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 info = ofd_info_init(env, NULL);
400 /* initialize lower device */
401 rc = next->ld_ops->ldo_prepare(env, dev, next);
405 rc = lfsck_register(env, ofd->ofd_osd, &ofd->ofd_dt_dev, false);
407 CERROR("%s: failed to initialize lfsck: rc = %d\n",
412 lsp.lsp_start = NULL;
413 lsp.lsp_namespace = ofd->ofd_namespace;
414 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
416 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
421 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
422 LASSERT(obd->obd_no_conn);
423 spin_lock(&obd->obd_dev_lock);
424 obd->obd_no_conn = 0;
425 spin_unlock(&obd->obd_dev_lock);
427 if (obd->obd_recovering == 0)
428 ofd_postrecov(env, ofd);
433 static int ofd_recovery_complete(const struct lu_env *env,
434 struct lu_device *dev)
436 struct ofd_device *ofd = ofd_dev(dev);
437 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
438 int rc = 0, max_precreate;
442 /* Grant space for object precreation on the self export.
443 * This initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
444 * is enough to create 10k objects. More space is then acquired for
445 * precreation in ofd_grant_create().
447 max_precreate = OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace / 2;
448 ofd_grant_connect(env, dev->ld_obd->obd_self_export, max_precreate,
450 rc = next->ld_ops->ldo_recovery_complete(env, next);
454 static struct lu_device_operations ofd_lu_ops = {
455 .ldo_object_alloc = ofd_object_alloc,
456 .ldo_process_config = ofd_process_config,
457 .ldo_recovery_complete = ofd_recovery_complete,
458 .ldo_prepare = ofd_prepare,
461 static int ofd_procfs_init(struct ofd_device *ofd)
463 struct lprocfs_static_vars lvars;
464 struct obd_device *obd = ofd_obd(ofd);
465 cfs_proc_dir_entry_t *entry;
470 /* lprocfs must be setup before the ofd so state can be safely added
471 * to /proc incrementally as the ofd is setup */
472 lprocfs_ofd_init_vars(&lvars);
473 rc = lprocfs_obd_setup(obd, lvars.obd_vars);
475 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
480 rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
482 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
484 GOTO(obd_cleanup, rc);
487 obd->obd_uses_nid_stats = 1;
489 entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
492 CERROR("%s: error %d setting up lprocfs for %s\n",
493 obd->obd_name, rc, "exports");
494 GOTO(obd_cleanup, rc);
496 obd->obd_proc_exports_entry = entry;
498 entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
499 lprocfs_nid_stats_clear_read,
500 lprocfs_nid_stats_clear_write, obd, NULL);
503 CERROR("%s: add proc entry 'clear' failed: %d.\n",
505 GOTO(obd_cleanup, rc);
508 ofd_stats_counter_init(obd->obd_stats);
510 rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
511 ofd_stats_counter_init);
513 GOTO(remove_entry_clear, rc);
516 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
518 lprocfs_obd_cleanup(obd);
519 lprocfs_free_obd_stats(obd);
524 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
526 struct obd_device *obd = ofd_obd(ofd);
527 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
528 cfs_proc_dir_entry_t *osd_root = osd_obd->obd_type->typ_procroot;
529 cfs_proc_dir_entry_t *osd_dir;
531 osd_dir = lprocfs_srch(osd_root, obd->obd_name);
535 if (lprocfs_srch(osd_dir, "brw_stats") != NULL)
536 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
537 "../../%s/%s/brw_stats",
538 osd_root->name, osd_dir->name);
540 if (lprocfs_srch(osd_dir, "read_cache_enable") != NULL)
541 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
542 "../../%s/%s/read_cache_enable",
543 osd_root->name, osd_dir->name);
545 if (lprocfs_srch(osd_dir, "readcache_max_filesize") != NULL)
546 lprocfs_add_symlink("readcache_max_filesize",
548 "../../%s/%s/readcache_max_filesize",
549 osd_root->name, osd_dir->name);
551 if (lprocfs_srch(osd_dir, "writethrough_cache_enable") != NULL)
552 lprocfs_add_symlink("writethrough_cache_enable",
554 "../../%s/%s/writethrough_cache_enable",
555 osd_root->name, osd_dir->name);
558 static void ofd_procfs_fini(struct ofd_device *ofd)
560 struct obd_device *obd = ofd_obd(ofd);
562 lprocfs_remove_proc_entry("writethrough_cache_enable",
563 obd->obd_proc_entry);
564 lprocfs_remove_proc_entry("readcache_max_filesize",
565 obd->obd_proc_entry);
566 lprocfs_remove_proc_entry("read_cache_enable", obd->obd_proc_entry);
567 lprocfs_remove_proc_entry("brw_stats", obd->obd_proc_entry);
568 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
569 lprocfs_free_per_client_stats(obd);
570 lprocfs_obd_cleanup(obd);
571 lprocfs_free_obd_stats(obd);
572 lprocfs_job_stats_fini(obd);
575 extern int ost_handle(struct ptlrpc_request *req);
577 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
579 return seq_site_fini(env, &ofd->ofd_seq_site);
582 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
584 struct seq_server_site *ss = &ofd->ofd_seq_site;
585 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
586 char *obd_name = ofd_name(ofd);
590 ss = &ofd->ofd_seq_site;
591 lu->ld_site->ld_seq_site = ss;
592 ss->ss_lu = lu->ld_site;
593 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
595 OBD_ALLOC_PTR(ss->ss_server_seq);
596 if (ss->ss_server_seq == NULL)
597 GOTO(out_free, rc = -ENOMEM);
599 OBD_ALLOC(name, strlen(obd_name) + 10);
601 OBD_FREE_PTR(ss->ss_server_seq);
602 ss->ss_server_seq = NULL;
603 GOTO(out_free, rc = -ENOMEM);
606 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
607 LUSTRE_SEQ_SERVER, ss);
609 CERROR("%s : seq server init error %d\n", obd_name, rc);
612 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
614 OBD_ALLOC_PTR(ss->ss_client_seq);
615 if (ss->ss_client_seq == NULL)
616 GOTO(out_free, rc = -ENOMEM);
618 snprintf(name, strlen(obd_name) + 6, "%p-super", obd_name);
619 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
622 CERROR("%s : seq client init error %d\n", obd_name, rc);
625 OBD_FREE(name, strlen(obd_name) + 10);
628 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
632 if (ss->ss_server_seq) {
633 seq_server_fini(ss->ss_server_seq, env);
634 OBD_FREE_PTR(ss->ss_server_seq);
635 ss->ss_server_seq = NULL;
638 if (ss->ss_client_seq) {
639 seq_client_fini(ss->ss_client_seq);
640 OBD_FREE_PTR(ss->ss_client_seq);
641 ss->ss_client_seq = NULL;
645 OBD_FREE(name, strlen(obd_name) + 10);
653 int ofd_set_info_hdl(struct tgt_session_info *tsi)
655 struct ptlrpc_request *req = tgt_ses_req(tsi);
656 struct ost_body *body = NULL, *repbody;
657 void *key, *val = NULL;
658 int keylen, vallen, rc = 0;
659 bool is_grant_shrink;
660 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
664 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
666 DEBUG_REQ(D_HA, req, "no set_info key");
667 RETURN(err_serious(-EFAULT));
669 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
672 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
674 DEBUG_REQ(D_HA, req, "no set_info val");
675 RETURN(err_serious(-EFAULT));
677 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
680 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
682 /* In this case the value is actually an RMF_OST_BODY, so we
683 * transmutate the type of this PTLRPC */
684 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
686 rc = req_capsule_server_pack(tsi->tsi_pill);
690 if (is_grant_shrink) {
691 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
693 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
696 /** handle grant shrink, similar to a read request */
697 ofd_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
699 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
701 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
703 } else if (KEY_IS(KEY_CAPA_KEY)) {
704 rc = ofd_update_capa_key(ofd, val);
705 } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
706 rc = tgt_adapt_sptlrpc_conf(tsi->tsi_tgt, 0);
708 CERROR("%s: Unsupported key %s\n",
709 tgt_name(tsi->tsi_tgt), (char *)key);
712 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
718 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
719 struct lu_fid *fid, struct ll_user_fiemap *fiemap)
721 struct ofd_object *fo;
724 fo = ofd_object_find(env, ofd, fid);
726 CERROR("%s: error finding object "DFID"\n",
727 ofd_name(ofd), PFID(fid));
731 ofd_read_lock(env, fo);
732 if (ofd_object_exists(fo))
733 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
736 ofd_read_unlock(env, fo);
737 ofd_object_put(env, fo);
741 struct locked_region {
743 struct lustre_handle lh;
746 static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
747 unsigned long long begin, unsigned long long end,
750 struct locked_region *region = NULL;
754 LASSERT(begin <= end);
755 OBD_ALLOC_PTR(region);
759 rc = tgt_extent_lock(ns, res_id, begin, end, ®ion->lh,
764 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
766 cfs_list_add(®ion->list, locked);
771 static int lock_zero_regions(struct ldlm_namespace *ns,
772 struct ldlm_res_id *res_id,
773 struct ll_user_fiemap *fiemap,
776 __u64 begin = fiemap->fm_start;
779 struct ll_fiemap_extent *fiemap_start = fiemap->fm_extents;
783 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
784 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
785 if (fiemap_start[i].fe_logical > begin) {
786 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
787 begin, fiemap_start[i].fe_logical);
788 rc = lock_region(ns, res_id, begin,
789 fiemap_start[i].fe_logical, locked);
794 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
797 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
798 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
799 begin, fiemap->fm_start + fiemap->fm_length);
800 rc = lock_region(ns, res_id, begin,
801 fiemap->fm_start + fiemap->fm_length, locked);
807 static void unlock_zero_regions(struct ldlm_namespace *ns, cfs_list_t *locked)
809 struct locked_region *entry, *temp;
811 cfs_list_for_each_entry_safe(entry, temp, locked, list) {
812 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
813 tgt_extent_unlock(&entry->lh, LCK_PR);
814 cfs_list_del(&entry->list);
819 int ofd_get_info_hdl(struct tgt_session_info *tsi)
821 struct obd_export *exp = tsi->tsi_exp;
822 struct ofd_device *ofd = ofd_exp(exp);
823 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
826 int replylen, rc = 0;
830 /* this common part for get_info rpc */
831 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
833 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
834 RETURN(err_serious(-EPROTO));
836 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
839 if (KEY_IS(KEY_LAST_ID)) {
841 struct ofd_seq *oseq;
843 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
844 rc = req_capsule_server_pack(tsi->tsi_pill);
846 RETURN(err_serious(rc));
848 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
850 oseq = ofd_seq_load(tsi->tsi_env, ofd,
851 (obd_seq)exp->exp_filter_data.fed_group);
855 *last_id = ofd_seq_last_oid(oseq);
856 ofd_seq_put(tsi->tsi_env, oseq);
857 } else if (KEY_IS(KEY_FIEMAP)) {
858 struct ll_fiemap_info_key *fm_key;
859 struct ll_user_fiemap *fiemap;
862 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
864 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
865 rc = tgt_validate_obdo(tsi, &fm_key->oa);
867 RETURN(err_serious(rc));
869 fid = &fm_key->oa.o_oi.oi_fid;
871 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
873 replylen = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
874 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
875 RCL_SERVER, replylen);
877 rc = req_capsule_server_pack(tsi->tsi_pill);
879 RETURN(err_serious(rc));
881 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
885 *fiemap = fm_key->fiemap;
886 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
888 /* LU-3219: Lock the sparse areas to make sure dirty
889 * flushed back from client, then call fiemap again. */
890 if (fm_key->oa.o_valid & OBD_MD_FLFLAGS &&
891 fm_key->oa.o_flags & OBD_FL_SRVLOCK) {
892 cfs_list_t locked = CFS_LIST_HEAD_INIT(locked);
894 ost_fid_build_resid(fid, &fti->fti_resid);
895 rc = lock_zero_regions(ofd->ofd_namespace,
896 &fti->fti_resid, fiemap,
898 if (rc == 0 && !cfs_list_empty(&locked)) {
899 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
901 unlock_zero_regions(ofd->ofd_namespace,
905 } else if (KEY_IS(KEY_LAST_FID)) {
906 struct ofd_device *ofd = ofd_exp(exp);
907 struct ofd_seq *oseq;
911 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
912 rc = req_capsule_server_pack(tsi->tsi_pill);
914 RETURN(err_serious(rc));
916 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
918 RETURN(err_serious(-EPROTO));
920 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
922 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
926 oseq = ofd_seq_load(tsi->tsi_env, ofd,
927 ostid_seq(&fti->fti_ostid));
929 RETURN(PTR_ERR(oseq));
931 rc = ostid_to_fid(fid, &oseq->os_oi,
932 ofd->ofd_lut.lut_lsd.lsd_osd_index);
936 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
939 ofd_seq_put(tsi->tsi_env, oseq);
941 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
945 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
951 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
953 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
954 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
955 struct ost_body *repbody;
956 struct lustre_handle lh = { 0 };
957 struct ofd_object *fo;
959 ldlm_mode_t lock_mode = LCK_PR;
964 LASSERT(tsi->tsi_ost_body != NULL);
966 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
970 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
971 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
973 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
974 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
977 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
980 rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace,
981 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
987 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
989 GOTO(out, rc = PTR_ERR(fo));
991 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
995 obdo_from_la(&repbody->oa, &fti->fti_attr,
996 OFD_VALID_FLAGS | LA_UID | LA_GID);
997 tgt_drop_id(tsi->tsi_exp, &repbody->oa);
999 /* Store object version in reply */
1000 curr_version = dt_version_get(tsi->tsi_env,
1001 ofd_object_child(fo));
1002 if ((__s64)curr_version != -EOPNOTSUPP) {
1003 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1004 repbody->oa.o_data_version = curr_version;
1008 ofd_object_put(tsi->tsi_env, fo);
1011 tgt_extent_unlock(&lh, lock_mode);
1013 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1016 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1017 repbody->oa.o_flags = OBD_FL_FLUSH;
1022 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1024 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1025 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1026 struct ost_body *body = tsi->tsi_ost_body;
1027 struct ost_body *repbody;
1028 struct ldlm_resource *res;
1029 struct ofd_object *fo;
1030 struct filter_fid *ff = NULL;
1035 LASSERT(body != NULL);
1037 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1038 if (repbody == NULL)
1041 repbody->oa.o_oi = body->oa.o_oi;
1042 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1044 /* This would be very bad - accidentally truncating a file when
1045 * changing the time or similar - bug 12203. */
1046 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1047 body->oa.o_size != OBD_OBJECT_EOF) {
1048 static char mdsinum[48];
1050 if (body->oa.o_valid & OBD_MD_FLFID)
1051 snprintf(mdsinum, sizeof(mdsinum) - 1,
1052 "of parent "DFID, body->oa.o_parent_seq,
1053 body->oa.o_parent_oid, 0);
1057 CERROR("%s: setattr from %s is trying to truncate object "DFID
1058 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1059 PFID(&tsi->tsi_fid), mdsinum);
1063 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1065 GOTO(out, rc = PTR_ERR(fo));
1067 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1068 fti->fti_attr.la_valid &= ~LA_TYPE;
1070 if (body->oa.o_valid & OBD_MD_FLFID) {
1071 ff = &fti->fti_mds_fid;
1072 ofd_prepare_fidea(ff, &body->oa);
1075 /* setting objects attributes (including owner/group) */
1076 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff);
1080 obdo_from_la(&repbody->oa, &fti->fti_attr,
1081 OFD_VALID_FLAGS | LA_UID | LA_GID);
1082 tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1084 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1088 ofd_object_put(tsi->tsi_env, fo);
1091 /* we do not call this before to avoid lu_object_find() in
1092 * ->lvbo_update() holding another reference on the object.
1093 * otherwise concurrent destroy can make the object unavailable
1094 * for 2nd lu_object_find() waiting for the first reference
1095 * to go... deadlock! */
1096 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1097 &tsi->tsi_resid, LDLM_EXTENT, 0);
1099 ldlm_res_lvbo_update(res, NULL, 0);
1100 ldlm_resource_putref(res);
1106 static int ofd_orphans_destroy(const struct lu_env *env,
1107 struct obd_export *exp,
1108 struct ofd_device *ofd, struct obdo *oa)
1110 struct ofd_thread_info *info = ofd_info(env);
1111 struct lu_fid *fid = &info->fti_fid;
1112 struct ost_id *oi = &oa->o_oi;
1113 struct ofd_seq *oseq;
1114 obd_seq seq = ostid_seq(oi);
1115 obd_id end_id = ostid_id(oi);
1123 oseq = ofd_seq_get(ofd, seq);
1125 CERROR("%s: Can not find seq for "DOSTID"\n",
1126 ofd_name(ofd), POSTID(oi));
1131 last = ofd_seq_last_oid(oseq);
1134 LASSERT(exp != NULL);
1135 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1137 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1138 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1140 while (oid > end_id) {
1141 rc = fid_set_id(fid, oid);
1142 if (unlikely(rc != 0))
1145 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1146 if (rc != 0 && rc != -ENOENT) /* this is pretty fatal... */
1147 CEMERG("%s: error destroying precreated id "DFID
1148 ": rc = %d\n", ofd_name(ofd), PFID(fid), rc);
1152 ofd_seq_last_oid_set(oseq, oid);
1153 /* update last_id on disk periodically so that if we
1154 * restart * we don't need to re-scan all of the just
1155 * deleted objects. */
1156 if ((oid & 511) == 0)
1157 ofd_seq_last_oid_write(env, ofd, oseq);
1161 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1162 ofd_name(ofd), seq, oid);
1165 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1167 /* don't reuse orphan object, return last used objid */
1168 ostid_set_id(oi, last);
1175 ofd_seq_put(env, oseq);
1179 static int ofd_create_hdl(struct tgt_session_info *tsi)
1181 struct ost_body *repbody;
1182 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1183 struct obdo *rep_oa;
1184 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1185 obd_seq seq = ostid_seq(&oa->o_oi);
1186 obd_id oid = ostid_id(&oa->o_oi);
1187 struct ofd_seq *oseq;
1193 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1196 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1197 if (repbody == NULL)
1200 rep_oa = &repbody->oa;
1201 rep_oa->o_oi = oa->o_oi;
1203 LASSERT(seq >= FID_SEQ_OST_MDT0);
1204 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1206 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1208 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1210 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1211 ofd_name(ofd), seq, PTR_ERR(oseq));
1215 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1216 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1217 if (!ofd_obd(ofd)->obd_recovering ||
1218 oid > ofd_seq_last_oid(oseq)) {
1219 CERROR("%s: recreate objid "DOSTID" > last id "LPU64
1220 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1221 ofd_seq_last_oid(oseq));
1222 GOTO(out_nolock, rc = -EINVAL);
1224 /* Do nothing here, we re-create objects during recovery
1225 * upon write replay, see ofd_preprw_write() */
1226 GOTO(out_nolock, rc = 0);
1228 /* former ofd_handle_precreate */
1229 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1230 (oa->o_flags & OBD_FL_DELORPHAN)) {
1231 /* destroy orphans */
1232 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1233 tsi->tsi_exp->exp_conn_cnt) {
1234 CERROR("%s: dropping old orphan cleanup request\n",
1236 GOTO(out_nolock, rc = 0);
1238 /* This causes inflight precreates to abort and drop lock */
1239 oseq->os_destroys_in_progress = 1;
1240 mutex_lock(&oseq->os_create_lock);
1241 if (!oseq->os_destroys_in_progress) {
1242 CERROR("%s:["LPU64"] destroys_in_progress already"
1243 " cleared\n", ofd_name(ofd), seq);
1244 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1247 diff = oid - ofd_seq_last_oid(oseq);
1248 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1249 ofd_seq_last_oid(oseq), diff);
1250 if (-diff > OST_MAX_PRECREATE) {
1251 /* FIXME: should reset precreate_next_id on MDS */
1253 } else if (diff < 0) {
1254 rc = ofd_orphans_destroy(tsi->tsi_env, tsi->tsi_exp,
1256 oseq->os_destroys_in_progress = 0;
1258 /* XXX: Used by MDS for the first time! */
1259 oseq->os_destroys_in_progress = 0;
1262 mutex_lock(&oseq->os_create_lock);
1263 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1264 tsi->tsi_exp->exp_conn_cnt) {
1265 CERROR("%s: dropping old precreate request\n",
1269 /* only precreate if seq is 0, IDIF or normal and also o_id
1270 * must be specfied */
1271 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1272 !fid_seq_is_idif(seq)) || oid == 0) {
1273 diff = 1; /* shouldn't we create this right now? */
1275 diff = oid - ofd_seq_last_oid(oseq);
1276 /* Do sync create if the seq is about to used up */
1277 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1278 if (unlikely(oid >= IDIF_MAX_OID - 1))
1280 } else if (fid_seq_is_norm(seq)) {
1282 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1285 CERROR("%s : invalid o_seq "DOSTID"\n",
1286 ofd_name(ofd), POSTID(&oa->o_oi));
1287 GOTO(out, rc = -EINVAL);
1292 cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
1297 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1298 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1299 /* don't enforce grant during orphan recovery */
1300 rc = ofd_grant_create(tsi->tsi_env,
1301 ofd_obd(ofd)->obd_self_export,
1304 CDEBUG(D_HA, "%s: failed to acquire grant "
1305 "space for precreate (%d): rc = %d\n",
1306 ofd_name(ofd), diff, rc);
1311 /* This can happen if a new OST is formatted and installed
1312 * in place of an old one at the same index. Instead of
1313 * precreating potentially millions of deleted old objects
1314 * (possibly filling the OST), only precreate the last batch.
1315 * LFSCK will eventually clean up any orphans. LU-14 */
1316 if (diff > 5 * OST_MAX_PRECREATE) {
1317 diff = OST_MAX_PRECREATE / 2;
1318 LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %u "
1319 "larger than the LAST_ID "DOSTID", only "
1320 "precreating the last %u objects.\n",
1321 ofd_name(ofd), POSTID(&oa->o_oi),
1322 5 * OST_MAX_PRECREATE,
1323 POSTID(&oseq->os_oi), diff);
1324 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1328 next_id = ofd_seq_last_oid(oseq) + 1;
1329 count = ofd_precreate_batch(ofd, diff);
1331 CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1332 " at "LPU64"\n", ofd_name(ofd),
1333 count, seq, next_id);
1335 if (cfs_time_after(jiffies, enough_time)) {
1336 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1337 " created at a rate of %d/s\n",
1338 ofd_name(ofd), created,
1340 created / DISK_TIMEOUT);
1344 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1345 oseq, count, sync_trans);
1349 } else if (rc < 0) {
1354 /* some objects got created, we can return
1355 * them, even if last creation failed */
1358 CERROR("%s: unable to precreate: rc = %d\n",
1361 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1362 !(oa->o_flags & OBD_FL_DELORPHAN))
1363 ofd_grant_commit(tsi->tsi_env,
1364 ofd_obd(ofd)->obd_self_export, rc);
1366 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1369 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_CREATE,
1372 mutex_unlock(&oseq->os_create_lock);
1375 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1377 ofd_seq_put(tsi->tsi_env, oseq);
1381 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1383 const struct ost_body *body = tsi->tsi_ost_body;
1384 struct ost_body *repbody;
1385 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1386 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1387 struct lu_fid *fid = &fti->fti_fid;
1394 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1397 /* This is old case for clients before Lustre 2.4 */
1398 /* If there's a DLM request, cancel the locks mentioned in it */
1399 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1401 struct ldlm_request *dlm;
1403 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1406 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0);
1409 *fid = body->oa.o_oi.oi_fid;
1410 oid = ostid_id(&body->oa.o_oi);
1413 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1415 /* check that o_misc makes sense */
1416 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1417 count = body->oa.o_misc;
1419 count = 1; /* default case - single destroy */
1421 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1422 POSTID(&body->oa.o_oi), count);
1427 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1428 if (lrc == -ENOENT) {
1430 "%s: destroying non-existent object "DFID"\n",
1431 ofd_name(ofd), PFID(fid));
1432 /* rewrite rc with -ENOENT only if it is 0 */
1435 } else if (lrc != 0) {
1436 CERROR("%s: error destroying object "DFID": %d\n",
1437 ofd_name(ofd), PFID(fid), lrc);
1443 lrc = fid_set_id(fid, oid);
1444 if (unlikely(lrc != 0 && count > 0))
1445 GOTO(out, rc = lrc);
1448 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1454 fid_to_ostid(fid, &repbody->oa.o_oi);
1458 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1460 struct obd_statfs *osfs;
1465 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1467 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1468 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
1470 CERROR("%s: statfs failed: rc = %d\n",
1471 tgt_name(tsi->tsi_tgt), rc);
1473 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1476 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1482 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1484 struct ost_body *body = tsi->tsi_ost_body;
1485 struct ost_body *repbody;
1486 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1487 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1488 struct ofd_object *fo = NULL;
1493 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1495 /* if no objid is specified, it means "sync whole filesystem" */
1496 if (!fid_is_zero(&tsi->tsi_fid)) {
1497 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1499 RETURN(PTR_ERR(fo));
1502 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1503 fo != NULL ? ofd_object_child(fo) : NULL);
1507 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1512 repbody->oa.o_oi = body->oa.o_oi;
1513 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1515 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1517 obdo_from_la(&repbody->oa, &fti->fti_attr,
1520 /* don't return rc from getattr */
1525 ofd_object_put(tsi->tsi_env, fo);
1529 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1531 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1532 struct ost_body *repbody;
1533 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1534 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1535 struct ldlm_resource *res;
1536 struct ofd_object *fo;
1537 struct filter_fid *ff = NULL;
1539 struct lustre_handle lh = { 0, };
1546 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1547 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
1549 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1550 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1551 RETURN(err_serious(-EPROTO));
1553 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1554 if (repbody == NULL)
1555 RETURN(err_serious(-ENOMEM));
1557 /* punch start,end are passed in o_size,o_blocks throught wire */
1561 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1564 /* standard truncate optimization: if file body is completely
1565 * destroyed, don't send data back to the server. */
1567 flags |= LDLM_FL_AST_DISCARD_DATA;
1569 repbody->oa.o_oi = oa->o_oi;
1570 repbody->oa.o_valid = OBD_MD_FLID;
1572 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1573 oa->o_flags & OBD_FL_SRVLOCK;
1576 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
1582 CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
1583 ", start = "LPD64", end = "LPD64"\n", PFID(&tsi->tsi_fid),
1584 oa->o_valid, start, end);
1586 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
1589 GOTO(out, rc = PTR_ERR(fo));
1591 la_from_obdo(&info->fti_attr, oa,
1592 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
1593 info->fti_attr.la_size = start;
1594 info->fti_attr.la_valid |= LA_SIZE;
1596 if (oa->o_valid & OBD_MD_FLFID) {
1597 ff = &info->fti_mds_fid;
1598 ofd_prepare_fidea(ff, oa);
1601 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
1606 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
1610 ofd_object_put(tsi->tsi_env, fo);
1613 tgt_extent_unlock(&lh, LCK_PW);
1615 /* we do not call this before to avoid lu_object_find() in
1616 * ->lvbo_update() holding another reference on the object.
1617 * otherwise concurrent destroy can make the object unavailable
1618 * for 2nd lu_object_find() waiting for the first reference
1619 * to go... deadlock! */
1620 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
1623 ldlm_res_lvbo_update(res, NULL, 0);
1624 ldlm_resource_putref(res);
1630 static int ofd_quotactl(struct tgt_session_info *tsi)
1632 struct obd_quotactl *oqctl, *repoqc;
1637 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
1639 RETURN(err_serious(-EPROTO));
1641 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
1643 RETURN(err_serious(-ENOMEM));
1645 /* report success for quota on/off for interoperability with current MDT
1647 if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1651 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
1653 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
1659 /* High priority request handlers for OFD */
1661 /* prolong locks for the current service time of the corresponding
1662 * portal (= OST_IO_PORTAL)
1664 static inline int prolong_timeout(struct ptlrpc_request *req)
1666 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
1669 return obd_timeout / 2;
1671 return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
1675 static int ofd_prolong_one_lock(struct tgt_session_info *tsi,
1676 struct ldlm_lock *lock,
1677 struct ldlm_extent *extent, int timeout)
1680 if (lock->l_flags & LDLM_FL_DESTROYED) /* lock already cancelled */
1683 /* XXX: never try to grab resource lock here because we're inside
1684 * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
1685 * res lock and then exp_bl_list_lock. */
1687 if (!(lock->l_flags & LDLM_FL_AST_SENT))
1688 /* ignore locks not being cancelled */
1691 LDLM_DEBUG(lock, "refreshed for req x"LPU64" ext("LPU64"->"LPU64") "
1692 "to %ds.\n", tgt_ses_req(tsi)->rq_xid, extent->start,
1693 extent->end, timeout);
1695 /* OK. this is a possible lock the user holds doing I/O
1696 * let's refresh eviction timer for it */
1697 ldlm_refresh_waiting_lock(lock, timeout);
1701 static int ofd_prolong_extent_locks(struct tgt_session_info *tsi,
1702 __u64 start, __u64 end)
1704 struct obd_export *exp = tsi->tsi_exp;
1705 struct obdo *oa = &tsi->tsi_ost_body->oa;
1706 struct ldlm_extent extent = {
1710 struct ldlm_lock *lock;
1711 int timeout = prolong_timeout(tgt_ses_req(tsi));
1716 if (oa->o_valid & OBD_MD_FLHANDLE) {
1717 /* mostly a request should be covered by only one lock, try
1719 lock = ldlm_handle2lock(&oa->o_handle);
1721 /* Fast path to check if the lock covers the whole IO
1722 * region exclusively. */
1723 if (lock->l_granted_mode == LCK_PW &&
1724 ldlm_extent_contain(&lock->l_policy_data.l_extent,
1727 LASSERT(lock->l_export == exp);
1728 lock_count = ofd_prolong_one_lock(tsi, lock,
1730 LDLM_LOCK_PUT(lock);
1733 LDLM_LOCK_PUT(lock);
1737 spin_lock_bh(&exp->exp_bl_list_lock);
1738 list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
1739 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1740 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
1742 if (!ldlm_res_eq(&tsi->tsi_resid, &lock->l_resource->lr_name))
1745 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
1749 lock_count += ofd_prolong_one_lock(tsi, lock, &extent, timeout);
1751 spin_unlock_bh(&exp->exp_bl_list_lock);
1757 * Returns 1 if the given PTLRPC matches the given LDLM lock, or 0 if it does
1760 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
1761 struct ldlm_lock *lock)
1763 struct niobuf_remote *rnb;
1764 struct obd_ioobj *ioo;
1766 struct ldlm_extent ext;
1767 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
1771 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1772 LASSERT(ioo != NULL);
1774 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1775 LASSERT(rnb != NULL);
1777 ext.start = rnb->offset;
1778 rnb += ioo->ioo_bufcnt - 1;
1779 ext.end = rnb->offset + rnb->len - 1;
1781 LASSERT(lock->l_resource != NULL);
1782 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
1786 if (opc == OST_READ)
1789 if (!(lock->l_granted_mode & mode))
1792 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
1796 * High-priority queue request check for whether the given PTLRPC request
1797 * (\a req) is blocking an LDLM lock cancel.
1799 * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1800 * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1802 * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue. This
1803 * function looks only at OST_READs and OST_WRITEs.
1805 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
1807 struct tgt_session_info *tsi;
1808 struct obd_ioobj *ioo;
1809 struct niobuf_remote *rnb;
1815 /* Don't use tgt_ses_info() to get session info, because lock_match()
1816 * can be called while request has no processing thread yet. */
1817 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1818 LASSERT(tsi != NULL);
1821 * Use LASSERT below because malformed RPCs should have
1822 * been filtered out in tgt_hpreq_handler().
1824 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1825 LASSERT(ioo != NULL);
1827 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1828 LASSERT(rnb != NULL);
1829 LASSERT(!(rnb->flags & OBD_BRW_SRVLOCK));
1831 start = rnb->offset;
1832 rnb += ioo->ioo_bufcnt - 1;
1833 end = rnb->offset + rnb->len - 1;
1835 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
1836 " ("LPU64"->"LPU64")\n",
1837 tgt_name(tsi->tsi_tgt), current->comm,
1838 PFID(&tsi->tsi_fid), start, end);
1840 lock_count = ofd_prolong_extent_locks(tsi, start, end);
1842 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1843 tgt_name(tsi->tsi_tgt), lock_count, req);
1845 RETURN(lock_count > 0);
1848 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
1850 ofd_rw_hpreq_check(req);
1854 * Like tgt_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1856 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
1857 struct ldlm_lock *lock)
1859 struct tgt_session_info *tsi;
1861 /* Don't use tgt_ses_info() to get session info, because lock_match()
1862 * can be called while request has no processing thread yet. */
1863 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1864 LASSERT(tsi != NULL);
1866 LASSERT(tsi->tsi_ost_body != NULL);
1867 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
1868 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1875 * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1877 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
1879 struct tgt_session_info *tsi;
1885 /* Don't use tgt_ses_info() to get session info, because lock_match()
1886 * can be called while request has no processing thread yet. */
1887 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1888 LASSERT(tsi != NULL);
1889 oa = &tsi->tsi_ost_body->oa;
1891 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
1892 oa->o_flags & OBD_FL_SRVLOCK));
1895 "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1896 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
1897 tsi->tsi_resid.name[1], oa->o_size, oa->o_blocks);
1899 lock_count = ofd_prolong_extent_locks(tsi, oa->o_size, oa->o_blocks);
1901 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1902 tgt_name(tsi->tsi_tgt), lock_count, req);
1904 RETURN(lock_count > 0);
1907 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
1909 ofd_punch_hpreq_check(req);
1912 struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
1913 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
1914 .hpreq_check = ofd_rw_hpreq_check,
1915 .hpreq_fini = ofd_rw_hpreq_fini
1918 struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
1919 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
1920 .hpreq_check = ofd_punch_hpreq_check,
1921 .hpreq_fini = ofd_punch_hpreq_fini
1924 /** Assign high priority operations to the IO requests */
1925 static void ofd_hp_brw(struct tgt_session_info *tsi)
1927 struct niobuf_remote *rnb;
1928 struct obd_ioobj *ioo;
1932 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1933 LASSERT(ioo != NULL); /* must exist after request preprocessing */
1934 if (ioo->ioo_bufcnt > 0) {
1935 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
1936 LASSERT(rnb != NULL); /* must exist after request preprocessing */
1938 /* no high priority if server lock is needed */
1939 if (rnb->flags & OBD_BRW_SRVLOCK)
1942 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
1945 static void ofd_hp_punch(struct tgt_session_info *tsi)
1947 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
1948 /* no high-priority if server lock is needed */
1949 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1950 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK)
1952 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
1955 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
1956 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
1957 #define OST_BRW_READ OST_READ
1958 #define OST_BRW_WRITE OST_WRITE
1960 static struct tgt_handler ofd_tgt_handlers[] = {
1961 TGT_RPC_HANDLER(OST_FIRST_OPC,
1962 0, OST_CONNECT, tgt_connect,
1963 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1964 TGT_RPC_HANDLER(OST_FIRST_OPC,
1965 0, OST_DISCONNECT, tgt_disconnect,
1966 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
1967 TGT_RPC_HANDLER(OST_FIRST_OPC,
1968 0, OST_SET_INFO, ofd_set_info_hdl,
1969 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
1970 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
1971 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
1972 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
1973 OST_SETATTR, ofd_setattr_hdl),
1974 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
1975 OST_CREATE, ofd_create_hdl),
1976 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
1977 OST_DESTROY, ofd_destroy_hdl),
1978 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
1979 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
1980 OST_BRW_READ, tgt_brw_read,
1982 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
1983 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
1985 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
1986 OST_PUNCH, ofd_punch_hdl,
1988 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
1989 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
1992 static struct tgt_opc_slice ofd_common_slice[] = {
1994 .tos_opc_start = OST_FIRST_OPC,
1995 .tos_opc_end = OST_LAST_OPC,
1996 .tos_hs = ofd_tgt_handlers
1999 .tos_opc_start = OBD_FIRST_OPC,
2000 .tos_opc_end = OBD_LAST_OPC,
2001 .tos_hs = tgt_obd_handlers
2004 .tos_opc_start = LDLM_FIRST_OPC,
2005 .tos_opc_end = LDLM_LAST_OPC,
2006 .tos_hs = tgt_dlm_handlers
2009 .tos_opc_start = UPDATE_OBJ,
2010 .tos_opc_end = UPDATE_LAST_OPC,
2011 .tos_hs = tgt_out_handlers
2014 .tos_opc_start = SEQ_FIRST_OPC,
2015 .tos_opc_end = SEQ_LAST_OPC,
2016 .tos_hs = seq_handlers
2023 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2024 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2026 const char *dev = lustre_cfg_string(cfg, 0);
2027 struct ofd_thread_info *info = NULL;
2028 struct obd_device *obd;
2029 struct obd_statfs *osfs;
2034 obd = class_name2obd(dev);
2036 CERROR("Cannot find obd with name %s\n", dev);
2040 rc = lu_env_refill((struct lu_env *)env);
2044 obd->u.obt.obt_magic = OBT_MAGIC;
2046 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2047 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2049 spin_lock_init(&m->ofd_flags_lock);
2050 m->ofd_raid_degraded = 0;
2051 m->ofd_syncjournal = 0;
2053 m->ofd_grant_compat_disable = 0;
2054 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2057 spin_lock_init(&m->ofd_osfs_lock);
2058 m->ofd_osfs_age = cfs_time_shift_64(-1000);
2059 m->ofd_osfs_unstable = 0;
2060 m->ofd_statfs_inflight = 0;
2061 m->ofd_osfs_inflight = 0;
2064 spin_lock_init(&m->ofd_grant_lock);
2065 m->ofd_tot_dirty = 0;
2066 m->ofd_tot_granted = 0;
2067 m->ofd_tot_pending = 0;
2068 m->ofd_seq_count = 0;
2070 spin_lock_init(&m->ofd_batch_lock);
2071 rwlock_init(&obd->u.filter.fo_sptlrpc_lock);
2072 sptlrpc_rule_set_init(&obd->u.filter.fo_sptlrpc_rset);
2074 obd->u.filter.fo_fl_oss_capa = 0;
2075 CFS_INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
2076 obd->u.filter.fo_capa_hash = init_capa_hash();
2077 if (obd->u.filter.fo_capa_hash == NULL)
2080 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2081 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2082 /* set this lu_device to obd, because error handling need it */
2083 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2085 rc = ofd_procfs_init(m);
2087 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2091 /* No connection accepted until configurations will finish */
2092 spin_lock(&obd->obd_dev_lock);
2093 obd->obd_no_conn = 1;
2094 spin_unlock(&obd->obd_dev_lock);
2095 obd->obd_replayable = 1;
2096 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2097 char *str = lustre_cfg_string(cfg, 4);
2099 if (strchr(str, 'n')) {
2100 CWARN("%s: recovery disabled\n", obd->obd_name);
2101 obd->obd_replayable = 0;
2105 info = ofd_info_init(env, NULL);
2109 rc = ofd_stack_init(env, m, cfg);
2111 CERROR("Can't init device stack, rc %d\n", rc);
2112 GOTO(err_fini_proc, rc);
2115 ofd_procfs_add_brw_stats_symlink(m);
2117 /* populate cached statfs data */
2118 osfs = &ofd_info(env)->fti_u.osfs;
2119 rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
2121 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
2122 GOTO(err_fini_stack, rc);
2124 if (!IS_PO2(osfs->os_bsize)) {
2125 CERROR("%s: blocksize (%d) is not a power of 2\n",
2126 obd->obd_name, osfs->os_bsize);
2127 GOTO(err_fini_stack, rc = -EPROTO);
2129 m->ofd_blockbits = fls(osfs->os_bsize) - 1;
2131 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2132 if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
2133 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2135 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2136 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2137 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2138 LDLM_NAMESPACE_SERVER,
2139 LDLM_NAMESPACE_GREEDY,
2141 if (m->ofd_namespace == NULL)
2142 GOTO(err_fini_stack, rc = -ENOMEM);
2143 /* set obd_namespace for compatibility with old code */
2144 obd->obd_namespace = m->ofd_namespace;
2145 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2146 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2147 m->ofd_namespace->ns_lvbp = m;
2149 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2150 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2152 dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
2154 /* Allow at most ddp_grant_reserved% of the available filesystem space
2155 * to be granted to clients, so that any errors in the grant overhead
2156 * calculations do not allow granting more space to clients than can be
2157 * written. Assumes that in aggregate the grant overhead calculations do
2158 * not have more than ddp_grant_reserved% estimation error in them. */
2159 m->ofd_grant_ratio =
2160 ofd_grant_ratio_conv(m->ofd_dt_conf.ddp_grant_reserved);
2162 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2163 OBD_FAIL_OST_ALL_REQUEST_NET,
2164 OBD_FAIL_OST_ALL_REPLY_NET);
2166 GOTO(err_free_ns, rc);
2168 rc = ofd_fs_setup(env, m, obd);
2170 GOTO(err_fini_lut, rc);
2174 tgt_fini(env, &m->ofd_lut);
2176 ldlm_namespace_free(m->ofd_namespace, 0, obd->obd_force);
2177 obd->obd_namespace = m->ofd_namespace = NULL;
2179 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
2185 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
2187 struct obd_device *obd = ofd_obd(m);
2188 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
2190 lfsck_stop(env, m->ofd_osd, true);
2191 lfsck_degister(env, m->ofd_osd);
2192 target_recovery_fini(obd);
2193 obd_exports_barrier(obd);
2194 obd_zombie_barrier();
2196 tgt_fini(env, &m->ofd_lut);
2197 ofd_fs_cleanup(env, m);
2199 ofd_free_capa_keys(m);
2200 cleanup_capa_hash(obd->u.filter.fo_capa_hash);
2202 if (m->ofd_namespace != NULL) {
2203 ldlm_namespace_free(m->ofd_namespace, NULL,
2204 d->ld_obd->obd_force);
2205 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
2208 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
2210 LASSERT(atomic_read(&d->ld_ref) == 0);
2211 server_put_mount(obd->obd_name);
2215 static struct lu_device *ofd_device_fini(const struct lu_env *env,
2216 struct lu_device *d)
2219 ofd_fini(env, ofd_dev(d));
2223 static struct lu_device *ofd_device_free(const struct lu_env *env,
2224 struct lu_device *d)
2226 struct ofd_device *m = ofd_dev(d);
2228 dt_device_fini(&m->ofd_dt_dev);
2233 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
2234 struct lu_device_type *t,
2235 struct lustre_cfg *cfg)
2237 struct ofd_device *m;
2238 struct lu_device *l;
2243 return ERR_PTR(-ENOMEM);
2245 l = &m->ofd_dt_dev.dd_lu_dev;
2246 dt_device_init(&m->ofd_dt_dev, t);
2247 rc = ofd_init0(env, m, t, cfg);
2249 ofd_device_free(env, l);
2256 /* thread context key constructor/destructor */
2257 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2259 static void ofd_key_exit(const struct lu_context *ctx,
2260 struct lu_context_key *key, void *data)
2262 struct ofd_thread_info *info = data;
2264 info->fti_env = NULL;
2265 info->fti_exp = NULL;
2268 info->fti_pre_version = 0;
2271 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2274 struct lu_context_key ofd_thread_key = {
2275 .lct_tags = LCT_DT_THREAD,
2276 .lct_init = ofd_key_init,
2277 .lct_fini = ofd_key_fini,
2278 .lct_exit = ofd_key_exit
2281 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
2282 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
2284 static struct lu_device_type_operations ofd_device_type_ops = {
2285 .ldto_init = ofd_type_init,
2286 .ldto_fini = ofd_type_fini,
2288 .ldto_start = ofd_type_start,
2289 .ldto_stop = ofd_type_stop,
2291 .ldto_device_alloc = ofd_device_alloc,
2292 .ldto_device_free = ofd_device_free,
2293 .ldto_device_fini = ofd_device_fini
2296 static struct lu_device_type ofd_device_type = {
2297 .ldt_tags = LU_DEVICE_DT,
2298 .ldt_name = LUSTRE_OST_NAME,
2299 .ldt_ops = &ofd_device_type_ops,
2300 .ldt_ctx_tags = LCT_DT_THREAD
2303 int __init ofd_init(void)
2305 struct lprocfs_static_vars lvars;
2308 rc = lu_kmem_init(ofd_caches);
2312 rc = ofd_fmd_init();
2314 lu_kmem_fini(ofd_caches);
2318 lprocfs_ofd_init_vars(&lvars);
2320 rc = class_register_type(&ofd_obd_ops, NULL, NULL,
2321 #ifndef HAVE_ONLY_PROCFS_SEQ
2324 LUSTRE_OST_NAME, &ofd_device_type);
2328 void __exit ofd_exit(void)
2331 lu_kmem_fini(ofd_caches);
2332 class_unregister_type(LUSTRE_OST_NAME);
2335 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
2336 MODULE_DESCRIPTION("Lustre Object Filtering Device");
2337 MODULE_LICENSE("GPL");
2339 module_init(ofd_init);
2340 module_exit(ofd_exit);