fo));
1073 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1074 fti->fti_attr.la_valid &= ~LA_TYPE;
1076 if (body->oa.o_valid & OBD_MD_FLFID) {
1077 ff = &fti->fti_mds_fid;
1078 ofd_prepare_fidea(ff, &body->oa);
1081 /* setting objects attributes (including owner/group) */
1082 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff);
1086 obdo_from_la(&repbody->oa, &fti->fti_attr,
1087 OFD_VALID_FLAGS | LA_UID | LA_GID);
1088 tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1090 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1094 ofd_object_put(tsi->tsi_env, fo);
1097 /* we do not call this before to avoid lu_object_find() in
1098 * ->lvbo_update() holding another reference on the object.
1099 * otherwise concurrent destroy can make the object unavailable
1100 * for 2nd lu_object_find() waiting for the first reference
1101 * to go... deadlock! */
1102 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1103 &tsi->tsi_resid, LDLM_EXTENT, 0);
1105 ldlm_res_lvbo_update(res, NULL, 0);
1106 ldlm_resource_putref(res);
1112 static int ofd_create_hdl(struct tgt_session_info *tsi)
1114 struct ost_body *repbody;
1115 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1116 struct obdo *rep_oa;
1117 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1118 obd_seq seq = ostid_seq(&oa->o_oi);
1119 obd_id oid = ostid_id(&oa->o_oi);
1120 struct ofd_seq *oseq;
1126 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1129 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1130 if (repbody == NULL)
1133 rep_oa = &repbody->oa;
1134 rep_oa->o_oi = oa->o_oi;
1136 LASSERT(seq >= FID_SEQ_OST_MDT0);
1137 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1139 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1141 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1143 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1144 ofd_name(ofd), seq, PTR_ERR(oseq));
1148 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1149 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1150 if (!ofd_obd(ofd)->obd_recovering ||
1151 oid > ofd_seq_last_oid(oseq)) {
1152 CERROR("%s: recreate objid "DOSTID" > last id "LPU64
1153 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1154 ofd_seq_last_oid(oseq));
1155 GOTO(out_nolock, rc = -EINVAL);
1157 /* Do nothing here, we re-create objects during recovery
1158 * upon write replay, see ofd_preprw_write() */
1159 GOTO(out_nolock, rc = 0);
1161 /* former ofd_handle_precreate */
1162 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1163 (oa->o_flags & OBD_FL_DELORPHAN)) {
1164 /* destroy orphans */
1165 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1166 tsi->tsi_exp->exp_conn_cnt) {
1167 CERROR("%s: dropping old orphan cleanup request\n",
1169 GOTO(out_nolock, rc = 0);
1171 /* This causes inflight precreates to abort and drop lock */
1172 oseq->os_destroys_in_progress = 1;
1173 mutex_lock(&oseq->os_create_lock);
1174 if (!oseq->os_destroys_in_progress) {
1175 CERROR("%s:["LPU64"] destroys_in_progress already"
1176 " cleared\n", ofd_name(ofd), seq);
1177 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1180 diff = oid - ofd_seq_last_oid(oseq);
1181 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1182 ofd_seq_last_oid(oseq), diff);
1183 if (-diff > OST_MAX_PRECREATE) {
1184 /* FIXME: should reset precreate_next_id on MDS */
1186 } else if (diff < 0) {
1187 rc = ofd_orphans_destroy(tsi->tsi_env, tsi->tsi_exp,
1189 oseq->os_destroys_in_progress = 0;
1191 /* XXX: Used by MDS for the first time! */
1192 oseq->os_destroys_in_progress = 0;
1195 mutex_lock(&oseq->os_create_lock);
1196 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1197 tsi->tsi_exp->exp_conn_cnt) {
1198 CERROR("%s: dropping old precreate request\n",
1202 /* only precreate if seq is 0, IDIF or normal and also o_id
1203 * must be specfied */
1204 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1205 !fid_seq_is_idif(seq)) || oid == 0) {
1206 diff = 1; /* shouldn't we create this right now? */
1208 diff = oid - ofd_seq_last_oid(oseq);
1209 /* Do sync create if the seq is about to used up */
1210 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1211 if (unlikely(oid >= IDIF_MAX_OID - 1))
1213 } else if (fid_seq_is_norm(seq)) {
1215 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1218 CERROR("%s : invalid o_seq "DOSTID"\n",
1219 ofd_name(ofd), POSTID(&oa->o_oi));
1220 GOTO(out, rc = -EINVAL);
1225 cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
1230 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1231 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1232 /* don't enforce grant during orphan recovery */
1233 rc = ofd_grant_create(tsi->tsi_env,
1234 ofd_obd(ofd)->obd_self_export,
1237 CDEBUG(D_HA, "%s: failed to acquire grant "
1238 "space for precreate (%d): rc = %d\n",
1239 ofd_name(ofd), diff, rc);
1244 /* This can happen if a new OST is formatted and installed
1245 * in place of an old one at the same index. Instead of
1246 * precreating potentially millions of deleted old objects
1247 * (possibly filling the OST), only precreate the last batch.
1248 * LFSCK will eventually clean up any orphans. LU-14 */
1249 if (diff > 5 * OST_MAX_PRECREATE) {
1250 diff = OST_MAX_PRECREATE / 2;
1251 LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %u "
1252 "larger than the LAST_ID "DOSTID", only "
1253 "precreating the last %u objects.\n",
1254 ofd_name(ofd), POSTID(&oa->o_oi),
1255 5 * OST_MAX_PRECREATE,
1256 POSTID(&oseq->os_oi), diff);
1257 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1261 next_id = ofd_seq_last_oid(oseq) + 1;
1262 count = ofd_precreate_batch(ofd, diff);
1264 CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1265 " at "LPU64"\n", ofd_name(ofd),
1266 count, seq, next_id);
1268 if (cfs_time_after(jiffies, enough_time)) {
1269 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1270 " created at a rate of %d/s\n",
1271 ofd_name(ofd), created,
1273 created / DISK_TIMEOUT);
1277 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1278 oseq, count, sync_trans);
1282 } else if (rc < 0) {
1287 /* some objects got created, we can return
1288 * them, even if last creation failed */
1291 CERROR("%s: unable to precreate: rc = %d\n",
1294 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1295 !(oa->o_flags & OBD_FL_DELORPHAN))
1296 ofd_grant_commit(tsi->tsi_env,
1297 ofd_obd(ofd)->obd_self_export, rc);
1299 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1302 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_CREATE,
1305 mutex_unlock(&oseq->os_create_lock);
1308 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1310 ofd_seq_put(tsi->tsi_env, oseq);
1314 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1316 const struct ost_body *body = tsi->tsi_ost_body;
1317 struct ost_body *repbody;
1318 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1319 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1320 struct lu_fid *fid = &fti->fti_fid;
1327 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1330 /* This is old case for clients before Lustre 2.4 */
1331 /* If there's a DLM request, cancel the locks mentioned in it */
1332 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1334 struct ldlm_request *dlm;
1336 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1339 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0);
1342 *fid = body->oa.o_oi.oi_fid;
1343 oid = ostid_id(&body->oa.o_oi);
1346 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1348 /* check that o_misc makes sense */
1349 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1350 count = body->oa.o_misc;
1352 count = 1; /* default case - single destroy */
1354 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1355 POSTID(&body->oa.o_oi), count);
1360 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1361 if (lrc == -ENOENT) {
1363 "%s: destroying non-existent object "DFID"\n",
1364 ofd_name(ofd), PFID(fid));
1365 /* rewrite rc with -ENOENT only if it is 0 */
1368 } else if (lrc != 0) {
1369 CERROR("%s: error destroying object "DFID": %d\n",
1370 ofd_name(ofd), PFID(fid), lrc);
1376 lrc = fid_set_id(fid, oid);
1377 if (unlikely(lrc != 0 && count > 0))
1378 GOTO(out, rc = lrc);
1381 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1387 fid_to_ostid(fid, &repbody->oa.o_oi);
1391 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1393 struct obd_statfs *osfs;
1398 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1400 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1401 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
1403 CERROR("%s: statfs failed: rc = %d\n",
1404 tgt_name(tsi->tsi_tgt), rc);
1406 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1409 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1415 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1417 struct ost_body *body = tsi->tsi_ost_body;
1418 struct ost_body *repbody;
1419 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1420 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1421 struct ofd_object *fo = NULL;
1426 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1428 /* if no objid is specified, it means "sync whole filesystem" */
1429 if (!fid_is_zero(&tsi->tsi_fid)) {
1430 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1432 RETURN(PTR_ERR(fo));
1435 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1436 fo != NULL ? ofd_object_child(fo) : NULL);
1440 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1445 repbody->oa.o_oi = body->oa.o_oi;
1446 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1448 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1450 obdo_from_la(&repbody->oa, &fti->fti_attr,
1453 /* don't return rc from getattr */
1458 ofd_object_put(tsi->tsi_env, fo);
1462 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1464 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1465 struct ost_body *repbody;
1466 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1467 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1468 struct ldlm_resource *res;
1469 struct ofd_object *fo;
1470 struct filter_fid *ff = NULL;
1472 struct lustre_handle lh = { 0, };
1479 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1480 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
1482 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1483 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1484 RETURN(err_serious(-EPROTO));
1486 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1487 if (repbody == NULL)
1488 RETURN(err_serious(-ENOMEM));
1490 /* punch start,end are passed in o_size,o_blocks throught wire */
1494 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1497 /* standard truncate optimization: if file body is completely
1498 * destroyed, don't send data back to the server. */
1500 flags |= LDLM_FL_AST_DISCARD_DATA;
1502 repbody->oa.o_oi = oa->o_oi;
1503 repbody->oa.o_valid = OBD_MD_FLID;
1505 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1506 oa->o_flags & OBD_FL_SRVLOCK;
1509 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
1515 CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
1516 ", start = "LPD64", end = "LPD64"\n", PFID(&tsi->tsi_fid),
1517 oa->o_valid, start, end);
1519 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
1522 GOTO(out, rc = PTR_ERR(fo));
1524 la_from_obdo(&info->fti_attr, oa,
1525 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
1526 info->fti_attr.la_size = start;
1527 info->fti_attr.la_valid |= LA_SIZE;
1529 if (oa->o_valid & OBD_MD_FLFID) {
1530 ff = &info->fti_mds_fid;
1531 ofd_prepare_fidea(ff, oa);
1534 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
1539 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
1543 ofd_object_put(tsi->tsi_env, fo);
1546 tgt_extent_unlock(&lh, LCK_PW);
1548 /* we do not call this before to avoid lu_object_find() in
1549 * ->lvbo_update() holding another reference on the object.
1550 * otherwise concurrent destroy can make the object unavailable
1551 * for 2nd lu_object_find() waiting for the first reference
1552 * to go... deadlock! */
1553 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
1556 ldlm_res_lvbo_update(res, NULL, 0);
1557 ldlm_resource_putref(res);
1563 static int ofd_quotactl(struct tgt_session_info *tsi)
1565 struct obd_quotactl *oqctl, *repoqc;
1570 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
1572 RETURN(err_serious(-EPROTO));
1574 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
1576 RETURN(err_serious(-ENOMEM));
1578 /* report success for quota on/off for interoperability with current MDT
1580 if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1584 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
1586 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
1592 /* High priority request handlers for OFD */
1594 /* prolong locks for the current service time of the corresponding
1595 * portal (= OST_IO_PORTAL)
1597 static inline int prolong_timeout(struct ptlrpc_request *req)
1599 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
1602 return obd_timeout / 2;
1604 return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
1608 static int ofd_prolong_one_lock(struct tgt_session_info *tsi,
1609 struct ldlm_lock *lock,
1610 struct ldlm_extent *extent, int timeout)
1613 if (lock->l_flags & LDLM_FL_DESTROYED) /* lock already cancelled */
1616 /* XXX: never try to grab resource lock here because we're inside
1617 * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
1618 * res lock and then exp_bl_list_lock. */
1620 if (!(lock->l_flags & LDLM_FL_AST_SENT))
1621 /* ignore locks not being cancelled */
1624 LDLM_DEBUG(lock, "refreshed for req x"LPU64" ext("LPU64"->"LPU64") "
1625 "to %ds.\n", tgt_ses_req(tsi)->rq_xid, extent->start,
1626 extent->end, timeout);
1628 /* OK. this is a possible lock the user holds doing I/O
1629 * let's refresh eviction timer for it */
1630 ldlm_refresh_waiting_lock(lock, timeout);
1634 static int ofd_prolong_extent_locks(struct tgt_session_info *tsi,
1635 __u64 start, __u64 end)
1637 struct obd_export *exp = tsi->tsi_exp;
1638 struct obdo *oa = &tsi->tsi_ost_body->oa;
1639 struct ldlm_extent extent = {
1643 struct ldlm_lock *lock;
1644 int timeout = prolong_timeout(tgt_ses_req(tsi));
1649 if (oa->o_valid & OBD_MD_FLHANDLE) {
1650 /* mostly a request should be covered by only one lock, try
1652 lock = ldlm_handle2lock(&oa->o_handle);
1654 /* Fast path to check if the lock covers the whole IO
1655 * region exclusively. */
1656 if (lock->l_granted_mode == LCK_PW &&
1657 ldlm_extent_contain(&lock->l_policy_data.l_extent,
1660 LASSERT(lock->l_export == exp);
1661 lock_count = ofd_prolong_one_lock(tsi, lock,
1663 LDLM_LOCK_PUT(lock);
1666 LDLM_LOCK_PUT(lock);
1670 spin_lock_bh(&exp->exp_bl_list_lock);
1671 list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
1672 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1673 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
1675 if (!ldlm_res_eq(&tsi->tsi_resid, &lock->l_resource->lr_name))
1678 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
1682 lock_count += ofd_prolong_one_lock(tsi, lock, &extent, timeout);
1684 spin_unlock_bh(&exp->exp_bl_list_lock);
1690 * Returns 1 if the given PTLRPC matches the given LDLM lock, or 0 if it does
1693 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
1694 struct ldlm_lock *lock)
1696 struct niobuf_remote *rnb;
1697 struct obd_ioobj *ioo;
1699 struct ldlm_extent ext;
1700 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
1704 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1705 LASSERT(ioo != NULL);
1707 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1708 LASSERT(rnb != NULL);
1710 ext.start = rnb->offset;
1711 rnb += ioo->ioo_bufcnt - 1;
1712 ext.end = rnb->offset + rnb->len - 1;
1714 LASSERT(lock->l_resource != NULL);
1715 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
1719 if (opc == OST_READ)
1722 if (!(lock->l_granted_mode & mode))
1725 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
1729 * High-priority queue request check for whether the given PTLRPC request
1730 * (\a req) is blocking an LDLM lock cancel.
1732 * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1733 * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1735 * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue. This
1736 * function looks only at OST_READs and OST_WRITEs.
1738 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
1740 struct tgt_session_info *tsi;
1741 struct obd_ioobj *ioo;
1742 struct niobuf_remote *rnb;
1748 /* Don't use tgt_ses_info() to get session info, because lock_match()
1749 * can be called while request has no processing thread yet. */
1750 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1751 LASSERT(tsi != NULL);
1754 * Use LASSERT below because malformed RPCs should have
1755 * been filtered out in tgt_hpreq_handler().
1757 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1758 LASSERT(ioo != NULL);
1760 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1761 LASSERT(rnb != NULL);
1762 LASSERT(!(rnb->flags & OBD_BRW_SRVLOCK));
1764 start = rnb->offset;
1765 rnb += ioo->ioo_bufcnt - 1;
1766 end = rnb->offset + rnb->len - 1;
1768 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
1769 " ("LPU64"->"LPU64")\n",
1770 tgt_name(tsi->tsi_tgt), current->comm,
1771 PFID(&tsi->tsi_fid), start, end);
1773 lock_count = ofd_prolong_extent_locks(tsi, start, end);
1775 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1776 tgt_name(tsi->tsi_tgt), lock_count, req);
1778 RETURN(lock_count > 0);
1781 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
1783 ofd_rw_hpreq_check(req);
1787 * Like tgt_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1789 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
1790 struct ldlm_lock *lock)
1792 struct tgt_session_info *tsi;
1794 /* Don't use tgt_ses_info() to get session info, because lock_match()
1795 * can be called while request has no processing thread yet. */
1796 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1797 LASSERT(tsi != NULL);
1799 LASSERT(tsi->tsi_ost_body != NULL);
1800 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
1801 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1808 * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1810 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
1812 struct tgt_session_info *tsi;
1818 /* Don't use tgt_ses_info() to get session info, because lock_match()
1819 * can be called while request has no processing thread yet. */
1820 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1821 LASSERT(tsi != NULL);
1822 oa = &tsi->tsi_ost_body->oa;
1824 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
1825 oa->o_flags & OBD_FL_SRVLOCK));
1828 "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1829 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
1830 tsi->tsi_resid.name[1], oa->o_size, oa->o_blocks);
1832 lock_count = ofd_prolong_extent_locks(tsi, oa->o_size, oa->o_blocks);
1834 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1835 tgt_name(tsi->tsi_tgt), lock_count, req);
1837 RETURN(lock_count > 0);
1840 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
1842 ofd_punch_hpreq_check(req);
1845 struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
1846 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
1847 .hpreq_check = ofd_rw_hpreq_check,
1848 .hpreq_fini = ofd_rw_hpreq_fini
1851 struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
1852 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
1853 .hpreq_check = ofd_punch_hpreq_check,
1854 .hpreq_fini = ofd_punch_hpreq_fini
1857 /** Assign high priority operations to the IO requests */
1858 static void ofd_hp_brw(struct tgt_session_info *tsi)
1860 struct niobuf_remote *rnb;
1861 struct obd_ioobj *ioo;
1865 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1866 LASSERT(ioo != NULL); /* must exist after request preprocessing */
1867 if (ioo->ioo_bufcnt > 0) {
1868 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
1869 LASSERT(rnb != NULL); /* must exist after request preprocessing */
1871 /* no high priority if server lock is needed */
1872 if (rnb->flags & OBD_BRW_SRVLOCK)
1875 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
1878 static void ofd_hp_punch(struct tgt_session_info *tsi)
1880 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
1881 /* no high-priority if server lock is needed */
1882 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1883 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK)
1885 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
1888 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
1889 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
1890 #define OST_BRW_READ OST_READ
1891 #define OST_BRW_WRITE OST_WRITE
1893 static struct tgt_handler ofd_tgt_handlers[] = {
1894 TGT_RPC_HANDLER(OST_FIRST_OPC,
1895 0, OST_CONNECT, tgt_connect,
1896 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1897 TGT_RPC_HANDLER(OST_FIRST_OPC,
1898 0, OST_DISCONNECT, tgt_disconnect,
1899 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
1900 TGT_RPC_HANDLER(OST_FIRST_OPC,
1901 0, OST_SET_INFO, ofd_set_info_hdl,
1902 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
1903 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
1904 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
1905 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
1906 OST_SETATTR, ofd_setattr_hdl),
1907 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
1908 OST_CREATE, ofd_create_hdl),
1909 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
1910 OST_DESTROY, ofd_destroy_hdl),
1911 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
1912 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
1913 OST_BRW_READ, tgt_brw_read,
1915 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
1916 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
1918 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
1919 OST_PUNCH, ofd_punch_hdl,
1921 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
1922 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
1925 static struct tgt_opc_slice ofd_common_slice[] = {
1927 .tos_opc_start = OST_FIRST_OPC,
1928 .tos_opc_end = OST_LAST_OPC,
1929 .tos_hs = ofd_tgt_handlers
1932 .tos_opc_start = OBD_FIRST_OPC,
1933 .tos_opc_end = OBD_LAST_OPC,
1934 .tos_hs = tgt_obd_handlers
1937 .tos_opc_start = LDLM_FIRST_OPC,
1938 .tos_opc_end = LDLM_LAST_OPC,
1939 .tos_hs = tgt_dlm_handlers
1942 .tos_opc_start = UPDATE_OBJ,
1943 .tos_opc_end = UPDATE_LAST_OPC,
1944 .tos_hs = tgt_out_handlers
1947 .tos_opc_start = SEQ_FIRST_OPC,
1948 .tos_opc_end = SEQ_LAST_OPC,
1949 .tos_hs = seq_handlers
1956 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
1957 struct lu_device_type *ldt, struct lustre_cfg *cfg)
1959 const char *dev = lustre_cfg_string(cfg, 0);
1960 struct ofd_thread_info *info = NULL;
1961 struct obd_device *obd;
1962 struct obd_statfs *osfs;
1967 obd = class_name2obd(dev);
1969 CERROR("Cannot find obd with name %s\n", dev);
1973 rc = lu_env_refill((struct lu_env *)env);
1977 obd->u.obt.obt_magic = OBT_MAGIC;
1979 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
1980 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
1982 spin_lock_init(&m->ofd_flags_lock);
1983 m->ofd_raid_degraded = 0;
1984 m->ofd_syncjournal = 0;
1986 m->ofd_grant_compat_disable = 0;
1987 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
1990 spin_lock_init(&m->ofd_osfs_lock);
1991 m->ofd_osfs_age = cfs_time_shift_64(-1000);
1992 m->ofd_osfs_unstable = 0;
1993 m->ofd_statfs_inflight = 0;
1994 m->ofd_osfs_inflight = 0;
1997 spin_lock_init(&m->ofd_grant_lock);
1998 m->ofd_tot_dirty = 0;
1999 m->ofd_tot_granted = 0;
2000 m->ofd_tot_pending = 0;
2001 m->ofd_seq_count = 0;
2003 spin_lock_init(&m->ofd_batch_lock);
2004 rwlock_init(&obd->u.filter.fo_sptlrpc_lock);
2005 sptlrpc_rule_set_init(&obd->u.filter.fo_sptlrpc_rset);
2007 obd->u.filter.fo_fl_oss_capa = 0;
2008 CFS_INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
2009 obd->u.filter.fo_capa_hash = init_capa_hash();
2010 if (obd->u.filter.fo_capa_hash == NULL)
2013 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2014 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2015 /* set this lu_device to obd, because error handling need it */
2016 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2018 rc = ofd_procfs_init(m);
2020 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2024 /* No connection accepted until configurations will finish */
2025 spin_lock(&obd->obd_dev_lock);
2026 obd->obd_no_conn = 1;
2027 spin_unlock(&obd->obd_dev_lock);
2028 obd->obd_replayable = 1;
2029 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2030 char *str = lustre_cfg_string(cfg, 4);
2032 if (strchr(str, 'n')) {
2033 CWARN("%s: recovery disabled\n", obd->obd_name);
2034 obd->obd_replayable = 0;
2038 info = ofd_info_init(env, NULL);
2042 rc = ofd_stack_init(env, m, cfg);
2044 CERROR("Can't init device stack, rc %d\n", rc);
2045 GOTO(err_fini_proc, rc);
2048 ofd_procfs_add_brw_stats_symlink(m);
2050 /* populate cached statfs data */
2051 osfs = &ofd_info(env)->fti_u.osfs;
2052 rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
2054 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
2055 GOTO(err_fini_stack, rc);
2057 if (!IS_PO2(osfs->os_bsize)) {
2058 CERROR("%s: blocksize (%d) is not a power of 2\n",
2059 obd->obd_name, osfs->os_bsize);
2060 GOTO(err_fini_stack, rc = -EPROTO);
2062 m->ofd_blockbits = fls(osfs->os_bsize) - 1;
2064 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2065 if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
2066 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2068 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2069 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2070 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2071 LDLM_NAMESPACE_SERVER,
2072 LDLM_NAMESPACE_GREEDY,
2074 if (m->ofd_namespace == NULL)
2075 GOTO(err_fini_stack, rc = -ENOMEM);
2076 /* set obd_namespace for compatibility with old code */
2077 obd->obd_namespace = m->ofd_namespace;
2078 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2079 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2080 m->ofd_namespace->ns_lvbp = m;
2082 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2083 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2085 dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
2087 /* Allow at most ddp_grant_reserved% of the available filesystem space
2088 * to be granted to clients, so that any errors in the grant overhead
2089 * calculations do not allow granting more space to clients than can be
2090 * written. Assumes that in aggregate the grant overhead calculations do
2091 * not have more than ddp_grant_reserved% estimation error in them. */
2092 m->ofd_grant_ratio =
2093 ofd_grant_ratio_conv(m->ofd_dt_conf.ddp_grant_reserved);
2095 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2096 OBD_FAIL_OST_ALL_REQUEST_NET,
2097 OBD_FAIL_OST_ALL_REPLY_NET);
2099 GOTO(err_free_ns, rc);
2101 rc = ofd_fs_setup(env, m, obd);
2103 GOTO(err_fini_lut, rc);
2107 tgt_fini(env, &m->ofd_lut);
2109 ldlm_namespace_free(m->ofd_namespace, 0, obd->obd_force);
2110 obd->obd_namespace = m->ofd_namespace = NULL;
2112 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
2118 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
2120 struct obd_device *obd = ofd_obd(m);
2121 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
2123 lfsck_stop(env, m->ofd_osd, true);
2124 lfsck_degister(env, m->ofd_osd);
2125 target_recovery_fini(obd);
2126 obd_exports_barrier(obd);
2127 obd_zombie_barrier();
2129 tgt_fini(env, &m->ofd_lut);
2130 ofd_fs_cleanup(env, m);
2132 ofd_free_capa_keys(m);
2133 cleanup_capa_hash(obd->u.filter.fo_capa_hash);
2135 if (m->ofd_namespace != NULL) {
2136 ldlm_namespace_free(m->ofd_namespace, NULL,
2137 d->ld_obd->obd_force);
2138 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
2141 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
2143 LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
2144 server_put_mount(obd->obd_name);
2148 static struct lu_device *ofd_device_fini(const struct lu_env *env,
2149 struct lu_device *d)
2152 ofd_fini(env, ofd_dev(d));
2156 static struct lu_device *ofd_device_free(const struct lu_env *env,
2157 struct lu_device *d)
2159 struct ofd_device *m = ofd_dev(d);
2161 dt_device_fini(&m->ofd_dt_dev);
2166 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
2167 struct lu_device_type *t,
2168 struct lustre_cfg *cfg)
2170 struct ofd_device *m;
2171 struct lu_device *l;
2176 return ERR_PTR(-ENOMEM);
2178 l = &m->ofd_dt_dev.dd_lu_dev;
2179 dt_device_init(&m->ofd_dt_dev, t);
2180 rc = ofd_init0(env, m, t, cfg);
2182 ofd_device_free(env, l);
2189 /* thread context key constructor/destructor */
2190 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2192 static void ofd_key_exit(const struct lu_context *ctx,
2193 struct lu_context_key *key, void *data)
2195 struct ofd_thread_info *info = data;
2197 info->fti_env = NULL;
2198 info->fti_exp = NULL;
2201 info->fti_pre_version = 0;
2204 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2207 struct lu_context_key ofd_thread_key = {
2208 .lct_tags = LCT_DT_THREAD,
2209 .lct_init = ofd_key_init,
2210 .lct_fini = ofd_key_fini,
2211 .lct_exit = ofd_key_exit
2214 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
2215 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
2217 static struct lu_device_type_operations ofd_device_type_ops = {
2218 .ldto_init = ofd_type_init,
2219 .ldto_fini = ofd_type_fini,
2221 .ldto_start = ofd_type_start,
2222 .ldto_stop = ofd_type_stop,
2224 .ldto_device_alloc = ofd_device_alloc,
2225 .ldto_device_free = ofd_device_free,
2226 .ldto_device_fini = ofd_device_fini
2229 static struct lu_device_type ofd_device_type = {
2230 .ldt_tags = LU_DEVICE_DT,
2231 .ldt_name = LUSTRE_OST_NAME,
2232 .ldt_ops = &ofd_device_type_ops,
2233 .ldt_ctx_tags = LCT_DT_THREAD
2236 int __init ofd_init(void)
2238 struct lprocfs_static_vars lvars;
2241 rc = lu_kmem_init(ofd_caches);
2245 rc = ofd_fmd_init();
2247 lu_kmem_fini(ofd_caches);
2251 lprocfs_ofd_init_vars(&lvars);
2253 rc = class_register_type(&ofd_obd_ops, NULL, NULL,
2254 #ifndef HAVE_ONLY_PROCFS_SEQ
2257 LUSTRE_OST_NAME, &ofd_device_type);
2261 void __exit ofd_exit(void)
2264 lu_kmem_fini(ofd_caches);
2265 class_unregister_type(LUSTRE_OST_NAME);
2268 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
2269 MODULE_DESCRIPTION("Lustre Object Filtering Device");
2270 MODULE_LICENSE("GPL");
2272 module_init(ofd_init);
2273 module_exit(ofd_exit);