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);
189 /* For interoperability, see mdt_interop_param[]. */
190 static struct cfg_interop_param ofd_interop_param[] = {
191 { "ost.quota_type", NULL },
195 /* Some parameters were moved from ofd to osd and only their
196 * symlinks were kept in ofd by LU-3106. They are:
197 * -writehthrough_cache_enable
198 * -readcache_max_filese
201 * Since they are not included by the static lprocfs var list,
202 * a pre-check is added for them to avoid "unknown param" error
203 * message confuses the customer. If they are matched in this
204 * check, they will be passed to the osd directly.
206 static bool match_symlink_param(char *param)
211 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
212 sval = strchr(param, '=');
214 paramlen = sval - param;
215 if (strncmp(param, "writethrough_cache_enable",
217 strncmp(param, "readcache_max_filesize",
219 strncmp(param, "read_cache_enable",
221 strncmp(param, "brw_stats", paramlen) == 0)
229 /* used by MGS to process specific configurations */
230 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
231 struct lustre_cfg *cfg)
233 struct ofd_device *m = ofd_dev(d);
234 struct dt_device *dt_next = m->ofd_osd;
235 struct lu_device *next = &dt_next->dd_lu_dev;
240 switch (cfg->lcfg_command) {
242 struct lprocfs_static_vars lvars;
244 /* For interoperability */
245 struct cfg_interop_param *ptr = NULL;
246 struct lustre_cfg *old_cfg = NULL;
249 param = lustre_cfg_string(cfg, 1);
251 CERROR("param is empty\n");
256 ptr = class_find_old_param(param, ofd_interop_param);
258 if (ptr->new_param == NULL) {
260 CWARN("For interoperability, skip this %s."
261 " It is obsolete.\n", ptr->old_param);
265 CWARN("Found old param %s, changed it to %s.\n",
266 ptr->old_param, ptr->new_param);
269 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
276 if (match_symlink_param(param)) {
277 rc = next->ld_ops->ldo_process_config(env, next, cfg);
281 lprocfs_ofd_init_vars(&lvars);
282 rc = class_process_proc_param(PARAM_OST, lvars.obd_vars, cfg,
284 if (rc > 0 || rc == -ENOSYS) {
285 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
287 /* we don't understand; pass it on */
288 rc = next->ld_ops->ldo_process_config(env, next, cfg);
292 case LCFG_SPTLRPC_CONF: {
297 /* others are passed further */
298 rc = next->ld_ops->ldo_process_config(env, next, cfg);
304 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
305 const struct lu_object_conf *conf)
307 struct ofd_device *d = ofd_dev(o->lo_dev);
308 struct lu_device *under;
309 struct lu_object *below;
314 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
315 PFID(lu_object_fid(o)));
317 under = &d->ofd_osd->dd_lu_dev;
318 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
320 lu_object_add(o, below);
327 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
329 struct ofd_object *of = ofd_obj(o);
330 struct lu_object_header *h;
335 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
336 PFID(lu_object_fid(o)));
339 lu_object_header_fini(h);
340 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
344 static int ofd_object_print(const struct lu_env *env, void *cookie,
345 lu_printer_t p, const struct lu_object *o)
347 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
350 struct lu_object_operations ofd_obj_ops = {
351 .loo_object_init = ofd_object_init,
352 .loo_object_free = ofd_object_free,
353 .loo_object_print = ofd_object_print
356 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
357 const struct lu_object_header *hdr,
360 struct ofd_object *of;
364 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
367 struct lu_object_header *h;
369 o = &of->ofo_obj.do_lu;
371 lu_object_header_init(h);
372 lu_object_init(o, h, d);
373 lu_object_add_top(h, o);
374 o->lo_ops = &ofd_obj_ops;
381 extern int ost_handle(struct ptlrpc_request *req);
383 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
384 enum lfsck_events event)
386 struct ofd_device *ofd = data;
387 struct obd_device *obd = ofd_obd(ofd);
390 case LE_LASTID_REBUILDING:
391 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
392 "on the device until the LAST_ID rebuilt successfully.\n",
394 down_write(&ofd->ofd_lastid_rwsem);
395 ofd->ofd_lastid_rebuilding = 1;
396 up_write(&ofd->ofd_lastid_rwsem);
398 case LE_LASTID_REBUILT: {
399 down_write(&ofd->ofd_lastid_rwsem);
400 ofd_seqs_free(env, ofd);
401 ofd->ofd_lastid_rebuilding = 0;
402 ofd->ofd_lastid_gen++;
403 up_write(&ofd->ofd_lastid_rwsem);
407 CERROR("%s: unknown lfsck event: rc = %d\n",
408 ofd_obd(ofd)->obd_name, event);
415 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
416 struct lu_device *dev)
418 struct ofd_thread_info *info;
419 struct ofd_device *ofd = ofd_dev(dev);
420 struct obd_device *obd = ofd_obd(ofd);
421 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
422 struct lfsck_start_param lsp;
427 info = ofd_info_init(env, NULL);
431 /* initialize lower device */
432 rc = next->ld_ops->ldo_prepare(env, dev, next);
436 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
437 ofd_lfsck_out_notify, ofd, false);
439 CERROR("%s: failed to initialize lfsck: rc = %d\n",
444 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
445 /* The LFSCK instance is registered just now, so it must be there when
446 * register the namespace to such instance. */
447 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
449 lsp.lsp_start = NULL;
450 lsp.lsp_index_valid = 0;
451 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
453 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
458 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
459 LASSERT(obd->obd_no_conn);
460 spin_lock(&obd->obd_dev_lock);
461 obd->obd_no_conn = 0;
462 spin_unlock(&obd->obd_dev_lock);
464 if (obd->obd_recovering == 0)
465 ofd_postrecov(env, ofd);
470 static int ofd_recovery_complete(const struct lu_env *env,
471 struct lu_device *dev)
473 struct ofd_device *ofd = ofd_dev(dev);
474 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
475 int rc = 0, max_precreate;
479 /* Grant space for object precreation on the self export.
480 * This initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
481 * is enough to create 10k objects. More space is then acquired for
482 * precreation in ofd_grant_create().
484 max_precreate = OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace / 2;
485 ofd_grant_connect(env, dev->ld_obd->obd_self_export, max_precreate,
487 rc = next->ld_ops->ldo_recovery_complete(env, next);
491 static struct lu_device_operations ofd_lu_ops = {
492 .ldo_object_alloc = ofd_object_alloc,
493 .ldo_process_config = ofd_process_config,
494 .ldo_recovery_complete = ofd_recovery_complete,
495 .ldo_prepare = ofd_prepare,
498 static int ofd_procfs_init(struct ofd_device *ofd)
500 struct lprocfs_static_vars lvars;
501 struct obd_device *obd = ofd_obd(ofd);
502 cfs_proc_dir_entry_t *entry;
507 /* lprocfs must be setup before the ofd so state can be safely added
508 * to /proc incrementally as the ofd is setup */
509 lprocfs_ofd_init_vars(&lvars);
510 rc = lprocfs_obd_setup(obd, lvars.obd_vars);
512 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
517 rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
519 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
521 GOTO(obd_cleanup, rc);
524 obd->obd_uses_nid_stats = 1;
526 entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
529 CERROR("%s: error %d setting up lprocfs for %s\n",
530 obd->obd_name, rc, "exports");
531 GOTO(obd_cleanup, rc);
533 obd->obd_proc_exports_entry = entry;
535 entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
536 lprocfs_nid_stats_clear_read,
537 lprocfs_nid_stats_clear_write, obd, NULL);
540 CERROR("%s: add proc entry 'clear' failed: %d.\n",
542 GOTO(obd_cleanup, rc);
545 ofd_stats_counter_init(obd->obd_stats);
547 rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
548 ofd_stats_counter_init);
550 GOTO(remove_entry_clear, rc);
553 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
555 lprocfs_obd_cleanup(obd);
556 lprocfs_free_obd_stats(obd);
561 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
563 struct obd_device *obd = ofd_obd(ofd);
564 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
565 cfs_proc_dir_entry_t *osd_root = osd_obd->obd_type->typ_procroot;
566 cfs_proc_dir_entry_t *osd_dir;
568 osd_dir = lprocfs_srch(osd_root, obd->obd_name);
572 if (lprocfs_srch(osd_dir, "brw_stats") != NULL)
573 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
574 "../../%s/%s/brw_stats",
575 osd_root->name, osd_dir->name);
577 if (lprocfs_srch(osd_dir, "read_cache_enable") != NULL)
578 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
579 "../../%s/%s/read_cache_enable",
580 osd_root->name, osd_dir->name);
582 if (lprocfs_srch(osd_dir, "readcache_max_filesize") != NULL)
583 lprocfs_add_symlink("readcache_max_filesize",
585 "../../%s/%s/readcache_max_filesize",
586 osd_root->name, osd_dir->name);
588 if (lprocfs_srch(osd_dir, "writethrough_cache_enable") != NULL)
589 lprocfs_add_symlink("writethrough_cache_enable",
591 "../../%s/%s/writethrough_cache_enable",
592 osd_root->name, osd_dir->name);
595 static void ofd_procfs_fini(struct ofd_device *ofd)
597 struct obd_device *obd = ofd_obd(ofd);
599 lprocfs_remove_proc_entry("writethrough_cache_enable",
600 obd->obd_proc_entry);
601 lprocfs_remove_proc_entry("readcache_max_filesize",
602 obd->obd_proc_entry);
603 lprocfs_remove_proc_entry("read_cache_enable", obd->obd_proc_entry);
604 lprocfs_remove_proc_entry("brw_stats", obd->obd_proc_entry);
605 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
606 lprocfs_free_per_client_stats(obd);
607 lprocfs_obd_cleanup(obd);
608 lprocfs_free_obd_stats(obd);
609 lprocfs_job_stats_fini(obd);
612 extern int ost_handle(struct ptlrpc_request *req);
614 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
616 return seq_site_fini(env, &ofd->ofd_seq_site);
619 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
621 struct seq_server_site *ss = &ofd->ofd_seq_site;
622 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
623 char *obd_name = ofd_name(ofd);
627 ss = &ofd->ofd_seq_site;
628 lu->ld_site->ld_seq_site = ss;
629 ss->ss_lu = lu->ld_site;
630 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
632 OBD_ALLOC_PTR(ss->ss_server_seq);
633 if (ss->ss_server_seq == NULL)
634 GOTO(out_free, rc = -ENOMEM);
636 OBD_ALLOC(name, strlen(obd_name) + 10);
638 OBD_FREE_PTR(ss->ss_server_seq);
639 ss->ss_server_seq = NULL;
640 GOTO(out_free, rc = -ENOMEM);
643 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
644 LUSTRE_SEQ_SERVER, ss);
646 CERROR("%s : seq server init error %d\n", obd_name, rc);
649 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
651 OBD_ALLOC_PTR(ss->ss_client_seq);
652 if (ss->ss_client_seq == NULL)
653 GOTO(out_free, rc = -ENOMEM);
655 snprintf(name, strlen(obd_name) + 6, "%p-super", obd_name);
656 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
659 CERROR("%s : seq client init error %d\n", obd_name, rc);
662 OBD_FREE(name, strlen(obd_name) + 10);
665 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
669 if (ss->ss_server_seq) {
670 seq_server_fini(ss->ss_server_seq, env);
671 OBD_FREE_PTR(ss->ss_server_seq);
672 ss->ss_server_seq = NULL;
675 if (ss->ss_client_seq) {
676 seq_client_fini(ss->ss_client_seq);
677 OBD_FREE_PTR(ss->ss_client_seq);
678 ss->ss_client_seq = NULL;
682 OBD_FREE(name, strlen(obd_name) + 10);
690 int ofd_set_info_hdl(struct tgt_session_info *tsi)
692 struct ptlrpc_request *req = tgt_ses_req(tsi);
693 struct ost_body *body = NULL, *repbody;
694 void *key, *val = NULL;
695 int keylen, vallen, rc = 0;
696 bool is_grant_shrink;
697 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
701 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
703 DEBUG_REQ(D_HA, req, "no set_info key");
704 RETURN(err_serious(-EFAULT));
706 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
709 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
711 DEBUG_REQ(D_HA, req, "no set_info val");
712 RETURN(err_serious(-EFAULT));
714 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
717 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
719 /* In this case the value is actually an RMF_OST_BODY, so we
720 * transmutate the type of this PTLRPC */
721 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
723 rc = req_capsule_server_pack(tsi->tsi_pill);
727 if (is_grant_shrink) {
728 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
730 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
733 /** handle grant shrink, similar to a read request */
734 ofd_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
736 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
738 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
740 } else if (KEY_IS(KEY_CAPA_KEY)) {
741 rc = ofd_update_capa_key(ofd, val);
742 } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
743 rc = tgt_adapt_sptlrpc_conf(tsi->tsi_tgt, 0);
745 CERROR("%s: Unsupported key %s\n",
746 tgt_name(tsi->tsi_tgt), (char *)key);
749 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
755 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
756 struct lu_fid *fid, struct ll_user_fiemap *fiemap)
758 struct ofd_object *fo;
761 fo = ofd_object_find(env, ofd, fid);
763 CERROR("%s: error finding object "DFID"\n",
764 ofd_name(ofd), PFID(fid));
768 ofd_read_lock(env, fo);
769 if (ofd_object_exists(fo))
770 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
773 ofd_read_unlock(env, fo);
774 ofd_object_put(env, fo);
778 struct locked_region {
780 struct lustre_handle lh;
783 static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
784 unsigned long long begin, unsigned long long end,
787 struct locked_region *region = NULL;
791 LASSERT(begin <= end);
792 OBD_ALLOC_PTR(region);
796 rc = tgt_extent_lock(ns, res_id, begin, end, ®ion->lh,
801 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
803 cfs_list_add(®ion->list, locked);
808 static int lock_zero_regions(struct ldlm_namespace *ns,
809 struct ldlm_res_id *res_id,
810 struct ll_user_fiemap *fiemap,
813 __u64 begin = fiemap->fm_start;
816 struct ll_fiemap_extent *fiemap_start = fiemap->fm_extents;
820 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
821 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
822 if (fiemap_start[i].fe_logical > begin) {
823 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
824 begin, fiemap_start[i].fe_logical);
825 rc = lock_region(ns, res_id, begin,
826 fiemap_start[i].fe_logical, locked);
831 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
834 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
835 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
836 begin, fiemap->fm_start + fiemap->fm_length);
837 rc = lock_region(ns, res_id, begin,
838 fiemap->fm_start + fiemap->fm_length, locked);
844 static void unlock_zero_regions(struct ldlm_namespace *ns, cfs_list_t *locked)
846 struct locked_region *entry, *temp;
848 cfs_list_for_each_entry_safe(entry, temp, locked, list) {
849 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
850 tgt_extent_unlock(&entry->lh, LCK_PR);
851 cfs_list_del(&entry->list);
856 int ofd_get_info_hdl(struct tgt_session_info *tsi)
858 struct obd_export *exp = tsi->tsi_exp;
859 struct ofd_device *ofd = ofd_exp(exp);
860 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
863 int replylen, rc = 0;
867 /* this common part for get_info rpc */
868 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
870 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
871 RETURN(err_serious(-EPROTO));
873 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
876 if (KEY_IS(KEY_LAST_ID)) {
878 struct ofd_seq *oseq;
880 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
881 rc = req_capsule_server_pack(tsi->tsi_pill);
883 RETURN(err_serious(rc));
885 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
887 oseq = ofd_seq_load(tsi->tsi_env, ofd,
888 (obd_seq)exp->exp_filter_data.fed_group);
892 *last_id = ofd_seq_last_oid(oseq);
893 ofd_seq_put(tsi->tsi_env, oseq);
894 } else if (KEY_IS(KEY_FIEMAP)) {
895 struct ll_fiemap_info_key *fm_key;
896 struct ll_user_fiemap *fiemap;
899 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
901 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
902 rc = tgt_validate_obdo(tsi, &fm_key->oa);
904 RETURN(err_serious(rc));
906 fid = &fm_key->oa.o_oi.oi_fid;
908 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
910 replylen = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
911 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
912 RCL_SERVER, replylen);
914 rc = req_capsule_server_pack(tsi->tsi_pill);
916 RETURN(err_serious(rc));
918 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
922 *fiemap = fm_key->fiemap;
923 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
925 /* LU-3219: Lock the sparse areas to make sure dirty
926 * flushed back from client, then call fiemap again. */
927 if (fm_key->oa.o_valid & OBD_MD_FLFLAGS &&
928 fm_key->oa.o_flags & OBD_FL_SRVLOCK) {
929 cfs_list_t locked = CFS_LIST_HEAD_INIT(locked);
931 ost_fid_build_resid(fid, &fti->fti_resid);
932 rc = lock_zero_regions(ofd->ofd_namespace,
933 &fti->fti_resid, fiemap,
935 if (rc == 0 && !cfs_list_empty(&locked)) {
936 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
938 unlock_zero_regions(ofd->ofd_namespace,
942 } else if (KEY_IS(KEY_LAST_FID)) {
943 struct ofd_device *ofd = ofd_exp(exp);
944 struct ofd_seq *oseq;
948 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
949 rc = req_capsule_server_pack(tsi->tsi_pill);
951 RETURN(err_serious(rc));
953 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
955 RETURN(err_serious(-EPROTO));
957 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
959 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
963 oseq = ofd_seq_load(tsi->tsi_env, ofd,
964 ostid_seq(&fti->fti_ostid));
966 RETURN(PTR_ERR(oseq));
968 rc = ostid_to_fid(fid, &oseq->os_oi,
969 ofd->ofd_lut.lut_lsd.lsd_osd_index);
973 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
976 ofd_seq_put(tsi->tsi_env, oseq);
978 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
982 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
988 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
990 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
991 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
992 struct ost_body *repbody;
993 struct lustre_handle lh = { 0 };
994 struct ofd_object *fo;
996 ldlm_mode_t lock_mode = LCK_PR;
1001 LASSERT(tsi->tsi_ost_body != NULL);
1003 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1004 if (repbody == NULL)
1007 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1008 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1010 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1011 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1014 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1017 rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace,
1018 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1024 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1026 GOTO(out, rc = PTR_ERR(fo));
1028 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1032 obdo_from_la(&repbody->oa, &fti->fti_attr,
1033 OFD_VALID_FLAGS | LA_UID | LA_GID);
1034 tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1036 /* Store object version in reply */
1037 curr_version = dt_version_get(tsi->tsi_env,
1038 ofd_object_child(fo));
1039 if ((__s64)curr_version != -EOPNOTSUPP) {
1040 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1041 repbody->oa.o_data_version = curr_version;
1045 ofd_object_put(tsi->tsi_env, fo);
1048 tgt_extent_unlock(&lh, lock_mode);
1050 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1053 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1054 repbody->oa.o_flags = OBD_FL_FLUSH;
1059 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1061 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1062 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1063 struct ost_body *body = tsi->tsi_ost_body;
1064 struct ost_body *repbody;
1065 struct ldlm_resource *res;
1066 struct ofd_object *fo;
1067 struct filter_fid *ff = NULL;
1072 LASSERT(body != NULL);
1074 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1075 if (repbody == NULL)
1078 repbody->oa.o_oi = body->oa.o_oi;
1079 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1081 /* This would be very bad - accidentally truncating a file when
1082 * changing the time or similar - bug 12203. */
1083 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1084 body->oa.o_size != OBD_OBJECT_EOF) {
1085 static char mdsinum[48];
1087 if (body->oa.o_valid & OBD_MD_FLFID)
1088 snprintf(mdsinum, sizeof(mdsinum) - 1,
1089 "of parent "DFID, body->oa.o_parent_seq,
1090 body->oa.o_parent_oid, 0);
1094 CERROR("%s: setattr from %s is trying to truncate object "DFID
1095 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1096 PFID(&tsi->tsi_fid), mdsinum);
1100 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1102 GOTO(out, rc = PTR_ERR(fo));
1104 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1105 fti->fti_attr.la_valid &= ~LA_TYPE;
1107 if (body->oa.o_valid & OBD_MD_FLFID) {
1108 ff = &fti->fti_mds_fid;
1109 ofd_prepare_fidea(ff, &body->oa);
1112 /* setting objects attributes (including owner/group) */
1113 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff);
1117 obdo_from_la(&repbody->oa, &fti->fti_attr,
1118 OFD_VALID_FLAGS | LA_UID | LA_GID);
1119 tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1121 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1125 ofd_object_put(tsi->tsi_env, fo);
1128 /* we do not call this before to avoid lu_object_find() in
1129 * ->lvbo_update() holding another reference on the object.
1130 * otherwise concurrent destroy can make the object unavailable
1131 * for 2nd lu_object_find() waiting for the first reference
1132 * to go... deadlock! */
1133 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1134 &tsi->tsi_resid, LDLM_EXTENT, 0);
1136 ldlm_res_lvbo_update(res, NULL, 0);
1137 ldlm_resource_putref(res);
1143 static int ofd_orphans_destroy(const struct lu_env *env,
1144 struct obd_export *exp,
1145 struct ofd_device *ofd, struct obdo *oa)
1147 struct ofd_thread_info *info = ofd_info(env);
1148 struct lu_fid *fid = &info->fti_fid;
1149 struct ost_id *oi = &oa->o_oi;
1150 struct ofd_seq *oseq;
1151 obd_seq seq = ostid_seq(oi);
1152 obd_id end_id = ostid_id(oi);
1160 oseq = ofd_seq_get(ofd, seq);
1162 CERROR("%s: Can not find seq for "DOSTID"\n",
1163 ofd_name(ofd), POSTID(oi));
1168 last = ofd_seq_last_oid(oseq);
1171 LASSERT(exp != NULL);
1172 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1174 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1177 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1178 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1180 while (oid > end_id) {
1181 rc = fid_set_id(fid, oid);
1182 if (unlikely(rc != 0))
1185 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1186 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1187 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1188 /* this is pretty fatal... */
1189 CEMERG("%s: error destroying precreated id "
1191 ofd_name(ofd), PFID(fid), rc);
1195 ofd_seq_last_oid_set(oseq, oid);
1196 /* update last_id on disk periodically so that if we
1197 * restart * we don't need to re-scan all of the just
1198 * deleted objects. */
1199 if ((oid & 511) == 0)
1200 ofd_seq_last_oid_write(env, ofd, oseq);
1204 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1205 ofd_name(ofd), seq, oid);
1209 ofd_seq_last_oid_set(oseq, oid);
1210 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1212 /* don't reuse orphan object, return last used objid */
1213 ostid_set_id(oi, last);
1220 ofd_seq_put(env, oseq);
1224 static int ofd_create_hdl(struct tgt_session_info *tsi)
1226 struct ost_body *repbody;
1227 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1228 struct obdo *rep_oa;
1229 struct obd_export *exp = tsi->tsi_exp;
1230 struct ofd_device *ofd = ofd_exp(exp);
1231 obd_seq seq = ostid_seq(&oa->o_oi);
1232 obd_id oid = ostid_id(&oa->o_oi);
1233 struct ofd_seq *oseq;
1239 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1242 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1243 if (repbody == NULL)
1246 down_read(&ofd->ofd_lastid_rwsem);
1247 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1248 * we may do that in the future.
1249 * Return -ENOSPC until the LAST_ID rebuilt. */
1250 if (unlikely(ofd->ofd_lastid_rebuilding))
1251 GOTO(out_sem, rc = -ENOSPC);
1253 rep_oa = &repbody->oa;
1254 rep_oa->o_oi = oa->o_oi;
1256 LASSERT(seq >= FID_SEQ_OST_MDT0);
1257 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1259 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1261 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1263 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1264 ofd_name(ofd), seq, PTR_ERR(oseq));
1265 GOTO(out_sem, rc = -EINVAL);
1268 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1269 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1270 if (!ofd_obd(ofd)->obd_recovering ||
1271 oid > ofd_seq_last_oid(oseq)) {
1272 CERROR("%s: recreate objid "DOSTID" > last id "LPU64
1273 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1274 ofd_seq_last_oid(oseq));
1275 GOTO(out_nolock, rc = -EINVAL);
1277 /* Do nothing here, we re-create objects during recovery
1278 * upon write replay, see ofd_preprw_write() */
1279 GOTO(out_nolock, rc = 0);
1281 /* former ofd_handle_precreate */
1282 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1283 (oa->o_flags & OBD_FL_DELORPHAN)) {
1284 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1286 /* destroy orphans */
1287 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1288 exp->exp_conn_cnt) {
1289 CERROR("%s: dropping old orphan cleanup request\n",
1291 GOTO(out_nolock, rc = 0);
1293 /* This causes inflight precreates to abort and drop lock */
1294 oseq->os_destroys_in_progress = 1;
1295 mutex_lock(&oseq->os_create_lock);
1296 if (!oseq->os_destroys_in_progress) {
1297 CERROR("%s:["LPU64"] destroys_in_progress already"
1298 " cleared\n", ofd_name(ofd), seq);
1299 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1302 diff = oid - ofd_seq_last_oid(oseq);
1303 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1304 ofd_seq_last_oid(oseq), diff);
1305 if (-diff > OST_MAX_PRECREATE) {
1306 /* FIXME: should reset precreate_next_id on MDS */
1308 } else if (diff < 0) {
1309 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1311 oseq->os_destroys_in_progress = 0;
1313 /* XXX: Used by MDS for the first time! */
1314 oseq->os_destroys_in_progress = 0;
1317 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1318 ofd->ofd_lastid_gen)) {
1319 ofd_obd_disconnect(exp);
1320 GOTO(out_nolock, rc = -ENOTCONN);
1323 mutex_lock(&oseq->os_create_lock);
1324 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1325 exp->exp_conn_cnt) {
1326 CERROR("%s: dropping old precreate request\n",
1330 /* only precreate if seq is 0, IDIF or normal and also o_id
1331 * must be specfied */
1332 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1333 !fid_seq_is_idif(seq)) || oid == 0) {
1334 diff = 1; /* shouldn't we create this right now? */
1336 diff = oid - ofd_seq_last_oid(oseq);
1337 /* Do sync create if the seq is about to used up */
1338 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1339 if (unlikely(oid >= IDIF_MAX_OID - 1))
1341 } else if (fid_seq_is_norm(seq)) {
1343 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1346 CERROR("%s : invalid o_seq "DOSTID"\n",
1347 ofd_name(ofd), POSTID(&oa->o_oi));
1348 GOTO(out, rc = -EINVAL);
1353 cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
1358 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1359 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1360 /* don't enforce grant during orphan recovery */
1361 rc = ofd_grant_create(tsi->tsi_env,
1362 ofd_obd(ofd)->obd_self_export,
1365 CDEBUG(D_HA, "%s: failed to acquire grant "
1366 "space for precreate (%d): rc = %d\n",
1367 ofd_name(ofd), diff, rc);
1372 /* This can happen if a new OST is formatted and installed
1373 * in place of an old one at the same index. Instead of
1374 * precreating potentially millions of deleted old objects
1375 * (possibly filling the OST), only precreate the last batch.
1376 * LFSCK will eventually clean up any orphans. LU-14 */
1377 if (diff > 5 * OST_MAX_PRECREATE) {
1378 diff = OST_MAX_PRECREATE / 2;
1379 LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %u "
1380 "larger than the LAST_ID "DOSTID", only "
1381 "precreating the last %u objects.\n",
1382 ofd_name(ofd), POSTID(&oa->o_oi),
1383 5 * OST_MAX_PRECREATE,
1384 POSTID(&oseq->os_oi), diff);
1385 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1389 next_id = ofd_seq_last_oid(oseq) + 1;
1390 count = ofd_precreate_batch(ofd, diff);
1392 CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1393 " at "LPU64"\n", ofd_name(ofd),
1394 count, seq, next_id);
1396 if (cfs_time_after(jiffies, enough_time)) {
1397 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1398 " created at a rate of %d/s\n",
1399 ofd_name(ofd), created,
1401 created / DISK_TIMEOUT);
1405 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1406 oseq, count, sync_trans);
1410 } else if (rc < 0) {
1415 /* some objects got created, we can return
1416 * them, even if last creation failed */
1419 CERROR("%s: unable to precreate: rc = %d\n",
1422 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1423 !(oa->o_flags & OBD_FL_DELORPHAN))
1424 ofd_grant_commit(tsi->tsi_env,
1425 ofd_obd(ofd)->obd_self_export, rc);
1427 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1430 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1433 mutex_unlock(&oseq->os_create_lock);
1436 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1438 ofd_seq_put(tsi->tsi_env, oseq);
1441 up_read(&ofd->ofd_lastid_rwsem);
1445 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1447 const struct ost_body *body = tsi->tsi_ost_body;
1448 struct ost_body *repbody;
1449 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1450 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1451 struct lu_fid *fid = &fti->fti_fid;
1458 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1461 /* This is old case for clients before Lustre 2.4 */
1462 /* If there's a DLM request, cancel the locks mentioned in it */
1463 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1465 struct ldlm_request *dlm;
1467 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1470 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0);
1473 *fid = body->oa.o_oi.oi_fid;
1474 oid = ostid_id(&body->oa.o_oi);
1477 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1479 /* check that o_misc makes sense */
1480 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1481 count = body->oa.o_misc;
1483 count = 1; /* default case - single destroy */
1485 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1486 POSTID(&body->oa.o_oi), count);
1491 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1492 if (lrc == -ENOENT) {
1494 "%s: destroying non-existent object "DFID"\n",
1495 ofd_name(ofd), PFID(fid));
1496 /* rewrite rc with -ENOENT only if it is 0 */
1499 } else if (lrc != 0) {
1500 CERROR("%s: error destroying object "DFID": %d\n",
1501 ofd_name(ofd), PFID(fid), lrc);
1507 lrc = fid_set_id(fid, oid);
1508 if (unlikely(lrc != 0 && count > 0))
1509 GOTO(out, rc = lrc);
1512 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1518 fid_to_ostid(fid, &repbody->oa.o_oi);
1522 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1524 struct obd_statfs *osfs;
1529 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1531 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1532 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
1534 CERROR("%s: statfs failed: rc = %d\n",
1535 tgt_name(tsi->tsi_tgt), rc);
1537 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1540 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1546 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1548 struct ost_body *body = tsi->tsi_ost_body;
1549 struct ost_body *repbody;
1550 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1551 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1552 struct ofd_object *fo = NULL;
1557 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1559 /* if no objid is specified, it means "sync whole filesystem" */
1560 if (!fid_is_zero(&tsi->tsi_fid)) {
1561 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1563 RETURN(PTR_ERR(fo));
1566 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1567 fo != NULL ? ofd_object_child(fo) : NULL,
1568 repbody->oa.o_size, repbody->oa.o_blocks);
1572 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1577 repbody->oa.o_oi = body->oa.o_oi;
1578 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1580 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1582 obdo_from_la(&repbody->oa, &fti->fti_attr,
1585 /* don't return rc from getattr */
1590 ofd_object_put(tsi->tsi_env, fo);
1594 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1596 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1597 struct ost_body *repbody;
1598 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1599 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1600 struct ldlm_resource *res;
1601 struct ofd_object *fo;
1602 struct filter_fid *ff = NULL;
1604 struct lustre_handle lh = { 0, };
1611 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1612 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
1614 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1615 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1616 RETURN(err_serious(-EPROTO));
1618 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1619 if (repbody == NULL)
1620 RETURN(err_serious(-ENOMEM));
1622 /* punch start,end are passed in o_size,o_blocks throught wire */
1626 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1629 /* standard truncate optimization: if file body is completely
1630 * destroyed, don't send data back to the server. */
1632 flags |= LDLM_FL_AST_DISCARD_DATA;
1634 repbody->oa.o_oi = oa->o_oi;
1635 repbody->oa.o_valid = OBD_MD_FLID;
1637 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1638 oa->o_flags & OBD_FL_SRVLOCK;
1641 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
1647 CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
1648 ", start = "LPD64", end = "LPD64"\n", PFID(&tsi->tsi_fid),
1649 oa->o_valid, start, end);
1651 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
1654 GOTO(out, rc = PTR_ERR(fo));
1656 la_from_obdo(&info->fti_attr, oa,
1657 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
1658 info->fti_attr.la_size = start;
1659 info->fti_attr.la_valid |= LA_SIZE;
1661 if (oa->o_valid & OBD_MD_FLFID) {
1662 ff = &info->fti_mds_fid;
1663 ofd_prepare_fidea(ff, oa);
1666 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
1667 ff, (struct obdo *)oa);
1671 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
1675 ofd_object_put(tsi->tsi_env, fo);
1678 tgt_extent_unlock(&lh, LCK_PW);
1680 /* we do not call this before to avoid lu_object_find() in
1681 * ->lvbo_update() holding another reference on the object.
1682 * otherwise concurrent destroy can make the object unavailable
1683 * for 2nd lu_object_find() waiting for the first reference
1684 * to go... deadlock! */
1685 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
1688 ldlm_res_lvbo_update(res, NULL, 0);
1689 ldlm_resource_putref(res);
1695 static int ofd_quotactl(struct tgt_session_info *tsi)
1697 struct obd_quotactl *oqctl, *repoqc;
1702 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
1704 RETURN(err_serious(-EPROTO));
1706 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
1708 RETURN(err_serious(-ENOMEM));
1710 /* report success for quota on/off for interoperability with current MDT
1712 if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1716 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
1718 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
1724 /* High priority request handlers for OFD */
1726 /* prolong locks for the current service time of the corresponding
1727 * portal (= OST_IO_PORTAL)
1729 static inline int prolong_timeout(struct ptlrpc_request *req)
1731 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
1734 return obd_timeout / 2;
1736 return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
1740 static int ofd_prolong_one_lock(struct tgt_session_info *tsi,
1741 struct ldlm_lock *lock,
1742 struct ldlm_extent *extent, int timeout)
1745 if (lock->l_flags & LDLM_FL_DESTROYED) /* lock already cancelled */
1748 /* XXX: never try to grab resource lock here because we're inside
1749 * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
1750 * res lock and then exp_bl_list_lock. */
1752 if (!(lock->l_flags & LDLM_FL_AST_SENT))
1753 /* ignore locks not being cancelled */
1756 LDLM_DEBUG(lock, "refreshed for req x"LPU64" ext("LPU64"->"LPU64") "
1757 "to %ds.\n", tgt_ses_req(tsi)->rq_xid, extent->start,
1758 extent->end, timeout);
1760 /* OK. this is a possible lock the user holds doing I/O
1761 * let's refresh eviction timer for it */
1762 ldlm_refresh_waiting_lock(lock, timeout);
1766 static int ofd_prolong_extent_locks(struct tgt_session_info *tsi,
1767 __u64 start, __u64 end)
1769 struct obd_export *exp = tsi->tsi_exp;
1770 struct obdo *oa = &tsi->tsi_ost_body->oa;
1771 struct ldlm_extent extent = {
1775 struct ldlm_lock *lock;
1776 int timeout = prolong_timeout(tgt_ses_req(tsi));
1781 if (oa->o_valid & OBD_MD_FLHANDLE) {
1782 /* mostly a request should be covered by only one lock, try
1784 lock = ldlm_handle2lock(&oa->o_handle);
1786 /* Fast path to check if the lock covers the whole IO
1787 * region exclusively. */
1788 if (lock->l_granted_mode == LCK_PW &&
1789 ldlm_extent_contain(&lock->l_policy_data.l_extent,
1792 LASSERT(lock->l_export == exp);
1793 lock_count = ofd_prolong_one_lock(tsi, lock,
1795 LDLM_LOCK_PUT(lock);
1798 LDLM_LOCK_PUT(lock);
1802 spin_lock_bh(&exp->exp_bl_list_lock);
1803 list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
1804 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1805 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
1807 if (!ldlm_res_eq(&tsi->tsi_resid, &lock->l_resource->lr_name))
1810 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
1814 lock_count += ofd_prolong_one_lock(tsi, lock, &extent, timeout);
1816 spin_unlock_bh(&exp->exp_bl_list_lock);
1822 * Returns 1 if the given PTLRPC matches the given LDLM lock, or 0 if it does
1825 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
1826 struct ldlm_lock *lock)
1828 struct niobuf_remote *rnb;
1829 struct obd_ioobj *ioo;
1831 struct ldlm_extent ext;
1832 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
1836 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1837 LASSERT(ioo != NULL);
1839 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1840 LASSERT(rnb != NULL);
1842 ext.start = rnb->offset;
1843 rnb += ioo->ioo_bufcnt - 1;
1844 ext.end = rnb->offset + rnb->len - 1;
1846 LASSERT(lock->l_resource != NULL);
1847 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
1851 if (opc == OST_READ)
1854 if (!(lock->l_granted_mode & mode))
1857 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
1861 * High-priority queue request check for whether the given PTLRPC request
1862 * (\a req) is blocking an LDLM lock cancel.
1864 * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1865 * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1867 * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue. This
1868 * function looks only at OST_READs and OST_WRITEs.
1870 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
1872 struct tgt_session_info *tsi;
1873 struct obd_ioobj *ioo;
1874 struct niobuf_remote *rnb;
1880 /* Don't use tgt_ses_info() to get session info, because lock_match()
1881 * can be called while request has no processing thread yet. */
1882 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1883 LASSERT(tsi != NULL);
1886 * Use LASSERT below because malformed RPCs should have
1887 * been filtered out in tgt_hpreq_handler().
1889 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1890 LASSERT(ioo != NULL);
1892 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1893 LASSERT(rnb != NULL);
1894 LASSERT(!(rnb->flags & OBD_BRW_SRVLOCK));
1896 start = rnb->offset;
1897 rnb += ioo->ioo_bufcnt - 1;
1898 end = rnb->offset + rnb->len - 1;
1900 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
1901 " ("LPU64"->"LPU64")\n",
1902 tgt_name(tsi->tsi_tgt), current->comm,
1903 PFID(&tsi->tsi_fid), start, end);
1905 lock_count = ofd_prolong_extent_locks(tsi, start, end);
1907 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1908 tgt_name(tsi->tsi_tgt), lock_count, req);
1910 RETURN(lock_count > 0);
1913 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
1915 ofd_rw_hpreq_check(req);
1919 * Like tgt_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1921 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
1922 struct ldlm_lock *lock)
1924 struct tgt_session_info *tsi;
1926 /* Don't use tgt_ses_info() to get session info, because lock_match()
1927 * can be called while request has no processing thread yet. */
1928 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1929 LASSERT(tsi != NULL);
1931 LASSERT(tsi->tsi_ost_body != NULL);
1932 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
1933 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1940 * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1942 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
1944 struct tgt_session_info *tsi;
1950 /* Don't use tgt_ses_info() to get session info, because lock_match()
1951 * can be called while request has no processing thread yet. */
1952 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1953 LASSERT(tsi != NULL);
1954 oa = &tsi->tsi_ost_body->oa;
1956 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
1957 oa->o_flags & OBD_FL_SRVLOCK));
1960 "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1961 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
1962 tsi->tsi_resid.name[1], oa->o_size, oa->o_blocks);
1964 lock_count = ofd_prolong_extent_locks(tsi, oa->o_size, oa->o_blocks);
1966 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1967 tgt_name(tsi->tsi_tgt), lock_count, req);
1969 RETURN(lock_count > 0);
1972 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
1974 ofd_punch_hpreq_check(req);
1977 struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
1978 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
1979 .hpreq_check = ofd_rw_hpreq_check,
1980 .hpreq_fini = ofd_rw_hpreq_fini
1983 struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
1984 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
1985 .hpreq_check = ofd_punch_hpreq_check,
1986 .hpreq_fini = ofd_punch_hpreq_fini
1989 /** Assign high priority operations to the IO requests */
1990 static void ofd_hp_brw(struct tgt_session_info *tsi)
1992 struct niobuf_remote *rnb;
1993 struct obd_ioobj *ioo;
1997 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1998 LASSERT(ioo != NULL); /* must exist after request preprocessing */
1999 if (ioo->ioo_bufcnt > 0) {
2000 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2001 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2003 /* no high priority if server lock is needed */
2004 if (rnb->flags & OBD_BRW_SRVLOCK)
2007 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2010 static void ofd_hp_punch(struct tgt_session_info *tsi)
2012 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2013 /* no high-priority if server lock is needed */
2014 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2015 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK)
2017 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2020 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2021 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2022 #define OST_BRW_READ OST_READ
2023 #define OST_BRW_WRITE OST_WRITE
2025 static struct tgt_handler ofd_tgt_handlers[] = {
2026 TGT_RPC_HANDLER(OST_FIRST_OPC,
2027 0, OST_CONNECT, tgt_connect,
2028 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2029 TGT_RPC_HANDLER(OST_FIRST_OPC,
2030 0, OST_DISCONNECT, tgt_disconnect,
2031 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2032 TGT_RPC_HANDLER(OST_FIRST_OPC,
2033 0, OST_SET_INFO, ofd_set_info_hdl,
2034 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2035 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2036 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
2037 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2038 OST_SETATTR, ofd_setattr_hdl),
2039 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2040 OST_CREATE, ofd_create_hdl),
2041 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2042 OST_DESTROY, ofd_destroy_hdl),
2043 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
2044 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2045 OST_BRW_READ, tgt_brw_read,
2047 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2048 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
2050 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2051 OST_PUNCH, ofd_punch_hdl,
2053 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
2054 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
2057 static struct tgt_opc_slice ofd_common_slice[] = {
2059 .tos_opc_start = OST_FIRST_OPC,
2060 .tos_opc_end = OST_LAST_OPC,
2061 .tos_hs = ofd_tgt_handlers
2064 .tos_opc_start = OBD_FIRST_OPC,
2065 .tos_opc_end = OBD_LAST_OPC,
2066 .tos_hs = tgt_obd_handlers
2069 .tos_opc_start = LDLM_FIRST_OPC,
2070 .tos_opc_end = LDLM_LAST_OPC,
2071 .tos_hs = tgt_dlm_handlers
2074 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2075 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2076 .tos_hs = tgt_out_handlers
2079 .tos_opc_start = SEQ_FIRST_OPC,
2080 .tos_opc_end = SEQ_LAST_OPC,
2081 .tos_hs = seq_handlers
2084 .tos_opc_start = LFSCK_FIRST_OPC,
2085 .tos_opc_end = LFSCK_LAST_OPC,
2086 .tos_hs = tgt_lfsck_handlers
2093 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2094 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2096 const char *dev = lustre_cfg_string(cfg, 0);
2097 struct ofd_thread_info *info = NULL;
2098 struct obd_device *obd;
2099 struct obd_statfs *osfs;
2104 obd = class_name2obd(dev);
2106 CERROR("Cannot find obd with name %s\n", dev);
2110 rc = lu_env_refill((struct lu_env *)env);
2114 obd->u.obt.obt_magic = OBT_MAGIC;
2116 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2117 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2119 spin_lock_init(&m->ofd_flags_lock);
2120 m->ofd_raid_degraded = 0;
2121 m->ofd_syncjournal = 0;
2123 m->ofd_grant_compat_disable = 0;
2124 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2127 spin_lock_init(&m->ofd_osfs_lock);
2128 m->ofd_osfs_age = cfs_time_shift_64(-1000);
2129 m->ofd_osfs_unstable = 0;
2130 m->ofd_statfs_inflight = 0;
2131 m->ofd_osfs_inflight = 0;
2134 spin_lock_init(&m->ofd_grant_lock);
2135 m->ofd_tot_dirty = 0;
2136 m->ofd_tot_granted = 0;
2137 m->ofd_tot_pending = 0;
2138 m->ofd_seq_count = 0;
2139 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2140 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2141 spin_lock_init(&m->ofd_inconsistency_lock);
2143 spin_lock_init(&m->ofd_batch_lock);
2144 rwlock_init(&obd->u.filter.fo_sptlrpc_lock);
2145 sptlrpc_rule_set_init(&obd->u.filter.fo_sptlrpc_rset);
2146 init_rwsem(&m->ofd_lastid_rwsem);
2148 obd->u.filter.fo_fl_oss_capa = 0;
2149 CFS_INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
2150 obd->u.filter.fo_capa_hash = init_capa_hash();
2151 if (obd->u.filter.fo_capa_hash == NULL)
2154 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2155 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2156 /* set this lu_device to obd, because error handling need it */
2157 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2159 rc = ofd_procfs_init(m);
2161 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2165 /* No connection accepted until configurations will finish */
2166 spin_lock(&obd->obd_dev_lock);
2167 obd->obd_no_conn = 1;
2168 spin_unlock(&obd->obd_dev_lock);
2169 obd->obd_replayable = 1;
2170 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2171 char *str = lustre_cfg_string(cfg, 4);
2173 if (strchr(str, 'n')) {
2174 CWARN("%s: recovery disabled\n", obd->obd_name);
2175 obd->obd_replayable = 0;
2179 info = ofd_info_init(env, NULL);
2183 rc = ofd_stack_init(env, m, cfg);
2185 CERROR("Can't init device stack, rc %d\n", rc);
2186 GOTO(err_fini_proc, rc);
2189 ofd_procfs_add_brw_stats_symlink(m);
2191 /* populate cached statfs data */
2192 osfs = &ofd_info(env)->fti_u.osfs;
2193 rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
2195 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
2196 GOTO(err_fini_stack, rc);
2198 if (!IS_PO2(osfs->os_bsize)) {
2199 CERROR("%s: blocksize (%d) is not a power of 2\n",
2200 obd->obd_name, osfs->os_bsize);
2201 GOTO(err_fini_stack, rc = -EPROTO);
2203 m->ofd_blockbits = fls(osfs->os_bsize) - 1;
2205 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2206 if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
2207 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2209 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2210 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2211 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2212 LDLM_NAMESPACE_SERVER,
2213 LDLM_NAMESPACE_GREEDY,
2215 if (m->ofd_namespace == NULL)
2216 GOTO(err_fini_stack, rc = -ENOMEM);
2217 /* set obd_namespace for compatibility with old code */
2218 obd->obd_namespace = m->ofd_namespace;
2219 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2220 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2221 m->ofd_namespace->ns_lvbp = m;
2223 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2224 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2226 dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
2228 /* Allow at most ddp_grant_reserved% of the available filesystem space
2229 * to be granted to clients, so that any errors in the grant overhead
2230 * calculations do not allow granting more space to clients than can be
2231 * written. Assumes that in aggregate the grant overhead calculations do
2232 * not have more than ddp_grant_reserved% estimation error in them. */
2233 m->ofd_grant_ratio =
2234 ofd_grant_ratio_conv(m->ofd_dt_conf.ddp_grant_reserved);
2236 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2237 OBD_FAIL_OST_ALL_REQUEST_NET,
2238 OBD_FAIL_OST_ALL_REPLY_NET);
2240 GOTO(err_free_ns, rc);
2242 rc = ofd_fs_setup(env, m, obd);
2244 GOTO(err_fini_lut, rc);
2246 rc = ofd_start_inconsistency_verification_thread(m);
2248 GOTO(err_fini_fs, rc);
2253 ofd_fs_cleanup(env, m);
2255 tgt_fini(env, &m->ofd_lut);
2257 ldlm_namespace_free(m->ofd_namespace, 0, obd->obd_force);
2258 obd->obd_namespace = m->ofd_namespace = NULL;
2260 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
2266 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
2268 struct obd_device *obd = ofd_obd(m);
2269 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
2270 struct lfsck_stop stop;
2272 stop.ls_status = LS_PAUSED;
2274 lfsck_stop(env, m->ofd_osd, &stop);
2275 target_recovery_fini(obd);
2276 obd_exports_barrier(obd);
2277 obd_zombie_barrier();
2279 tgt_fini(env, &m->ofd_lut);
2280 ofd_stop_inconsistency_verification_thread(m);
2281 lfsck_degister(env, m->ofd_osd);
2282 ofd_fs_cleanup(env, m);
2284 ofd_free_capa_keys(m);
2285 cleanup_capa_hash(obd->u.filter.fo_capa_hash);
2287 if (m->ofd_namespace != NULL) {
2288 ldlm_namespace_free(m->ofd_namespace, NULL,
2289 d->ld_obd->obd_force);
2290 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
2293 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
2295 LASSERT(atomic_read(&d->ld_ref) == 0);
2296 server_put_mount(obd->obd_name, true);
2300 static struct lu_device *ofd_device_fini(const struct lu_env *env,
2301 struct lu_device *d)
2304 ofd_fini(env, ofd_dev(d));
2308 static struct lu_device *ofd_device_free(const struct lu_env *env,
2309 struct lu_device *d)
2311 struct ofd_device *m = ofd_dev(d);
2313 dt_device_fini(&m->ofd_dt_dev);
2318 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
2319 struct lu_device_type *t,
2320 struct lustre_cfg *cfg)
2322 struct ofd_device *m;
2323 struct lu_device *l;
2328 return ERR_PTR(-ENOMEM);
2330 l = &m->ofd_dt_dev.dd_lu_dev;
2331 dt_device_init(&m->ofd_dt_dev, t);
2332 rc = ofd_init0(env, m, t, cfg);
2334 ofd_device_free(env, l);
2341 /* thread context key constructor/destructor */
2342 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2344 static void ofd_key_exit(const struct lu_context *ctx,
2345 struct lu_context_key *key, void *data)
2347 struct ofd_thread_info *info = data;
2349 info->fti_env = NULL;
2350 info->fti_exp = NULL;
2353 info->fti_pre_version = 0;
2356 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2359 struct lu_context_key ofd_thread_key = {
2360 .lct_tags = LCT_DT_THREAD,
2361 .lct_init = ofd_key_init,
2362 .lct_fini = ofd_key_fini,
2363 .lct_exit = ofd_key_exit
2366 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
2367 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
2369 static struct lu_device_type_operations ofd_device_type_ops = {
2370 .ldto_init = ofd_type_init,
2371 .ldto_fini = ofd_type_fini,
2373 .ldto_start = ofd_type_start,
2374 .ldto_stop = ofd_type_stop,
2376 .ldto_device_alloc = ofd_device_alloc,
2377 .ldto_device_free = ofd_device_free,
2378 .ldto_device_fini = ofd_device_fini
2381 static struct lu_device_type ofd_device_type = {
2382 .ldt_tags = LU_DEVICE_DT,
2383 .ldt_name = LUSTRE_OST_NAME,
2384 .ldt_ops = &ofd_device_type_ops,
2385 .ldt_ctx_tags = LCT_DT_THREAD
2388 int __init ofd_init(void)
2390 struct lprocfs_static_vars lvars;
2393 rc = lu_kmem_init(ofd_caches);
2397 rc = ofd_fmd_init();
2399 lu_kmem_fini(ofd_caches);
2403 lprocfs_ofd_init_vars(&lvars);
2405 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
2406 #ifndef HAVE_ONLY_PROCFS_SEQ
2409 LUSTRE_OST_NAME, &ofd_device_type);
2413 void __exit ofd_exit(void)
2416 lu_kmem_fini(ofd_caches);
2417 class_unregister_type(LUSTRE_OST_NAME);
2420 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
2421 MODULE_DESCRIPTION("Lustre Object Filtering Device");
2422 MODULE_LICENSE("GPL");
2424 module_init(ofd_init);
2425 module_exit(ofd_exit);