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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/ofd/ofd_dev.c
34 * This file contains OSD API methods for OBD Filter Device (OFD),
35 * request handlers and supplemental functions to set OFD up and clean it up.
37 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
38 * Author: Mike Pershin <mike.pershin@intel.com>
39 * Author: Johann Lombardi <johann.lombardi@intel.com>
42 * The OBD Filter Device (OFD) module belongs to the Object Storage
43 * Server stack and connects the RPC oriented Unified Target (TGT)
44 * layer (see lustre/include/lu_target.h) to the storage oriented OSD
45 * layer (see Documentation/osd-api.txt).
53 * OFD implements the LU and OBD device APIs and is responsible for:
55 * - Handling client requests (create, destroy, bulk IO, setattr,
56 * get_info, set_info, statfs) for the objects belonging to the OST
57 * (together with TGT).
59 * - Providing grant space management which allows clients to reserve
60 * disk space for data writeback. OFD tracks grants on global and
63 * - Handling object precreation requests from MDTs.
65 * - Operating the LDLM service that allows clients to maintain object
66 * data cache coherence.
69 #define DEBUG_SUBSYSTEM S_FILTER
71 #include <obd_class.h>
72 #include <obd_cksum.h>
73 #include <uapi/linux/lustre/lustre_param.h>
74 #include <lustre_fid.h>
75 #include <lustre_lfsck.h>
76 #include <lustre_dlm.h>
77 #include <lustre_quota.h>
78 #include <lustre_nodemap.h>
79 #include <lustre_log.h>
81 #include "ofd_internal.h"
83 /* Slab for OFD object allocation */
84 static struct kmem_cache *ofd_object_kmem;
85 static struct lu_kmem_descr ofd_caches[] = {
87 .ckd_cache = &ofd_object_kmem,
88 .ckd_name = "ofd_obj",
89 .ckd_size = sizeof(struct ofd_object)
97 * Connect OFD to the next device in the stack.
99 * This function is used for device stack configuration and links OFD
100 * device with bottom OSD device.
102 * \param[in] env execution environment
103 * \param[in] m OFD device
104 * \param[in] next name of next device in the stack
105 * \param[out] exp export to return
107 * \retval 0 and export in \a exp if successful
108 * \retval negative value on error
110 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
111 const char *next, struct obd_export **exp)
113 struct obd_connect_data *data = NULL;
114 struct obd_device *obd;
120 GOTO(out, rc = -ENOMEM);
122 obd = class_name2obd(next);
124 CERROR("%s: can't locate next device: %s\n",
126 GOTO(out, rc = -ENOTCONN);
129 data->ocd_connect_flags = OBD_CONNECT_VERSION;
130 data->ocd_version = LUSTRE_VERSION_CODE;
132 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
134 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
135 ofd_name(m), next, rc);
139 m->ofd_dt_dev.dd_lu_dev.ld_site =
140 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
141 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
142 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
143 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
152 * Initialize stack of devices.
154 * This function initializes OFD-OSD device stack to serve OST requests
156 * \param[in] env execution environment
157 * \param[in] m OFD device
158 * \param[in] cfg Lustre config for this server
160 * \retval 0 if successful
161 * \retval negative value on error
163 static int ofd_stack_init(const struct lu_env *env,
164 struct ofd_device *m, struct lustre_cfg *cfg,
167 const char *dev = lustre_cfg_string(cfg, 0);
169 struct ofd_thread_info *info = ofd_info(env);
170 struct lustre_mount_info *lmi;
171 struct lustre_mount_data *lmd;
177 lmi = server_get_mount(dev);
179 CERROR("Cannot get mount info for %s!\n", dev);
183 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
185 if (lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
186 m->ofd_skip_lfsck = 1;
187 if (lmd->lmd_flags & LMD_FLG_NO_PRECREATE)
188 m->ofd_no_precreate = 1;
189 *lmd_flags = lmd->lmd_flags;
192 /* find bottom osd */
193 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
197 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
198 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
199 OBD_FREE(osdname, MTI_NAME_MAXLEN);
203 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
205 m->ofd_osd = lu2dt_dev(d);
207 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
208 "%s-osd", lustre_cfg_string(cfg, 0));
214 * Finalize the device stack OFD-OSD.
216 * This function cleans OFD-OSD device stack and
217 * disconnects OFD from the OSD.
219 * \param[in] env execution environment
220 * \param[in] m OFD device
221 * \param[in] top top device of stack
223 * \retval 0 if successful
224 * \retval negative value on error
226 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
227 struct lu_device *top)
229 struct obd_device *obd = ofd_obd(m);
230 struct lustre_cfg_bufs bufs;
231 struct lustre_cfg *lcfg;
236 lu_site_purge(env, top->ld_site, ~0);
237 /* process cleanup, pass mdt obd name to get obd umount flags */
238 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
243 lustre_cfg_bufs_set_string(&bufs, 1, flags);
244 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
247 lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs);
250 top->ld_ops->ldo_process_config(env, top, lcfg);
251 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
253 if (m->ofd_los != NULL) {
254 local_oid_storage_fini(env, m->ofd_los);
258 lu_site_purge(env, top->ld_site, ~0);
259 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
260 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
261 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
264 LASSERT(m->ofd_osd_exp);
265 obd_disconnect(m->ofd_osd_exp);
270 static void ofd_stack_pre_fini(const struct lu_env *env, struct ofd_device *m,
271 struct lu_device *top)
273 struct lustre_cfg_bufs bufs;
274 struct lustre_cfg *lcfg;
279 lustre_cfg_bufs_reset(&bufs, ofd_name(m));
280 lustre_cfg_bufs_set_string(&bufs, 1, NULL);
281 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
283 CERROR("%s: failed to trigger LCFG_PRE_CLEANUP\n", ofd_name(m));
285 lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, &bufs);
286 top->ld_ops->ldo_process_config(env, top, lcfg);
287 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
288 lcfg->lcfg_buflens));
294 /* For interoperability, see mdt_interop_param[]. */
295 static struct cfg_interop_param ofd_interop_param[] = {
296 { "ost.quota_type", NULL },
301 * Check if parameters are symlinks to the OSD.
303 * Some parameters were moved from ofd to osd and only their
304 * symlinks were kept in ofd by LU-3106. They are:
305 * -writehthrough_cache_enable
306 * -readcache_max_filesize
310 * Since they are not included by the static lprocfs var list, a pre-check
311 * is added for them to avoid "unknown param" errors. If they are matched
312 * in this check, they will be passed to the OSD directly.
314 * \param[in] param parameters to check
316 * \retval true if param is symlink to OSD param
319 static bool match_symlink_param(char *param)
324 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
325 sval = strchr(param, '=');
327 paramlen = sval - param;
328 if (strncmp(param, "brw_stats", paramlen) == 0)
337 * Process various configuration parameters.
339 * This function is used by MGS to process specific configurations and
340 * pass them through to the next device in server stack, i.e. the OSD.
342 * \param[in] env execution environment
343 * \param[in] d LU device of OFD
344 * \param[in] cfg parameters to process
346 * \retval 0 if successful
347 * \retval negative value on error
349 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
350 struct lustre_cfg *cfg)
352 struct ofd_device *m = ofd_dev(d);
353 struct dt_device *dt_next = m->ofd_osd;
354 struct lu_device *next = &dt_next->dd_lu_dev;
359 switch (cfg->lcfg_command) {
361 /* For interoperability */
362 struct cfg_interop_param *ptr = NULL;
363 struct lustre_cfg *old_cfg = NULL;
367 param = lustre_cfg_string(cfg, 1);
369 CERROR("param is empty\n");
374 ptr = class_find_old_param(param, ofd_interop_param);
376 if (ptr->new_param == NULL) {
378 CWARN("For interoperability, skip this %s."
379 " It is obsolete.\n", ptr->old_param);
383 CWARN("Found old param %s, changed it to %s.\n",
384 ptr->old_param, ptr->new_param);
387 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
394 if (match_symlink_param(param)) {
395 rc = next->ld_ops->ldo_process_config(env, next, cfg);
399 count = class_modify_config(cfg, PARAM_OST,
400 &d->ld_obd->obd_kset.kobj);
405 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
407 /* we don't understand; pass it on */
408 rc = next->ld_ops->ldo_process_config(env, next, cfg);
411 case LCFG_SPTLRPC_CONF: {
416 /* others are passed further */
417 rc = next->ld_ops->ldo_process_config(env, next, cfg);
424 * Implementation of lu_object_operations::loo_object_init for OFD
426 * Allocate just the next object (OSD) in stack.
428 * \param[in] env execution environment
429 * \param[in] o lu_object of OFD object
430 * \param[in] conf additional configuration parameters, not used here
432 * \retval 0 if successful
433 * \retval negative value on error
435 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
436 const struct lu_object_conf *conf)
438 struct ofd_device *d = ofd_dev(o->lo_dev);
439 struct lu_device *under;
440 struct lu_object *below;
445 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
446 PFID(lu_object_fid(o)));
448 under = &d->ofd_osd->dd_lu_dev;
449 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
451 lu_object_add(o, below);
459 * Implementation of lu_object_operations::loo_object_free.
461 * Finish OFD object lifecycle and free its memory.
463 * \param[in] env execution environment
464 * \param[in] o LU object of OFD object
466 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
468 struct ofd_object *of = ofd_obj(o);
469 struct lu_object_header *h;
474 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
475 PFID(lu_object_fid(o)));
478 lu_object_header_fini(h);
479 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
484 * Implementation of lu_object_operations::loo_object_print.
486 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
487 * LU_OBJECT_DEBUG() for more details about the compound object printing.
489 * \param[in] env execution environment
490 * \param[in] cookie opaque data passed to the printer function
491 * \param[in] p printer function to use
492 * \param[in] o LU object of OFD object
494 * \retval 0 if successful
495 * \retval negative value on error
497 static int ofd_object_print(const struct lu_env *env, void *cookie,
498 lu_printer_t p, const struct lu_object *o)
500 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
503 static struct lu_object_operations ofd_obj_ops = {
504 .loo_object_init = ofd_object_init,
505 .loo_object_free = ofd_object_free,
506 .loo_object_print = ofd_object_print
510 * Implementation of lu_device_operations::lod_object_alloc.
512 * This function allocates OFD part of compound OFD-OSD object and
513 * initializes its header, because OFD is the top device in stack
515 * \param[in] env execution environment
516 * \param[in] hdr object header, NULL for OFD
517 * \param[in] d lu_device
519 * \retval allocated object if successful
520 * \retval NULL value on failed allocation
522 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
523 const struct lu_object_header *hdr,
526 struct ofd_object *of;
530 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
533 struct lu_object_header *h;
535 o = &of->ofo_obj.do_lu;
537 lu_object_header_init(h);
538 lu_object_init(o, h, d);
539 lu_object_add_top(h, o);
540 o->lo_ops = &ofd_obj_ops;
548 * Return the result of LFSCK run to the OFD.
550 * Notify OFD about result of LFSCK run. That may block the new object
551 * creation until problem is fixed by LFSCK.
553 * \param[in] env execution environment
554 * \param[in] data pointer to the OFD device
555 * \param[in] event LFSCK event type
557 * \retval 0 if successful
558 * \retval negative value on unknown event
560 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
561 enum lfsck_events event)
563 struct ofd_device *ofd = data;
564 struct obd_device *obd = ofd_obd(ofd);
567 case LE_LASTID_REBUILDING:
568 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
569 "on the device until the LAST_ID rebuilt successfully.\n",
571 down_write(&ofd->ofd_lastid_rwsem);
572 ofd->ofd_lastid_rebuilding = 1;
573 up_write(&ofd->ofd_lastid_rwsem);
575 case LE_LASTID_REBUILT: {
576 down_write(&ofd->ofd_lastid_rwsem);
577 ofd_seqs_free(env, ofd);
578 ofd->ofd_lastid_rebuilding = 0;
579 ofd->ofd_lastid_gen++;
580 up_write(&ofd->ofd_lastid_rwsem);
581 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
586 CERROR("%s: unknown lfsck event: rc = %d\n",
587 ofd_name(ofd), event);
595 * Implementation of lu_device_operations::ldo_prepare.
597 * This method is called after layer has been initialized and before it starts
598 * serving user requests. In OFD it starts lfsk check routines and initializes
601 * \param[in] env execution environment
602 * \param[in] pdev higher device in stack, NULL for OFD
603 * \param[in] dev lu_device of OFD device
605 * \retval 0 if successful
606 * \retval negative value on error
608 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
609 struct lu_device *dev)
611 struct ofd_thread_info *info;
612 struct ofd_device *ofd = ofd_dev(dev);
613 struct obd_device *obd = ofd_obd(ofd);
614 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
619 info = ofd_info_init(env, NULL);
623 /* initialize lower device */
624 rc = next->ld_ops->ldo_prepare(env, dev, next);
628 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
629 ofd_lfsck_out_notify, ofd, false);
631 CERROR("%s: failed to initialize lfsck: rc = %d\n",
636 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
637 /* The LFSCK instance is registered just now, so it must be there when
638 * register the namespace to such instance. */
639 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
641 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
642 OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_OST_PREPARE_DELAY, OBD_FAIL_ONCE,
643 (OBD_TIMEOUT_DEFAULT + 1) / 4);
644 LASSERT(obd->obd_no_conn);
645 spin_lock(&obd->obd_dev_lock);
646 obd->obd_no_conn = 0;
647 spin_unlock(&obd->obd_dev_lock);
649 if (obd->obd_recovering == 0)
650 ofd_postrecov(env, ofd);
656 * Implementation of lu_device_operations::ldo_recovery_complete.
658 * This method notifies all layers about 'recovery complete' event. That means
659 * device is in full state and consistent. An OFD calculates available grant
660 * space upon this event.
662 * \param[in] env execution environment
663 * \param[in] dev lu_device of OFD device
665 * \retval 0 if successful
666 * \retval negative value on error
668 static int ofd_recovery_complete(const struct lu_env *env,
669 struct lu_device *dev)
671 struct ofd_thread_info *oti = ofd_info(env);
672 struct ofd_device *ofd = ofd_dev(dev);
673 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
679 * Grant space for object precreation on the self export.
680 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
681 * is enough to create 10k objects. More space is then acquired for
682 * precreation in tgt_grant_create().
684 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
685 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
686 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
687 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
688 OBD_CONNECT_GRANT_PARAM;
689 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
691 rc = next->ld_ops->ldo_recovery_complete(env, next);
696 * lu_device_operations matrix for OFD device.
698 static struct lu_device_operations ofd_lu_ops = {
699 .ldo_object_alloc = ofd_object_alloc,
700 .ldo_process_config = ofd_process_config,
701 .ldo_recovery_complete = ofd_recovery_complete,
702 .ldo_prepare = ofd_prepare,
705 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
707 * Expose OSD statistics to OFD layer.
709 * The osd interfaces to the backend file system exposes useful data
710 * such as brw_stats and read or write cache states. This same data
711 * needs to be exposed into the obdfilter (ofd) layer to maintain
712 * backwards compatibility. This function creates the symlinks in the
713 * proc layer to enable this.
715 * \param[in] ofd OFD device
717 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
719 struct obd_device *obd = ofd_obd(ofd);
720 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
721 struct kobj_type *osd_type;
724 osd_type = get_ktype(&ofd->ofd_osd->dd_kobj);
725 for (i = 0; osd_type->default_attrs[i]; i++) {
726 if (strcmp(osd_type->default_attrs[i]->name,
727 "read_cache_enable") == 0) {
728 ofd->ofd_read_cache_enable =
729 osd_type->default_attrs[i];
732 if (strcmp(osd_type->default_attrs[i]->name,
733 "readcache_max_filesize") == 0) {
734 ofd->ofd_read_cache_max_filesize =
735 osd_type->default_attrs[i];
738 if (strcmp(osd_type->default_attrs[i]->name,
739 "writethrough_cache_enable") == 0) {
740 ofd->ofd_write_cache_enable =
741 osd_type->default_attrs[i];
745 if (obd->obd_proc_entry == NULL)
748 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
749 "../../%s/%s/brw_stats",
750 osd_obd->obd_type->typ_name, obd->obd_name);
755 * Cleanup all procfs entries in OFD.
757 * \param[in] ofd OFD device
759 static void ofd_procfs_fini(struct ofd_device *ofd)
761 struct obd_device *obd = ofd_obd(ofd);
763 tgt_tunables_fini(&ofd->ofd_lut);
764 lprocfs_free_per_client_stats(obd);
765 lprocfs_obd_cleanup(obd);
766 lprocfs_free_obd_stats(obd);
767 lprocfs_job_stats_fini(obd);
771 * Stop SEQ/FID server on OFD.
773 * \param[in] env execution environment
774 * \param[in] ofd OFD device
776 * \retval 0 if successful
777 * \retval negative value on error
779 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
781 return seq_site_fini(env, &ofd->ofd_seq_site);
785 * Start SEQ/FID server on OFD.
787 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
788 * It also connects to the master server to get own FID sequence (SEQ) range
789 * to this particular OFD. Typically that happens when the OST is first
790 * formatted or in the rare case that it exhausts the local sequence range.
792 * The sequence range is allocated out to the MDTs for OST object allocations,
793 * and not directly to the clients.
795 * \param[in] env execution environment
796 * \param[in] ofd OFD device
798 * \retval 0 if successful
799 * \retval negative value on error
801 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
803 struct seq_server_site *ss = &ofd->ofd_seq_site;
804 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
805 char *obd_name = ofd_name(ofd);
807 int len = strlen(obd_name) + 7;
810 ss = &ofd->ofd_seq_site;
811 lu->ld_site->ld_seq_site = ss;
812 ss->ss_lu = lu->ld_site;
813 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
815 OBD_ALLOC(name, len);
819 OBD_ALLOC_PTR(ss->ss_server_seq);
820 if (ss->ss_server_seq == NULL)
821 GOTO(out_name, rc = -ENOMEM);
823 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
824 LUSTRE_SEQ_SERVER, ss);
826 CERROR("%s: seq server init error: rc = %d\n", obd_name, rc);
827 GOTO(out_server, rc);
829 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
831 OBD_ALLOC_PTR(ss->ss_client_seq);
832 if (ss->ss_client_seq == NULL)
833 GOTO(out_server, rc = -ENOMEM);
835 snprintf(name, len, "%s-super", obd_name);
836 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
839 CERROR("%s: seq client init error: rc = %d\n", obd_name, rc);
840 GOTO(out_client, rc);
843 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
847 seq_client_fini(ss->ss_client_seq);
848 OBD_FREE_PTR(ss->ss_client_seq);
849 ss->ss_client_seq = NULL;
851 seq_server_fini(ss->ss_server_seq, env);
852 OBD_FREE_PTR(ss->ss_server_seq);
853 ss->ss_server_seq = NULL;
862 * OFD request handler for OST_SET_INFO RPC.
864 * This is OFD-specific part of request handling
866 * \param[in] tsi target session environment for this request
868 * \retval 0 if successful
869 * \retval negative value on error
871 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
873 struct ptlrpc_request *req = tgt_ses_req(tsi);
874 struct ost_body *body = NULL, *repbody;
875 void *key, *val = NULL;
876 int keylen, vallen, rc = 0;
877 bool is_grant_shrink;
881 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
883 DEBUG_REQ(D_HA, req, "no set_info key");
884 RETURN(err_serious(-EFAULT));
886 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
889 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
891 DEBUG_REQ(D_HA, req, "no set_info val");
892 RETURN(err_serious(-EFAULT));
894 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
897 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
899 /* In this case the value is actually an RMF_OST_BODY, so we
900 * transmutate the type of this PTLRPC */
901 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
903 rc = req_capsule_server_pack(tsi->tsi_pill);
907 if (is_grant_shrink) {
908 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
910 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
913 /** handle grant shrink, similar to a read request */
914 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
916 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
918 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
921 CERROR("%s: Unsupported key %s\n",
922 tgt_name(tsi->tsi_tgt), (char *)key);
925 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
932 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
934 * This function returns a list of extents which describes how a file's
935 * blocks are laid out on the disk.
937 * \param[in] env execution environment
938 * \param[in] ofd OFD device
939 * \param[in] fid FID of object
940 * \param[in] fiemap fiemap structure to fill with data
942 * \retval 0 if \a fiemap is filled with data successfully
943 * \retval negative value on error
945 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
946 struct lu_fid *fid, struct fiemap *fiemap)
948 struct ofd_object *fo;
951 fo = ofd_object_find(env, ofd, fid);
953 CERROR("%s: error finding object "DFID"\n",
954 ofd_name(ofd), PFID(fid));
958 ofd_read_lock(env, fo);
959 if (ofd_object_exists(fo))
960 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
963 ofd_read_unlock(env, fo);
964 ofd_object_put(env, fo);
969 static int ofd_lock_unlock_region(const struct lu_env *env,
970 struct ldlm_namespace *ns,
971 struct ldlm_res_id *res_id,
972 unsigned long long begin,
973 unsigned long long end)
977 struct lustre_handle lh = { 0 };
979 LASSERT(begin <= end);
981 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
985 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
986 tgt_extent_unlock(&lh, LCK_PR);
992 * Lock the sparse areas of given resource.
994 * The locking of sparse areas will cause dirty data to be flushed back from
995 * clients. This is used when getting the FIEMAP of an object to make sure
996 * there is no unaccounted cached data on clients.
998 * This function goes through \a fiemap list of extents and locks only sparse
999 * areas between extents.
1001 * \param[in] ns LDLM namespace
1002 * \param[in] res_id resource ID
1003 * \param[in] fiemap file extents mapping on disk
1004 * \param[in] locked list head of regions list
1006 * \retval 0 if successful
1007 * \retval negative value on error
1009 static int lock_zero_regions(const struct lu_env *env,
1010 struct ldlm_namespace *ns,
1011 struct ldlm_res_id *res_id,
1012 struct fiemap *fiemap)
1014 __u64 begin = fiemap->fm_start;
1017 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1021 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1022 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1023 if (fiemap_start[i].fe_logical > begin) {
1024 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1025 begin, fiemap_start[i].fe_logical);
1026 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1027 fiemap_start[i].fe_logical);
1032 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1035 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1036 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1037 begin, fiemap->fm_start + fiemap->fm_length);
1038 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1039 fiemap->fm_start + fiemap->fm_length);
1047 * OFD request handler for OST_GET_INFO RPC.
1049 * This is OFD-specific part of request handling. The OFD-specific keys are:
1050 * - KEY_LAST_ID (obsolete)
1054 * This function reads needed data from storage and fills reply with it.
1056 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1057 * and is kept for compatibility.
1059 * \param[in] tsi target session environment for this request
1061 * \retval 0 if successful
1062 * \retval negative value on error
1064 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1066 struct obd_export *exp = tsi->tsi_exp;
1067 struct ofd_device *ofd = ofd_exp(exp);
1068 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1071 int replylen, rc = 0;
1075 /* this common part for get_info rpc */
1076 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1078 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1079 RETURN(err_serious(-EPROTO));
1081 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1084 if (KEY_IS(KEY_LAST_ID)) {
1086 struct ofd_seq *oseq;
1088 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1089 rc = req_capsule_server_pack(tsi->tsi_pill);
1091 RETURN(err_serious(rc));
1093 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1095 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1096 (u64)exp->exp_filter_data.fed_group);
1100 *last_id = ofd_seq_last_oid(oseq);
1101 ofd_seq_put(tsi->tsi_env, oseq);
1102 } else if (KEY_IS(KEY_FIEMAP)) {
1103 struct ll_fiemap_info_key *fm_key;
1104 struct fiemap *fiemap;
1107 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1109 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1110 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1112 RETURN(err_serious(rc));
1114 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1116 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1118 replylen = fiemap_count_to_size(
1119 fm_key->lfik_fiemap.fm_extent_count);
1120 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1121 RCL_SERVER, replylen);
1123 rc = req_capsule_server_pack(tsi->tsi_pill);
1125 RETURN(err_serious(rc));
1127 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1131 *fiemap = fm_key->lfik_fiemap;
1132 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1134 /* LU-3219: Lock the sparse areas to make sure dirty
1135 * flushed back from client, then call fiemap again. */
1136 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1137 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1138 ost_fid_build_resid(fid, &fti->fti_resid);
1139 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1140 &fti->fti_resid, fiemap);
1142 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1145 } else if (KEY_IS(KEY_LAST_FID)) {
1146 struct ofd_device *ofd = ofd_exp(exp);
1147 struct ofd_seq *oseq;
1151 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1152 rc = req_capsule_server_pack(tsi->tsi_pill);
1154 RETURN(err_serious(rc));
1156 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1158 RETURN(err_serious(-EPROTO));
1160 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1162 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1166 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1167 ostid_seq(&fti->fti_ostid));
1169 RETURN(PTR_ERR(oseq));
1171 rc = ostid_to_fid(fid, &oseq->os_oi,
1172 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1176 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1179 ofd_seq_put(tsi->tsi_env, oseq);
1181 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1185 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1192 * OFD request handler for OST_GETATTR RPC.
1194 * This is OFD-specific part of request handling. It finds the OFD object
1195 * by its FID, gets attributes from storage and packs result to the reply.
1197 * \param[in] tsi target session environment for this request
1199 * \retval 0 if successful
1200 * \retval negative value on error
1202 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1204 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1205 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1206 struct ost_body *repbody;
1207 struct lustre_handle lh = { 0 };
1208 struct ofd_object *fo;
1210 enum ldlm_mode lock_mode = LCK_PR;
1215 LASSERT(tsi->tsi_ost_body != NULL);
1217 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1218 if (repbody == NULL)
1221 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1222 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1224 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1225 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1228 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1231 rc = tgt_extent_lock(tsi->tsi_env,
1232 tsi->tsi_tgt->lut_obd->obd_namespace,
1233 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1239 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1241 GOTO(out, rc = PTR_ERR(fo));
1243 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1247 obdo_from_la(&repbody->oa, &fti->fti_attr,
1248 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1250 /* Store object version in reply */
1251 curr_version = dt_version_get(tsi->tsi_env,
1252 ofd_object_child(fo));
1253 if ((__s64)curr_version != -EOPNOTSUPP) {
1254 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1255 repbody->oa.o_data_version = curr_version;
1258 if (fo->ofo_ff.ff_layout_version > 0) {
1259 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1260 repbody->oa.o_layout_version =
1261 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1263 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1264 PFID(&tsi->tsi_fid),
1265 repbody->oa.o_layout_version);
1269 ofd_object_put(tsi->tsi_env, fo);
1272 tgt_extent_unlock(&lh, lock_mode);
1274 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1277 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1278 repbody->oa.o_flags = OBD_FL_FLUSH;
1284 * OFD request handler for OST_SETATTR RPC.
1286 * This is OFD-specific part of request handling. It finds the OFD object
1287 * by its FID, sets attributes from request and packs result to the reply.
1289 * \param[in] tsi target session environment for this request
1291 * \retval 0 if successful
1292 * \retval negative value on error
1294 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1296 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1297 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1298 struct ost_body *body = tsi->tsi_ost_body;
1299 struct ost_body *repbody;
1300 struct ldlm_resource *res;
1301 struct ofd_object *fo;
1306 LASSERT(body != NULL);
1308 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1309 if (repbody == NULL)
1312 repbody->oa.o_oi = body->oa.o_oi;
1313 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1315 /* This would be very bad - accidentally truncating a file when
1316 * changing the time or similar - bug 12203. */
1317 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1318 body->oa.o_size != OBD_OBJECT_EOF) {
1319 static char mdsinum[48];
1321 if (body->oa.o_valid & OBD_MD_FLFID)
1322 snprintf(mdsinum, sizeof(mdsinum) - 1,
1323 "of parent "DFID, body->oa.o_parent_seq,
1324 body->oa.o_parent_oid, 0);
1328 CERROR("%s: setattr from %s is trying to truncate object "DFID
1329 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1330 PFID(&tsi->tsi_fid), mdsinum);
1334 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1336 GOTO(out, rc = PTR_ERR(fo));
1338 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1339 fti->fti_attr.la_valid &= ~LA_TYPE;
1341 /* setting objects attributes (including owner/group) */
1342 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1346 obdo_from_la(&repbody->oa, &fti->fti_attr,
1347 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1349 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1353 ofd_object_put(tsi->tsi_env, fo);
1356 /* we do not call this before to avoid lu_object_find() in
1357 * ->lvbo_update() holding another reference on the object.
1358 * otherwise concurrent destroy can make the object unavailable
1359 * for 2nd lu_object_find() waiting for the first reference
1360 * to go... deadlock! */
1361 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1362 &tsi->tsi_resid, LDLM_EXTENT, 0);
1364 ldlm_res_lvbo_update(res, NULL, 0);
1365 ldlm_resource_putref(res);
1372 * Destroy OST orphans.
1374 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1375 * set then we must destroy possible orphaned objects.
1377 * \param[in] env execution environment
1378 * \param[in] exp OBD export
1379 * \param[in] ofd OFD device
1380 * \param[in] oa obdo structure for reply
1382 * \retval 0 if successful
1383 * \retval negative value on error
1385 static int ofd_orphans_destroy(const struct lu_env *env,
1386 struct obd_export *exp,
1387 struct ofd_device *ofd, struct obdo *oa)
1389 struct ofd_thread_info *info = ofd_info(env);
1390 struct lu_fid *fid = &info->fti_fid;
1391 struct ost_id *oi = &oa->o_oi;
1392 struct ofd_seq *oseq;
1393 u64 seq = ostid_seq(oi);
1394 u64 end_id = ostid_id(oi);
1402 oseq = ofd_seq_get(ofd, seq);
1404 CERROR("%s: Can not find seq for "DOSTID"\n",
1405 ofd_name(ofd), POSTID(oi));
1410 last = ofd_seq_last_oid(oseq);
1413 LASSERT(exp != NULL);
1414 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1416 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1419 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1420 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1422 while (oid > end_id) {
1423 rc = fid_set_id(fid, oid);
1424 if (unlikely(rc != 0))
1427 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1428 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1429 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1430 /* this is pretty fatal... */
1431 CEMERG("%s: error destroying precreated id "
1433 ofd_name(ofd), PFID(fid), rc);
1437 ofd_seq_last_oid_set(oseq, oid);
1438 /* update last_id on disk periodically so that if we
1439 * restart * we don't need to re-scan all of the just
1440 * deleted objects. */
1441 if ((oid & 511) == 0)
1442 ofd_seq_last_oid_write(env, ofd, oseq);
1446 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1447 ofd_name(ofd), seq, oid);
1451 ofd_seq_last_oid_set(oseq, oid);
1452 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1454 /* don't reuse orphan object, return last used objid */
1455 rc = ostid_set_id(oi, last);
1461 ofd_seq_put(env, oseq);
1466 * OFD request handler for OST_CREATE RPC.
1468 * This is OFD-specific part of request handling. Its main purpose is to
1469 * create new data objects on OST, but it also used to destroy orphans.
1471 * \param[in] tsi target session environment for this request
1473 * \retval 0 if successful
1474 * \retval negative value on error
1476 static int ofd_create_hdl(struct tgt_session_info *tsi)
1478 struct ptlrpc_request *req = tgt_ses_req(tsi);
1479 struct ost_body *repbody;
1480 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1481 struct obdo *rep_oa;
1482 struct obd_export *exp = tsi->tsi_exp;
1483 struct ofd_device *ofd = ofd_exp(exp);
1484 u64 seq = ostid_seq(&oa->o_oi);
1485 u64 oid = ostid_id(&oa->o_oi);
1486 struct ofd_seq *oseq;
1494 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1497 if (ofd->ofd_no_precreate)
1500 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1501 if (repbody == NULL)
1504 down_read(&ofd->ofd_lastid_rwsem);
1505 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1506 * we may do that in the future.
1507 * Return -ENOSPC until the LAST_ID rebuilt. */
1508 if (unlikely(ofd->ofd_lastid_rebuilding))
1509 GOTO(out_sem, rc = -ENOSPC);
1511 rep_oa = &repbody->oa;
1512 rep_oa->o_oi = oa->o_oi;
1514 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1516 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1518 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1520 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1521 ofd_name(ofd), seq, PTR_ERR(oseq));
1522 GOTO(out_sem, rc = -EINVAL);
1525 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1526 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1527 if (!ofd_obd(ofd)->obd_recovering ||
1528 oid > ofd_seq_last_oid(oseq)) {
1529 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1530 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1531 ofd_seq_last_oid(oseq));
1532 GOTO(out_nolock, rc = -EINVAL);
1534 /* Do nothing here, we re-create objects during recovery
1535 * upon write replay, see ofd_preprw_write() */
1536 GOTO(out_nolock, rc = 0);
1538 /* former ofd_handle_precreate */
1539 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1540 (oa->o_flags & OBD_FL_DELORPHAN)) {
1541 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1543 /* destroy orphans */
1544 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1545 exp->exp_conn_cnt) {
1546 CERROR("%s: dropping old orphan cleanup request\n",
1548 GOTO(out_nolock, rc = 0);
1550 /* This causes inflight precreates to abort and drop lock */
1551 oseq->os_destroys_in_progress = 1;
1552 mutex_lock(&oseq->os_create_lock);
1553 if (!oseq->os_destroys_in_progress) {
1555 "%s:[%llu] destroys_in_progress already cleared\n",
1556 ofd_name(ofd), seq);
1557 rc = ostid_set_id(&rep_oa->o_oi,
1558 ofd_seq_last_oid(oseq));
1561 diff = oid - ofd_seq_last_oid(oseq);
1562 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1563 ofd_seq_last_oid(oseq), diff);
1564 if (-diff > OST_MAX_PRECREATE) {
1565 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1566 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1567 "(%llu), trust the OST\n",
1568 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1569 PFID(&oseq->os_oi.oi_fid),
1570 ofd_seq_last_oid(oseq));
1572 /* Let MDS know that we are so far ahead. */
1573 rc = ostid_set_id(&rep_oa->o_oi,
1574 ofd_seq_last_oid(oseq) + 1);
1575 } else if (diff < 0) {
1576 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1578 oseq->os_destroys_in_progress = 0;
1580 /* XXX: Used by MDS for the first time! */
1581 oseq->os_destroys_in_progress = 0;
1584 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1585 ofd->ofd_lastid_gen)) {
1586 /* Keep the export ref so we can send the reply. */
1587 ofd_obd_disconnect(class_export_get(exp));
1588 GOTO(out_nolock, rc = -ENOTCONN);
1591 mutex_lock(&oseq->os_create_lock);
1592 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1593 exp->exp_conn_cnt) {
1594 CERROR("%s: dropping old precreate request\n",
1598 /* only precreate if seq is 0, IDIF or normal and also o_id
1599 * must be specfied */
1600 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1601 !fid_seq_is_idif(seq)) || oid == 0) {
1602 diff = 1; /* shouldn't we create this right now? */
1604 diff = oid - ofd_seq_last_oid(oseq);
1605 /* Do sync create if the seq is about to used up */
1606 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1607 if (unlikely(oid >= IDIF_MAX_OID - 1))
1609 } else if (fid_seq_is_norm(seq)) {
1611 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1614 CERROR("%s : invalid o_seq "DOSTID"\n",
1615 ofd_name(ofd), POSTID(&oa->o_oi));
1616 GOTO(out, rc = -EINVAL);
1621 CERROR("%s: invalid precreate request for "
1622 DOSTID", last_id %llu. "
1623 "Likely MDS last_id corruption\n",
1624 ofd_name(ofd), POSTID(&oa->o_oi),
1625 ofd_seq_last_oid(oseq));
1626 GOTO(out, rc = -EINVAL);
1631 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1637 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1638 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1639 /* don't enforce grant during orphan recovery */
1640 granted = tgt_grant_create(tsi->tsi_env,
1641 ofd_obd(ofd)->obd_self_export,
1646 CDEBUG(D_HA, "%s: failed to acquire grant "
1647 "space for precreate (%lld): rc = %d\n",
1648 ofd_name(ofd), diff, rc);
1653 /* This can happen if a new OST is formatted and installed
1654 * in place of an old one at the same index. Instead of
1655 * precreating potentially millions of deleted old objects
1656 * (possibly filling the OST), only precreate the last batch.
1657 * LFSCK will eventually clean up any orphans. LU-14 */
1658 if (diff > 5 * OST_MAX_PRECREATE) {
1659 diff = OST_MAX_PRECREATE / 2;
1660 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1661 "OST replaced or reformatted: "
1662 "LFSCK will clean up",
1665 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1666 "%u larger than the LAST_ID "DOSTID", only "
1667 "precreating the last %lld objects.\n",
1668 ofd_name(ofd), POSTID(&oa->o_oi),
1669 5 * OST_MAX_PRECREATE,
1670 POSTID(&oseq->os_oi), diff);
1671 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1675 next_id = ofd_seq_last_oid(oseq) + 1;
1676 count = ofd_precreate_batch(ofd, (int)diff);
1678 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1679 " at %llu\n", ofd_name(ofd),
1680 count, seq, next_id);
1682 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1683 && ktime_get_seconds() > enough_time) {
1684 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1685 " created at a rate of %d/s\n",
1686 ofd_name(ofd), created, diff + created,
1687 created / DISK_TIMEOUT);
1691 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1692 oseq, count, sync_trans);
1696 } else if (rc < 0) {
1702 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1703 LCONSOLE_WARN("%s: can't create the same count of"
1704 " objects when replaying the request"
1705 " (diff is %lld). see LU-4621\n",
1706 ofd_name(ofd), diff);
1709 /* some objects got created, we can return
1710 * them, even if last creation failed */
1713 CERROR("%s: unable to precreate: rc = %d\n",
1716 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1717 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1718 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1723 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1727 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1729 if (unlikely(!oseq->os_last_id_synced))
1730 oseq->os_last_id_synced = 1;
1732 mutex_unlock(&oseq->os_create_lock);
1735 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1736 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1737 struct lu_fid *fid = &info->fti_fid;
1739 /* For compatible purpose, it needs to convert back to
1740 * OST ID before put it on wire. */
1741 *fid = rep_oa->o_oi.oi_fid;
1742 fid_to_ostid(fid, &rep_oa->o_oi);
1744 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1746 ofd_seq_put(tsi->tsi_env, oseq);
1749 up_read(&ofd->ofd_lastid_rwsem);
1754 * OFD request handler for OST_DESTROY RPC.
1756 * This is OFD-specific part of request handling. It destroys data objects
1757 * related to destroyed object on MDT.
1759 * \param[in] tsi target session environment for this request
1761 * \retval 0 if successful
1762 * \retval negative value on error
1764 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1766 const struct ost_body *body = tsi->tsi_ost_body;
1767 struct ost_body *repbody;
1768 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1769 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1770 struct lu_fid *fid = &fti->fti_fid;
1777 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1780 /* This is old case for clients before Lustre 2.4 */
1781 /* If there's a DLM request, cancel the locks mentioned in it */
1782 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1784 struct ldlm_request *dlm;
1786 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1789 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1792 *fid = body->oa.o_oi.oi_fid;
1793 oid = ostid_id(&body->oa.o_oi);
1796 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1798 /* check that o_misc makes sense */
1799 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1800 count = body->oa.o_misc;
1802 count = 1; /* default case - single destroy */
1804 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1805 POSTID(&body->oa.o_oi), count);
1810 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1811 if (lrc == -ENOENT) {
1813 "%s: destroying non-existent object "DFID"\n",
1814 ofd_name(ofd), PFID(fid));
1815 /* rewrite rc with -ENOENT only if it is 0 */
1818 } else if (lrc != 0) {
1819 CERROR("%s: error destroying object "DFID": %d\n",
1820 ofd_name(ofd), PFID(fid), lrc);
1826 lrc = fid_set_id(fid, oid);
1827 if (unlikely(lrc != 0 && count > 0))
1828 GOTO(out, rc = lrc);
1831 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1837 fid_to_ostid(fid, &repbody->oa.o_oi);
1842 * OFD request handler for OST_STATFS RPC.
1844 * This function gets statfs data from storage as part of request
1847 * \param[in] tsi target session environment for this request
1849 * \retval 0 if successful
1850 * \retval negative value on error
1852 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1854 struct obd_statfs *osfs;
1859 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1861 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1863 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1864 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1866 CERROR("%s: statfs failed: rc = %d\n",
1867 tgt_name(tsi->tsi_tgt), rc);
1869 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1872 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1879 * OFD request handler for OST_SYNC RPC.
1881 * Sync object data or all filesystem data to the disk and pack the
1884 * \param[in] tsi target session environment for this request
1886 * \retval 0 if successful
1887 * \retval negative value on error
1889 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1891 struct ost_body *body = tsi->tsi_ost_body;
1892 struct ost_body *repbody;
1893 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1894 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1895 struct ofd_object *fo = NULL;
1900 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1902 /* if no objid is specified, it means "sync whole filesystem" */
1903 if (!fid_is_zero(&tsi->tsi_fid)) {
1904 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1906 RETURN(PTR_ERR(fo));
1909 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1910 fo != NULL ? ofd_object_child(fo) : NULL,
1911 repbody->oa.o_size, repbody->oa.o_blocks);
1915 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1920 repbody->oa.o_oi = body->oa.o_oi;
1921 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1923 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1925 obdo_from_la(&repbody->oa, &fti->fti_attr,
1928 /* don't return rc from getattr */
1933 ofd_object_put(tsi->tsi_env, fo);
1938 * OFD request handler for OST_PUNCH RPC.
1940 * This is part of request processing. Validate request fields,
1941 * punch (truncate) the given OFD object and pack reply.
1943 * \param[in] tsi target session environment for this request
1945 * \retval 0 if successful
1946 * \retval negative value on error
1948 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1950 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1951 struct ost_body *repbody;
1952 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1953 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1954 struct ldlm_resource *res;
1955 struct ofd_object *fo;
1957 struct lustre_handle lh = { 0, };
1964 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
1966 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1967 BUILD_BUG_ON(!(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK));
1969 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1970 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1971 RETURN(err_serious(-EPROTO));
1973 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1974 if (repbody == NULL)
1975 RETURN(err_serious(-ENOMEM));
1977 /* punch start,end are passed in o_size,o_blocks throught wire */
1981 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1984 /* standard truncate optimization: if file body is completely
1985 * destroyed, don't send data back to the server. */
1987 flags |= LDLM_FL_AST_DISCARD_DATA;
1989 repbody->oa.o_oi = oa->o_oi;
1990 repbody->oa.o_valid = OBD_MD_FLID;
1992 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1993 oa->o_flags & OBD_FL_SRVLOCK;
1996 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
1997 end, &lh, LCK_PW, &flags);
2002 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2003 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2004 oa->o_valid, start, end);
2006 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2009 GOTO(out, rc = PTR_ERR(fo));
2011 la_from_obdo(&info->fti_attr, oa,
2012 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2013 info->fti_attr.la_size = start;
2014 info->fti_attr.la_valid |= LA_SIZE;
2016 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2021 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2025 ofd_object_put(tsi->tsi_env, fo);
2028 tgt_extent_unlock(&lh, LCK_PW);
2030 /* we do not call this before to avoid lu_object_find() in
2031 * ->lvbo_update() holding another reference on the object.
2032 * otherwise concurrent destroy can make the object unavailable
2033 * for 2nd lu_object_find() waiting for the first reference
2034 * to go... deadlock! */
2035 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2038 struct ost_lvb *res_lvb;
2040 ldlm_res_lvbo_update(res, NULL, 0);
2041 res_lvb = res->lr_lvb_data;
2042 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2043 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2044 ldlm_resource_putref(res);
2050 static int ofd_ladvise_prefetch(const struct lu_env *env,
2051 struct ofd_object *fo,
2052 struct niobuf_local *lnb,
2053 __u64 start, __u64 end, enum dt_bufs_type dbt)
2055 struct ofd_thread_info *info = ofd_info(env);
2056 pgoff_t start_index, end_index, pages;
2057 struct niobuf_remote rnb;
2058 unsigned long nr_local;
2064 ofd_read_lock(env, fo);
2065 if (!ofd_object_exists(fo))
2066 GOTO(out_unlock, rc = -ENOENT);
2068 rc = ofd_attr_get(env, fo, &info->fti_attr);
2070 GOTO(out_unlock, rc);
2072 if (end > info->fti_attr.la_size)
2073 end = info->fti_attr.la_size;
2076 GOTO(out_unlock, rc);
2078 /* We need page aligned offset and length */
2079 start_index = start >> PAGE_SHIFT;
2080 end_index = (end - 1) >> PAGE_SHIFT;
2081 pages = end_index - start_index + 1;
2083 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2084 PTLRPC_MAX_BRW_PAGES;
2085 rnb.rnb_offset = start_index << PAGE_SHIFT;
2086 rnb.rnb_len = nr_local << PAGE_SHIFT;
2087 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb,
2088 PTLRPC_MAX_BRW_PAGES, dbt);
2089 if (unlikely(rc < 0))
2092 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2093 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2096 start_index += nr_local;
2101 ofd_read_unlock(env, fo);
2106 * OFD request handler for OST_LADVISE RPC.
2108 * Tune cache or perfetch policies according to advices.
2110 * \param[in] tsi target session environment for this request
2112 * \retval 0 if successful
2113 * \retval negative errno on error
2115 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2117 struct ptlrpc_request *req = tgt_ses_req(tsi);
2118 struct obd_export *exp = tsi->tsi_exp;
2119 struct ofd_device *ofd = ofd_exp(exp);
2120 struct ost_body *body, *repbody;
2121 struct ofd_thread_info *info;
2122 struct ofd_object *fo;
2123 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2124 const struct lu_env *env = svc_thread->t_env;
2125 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2126 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2127 struct lu_ladvise *ladvise;
2129 struct ladvise_hdr *ladvise_hdr;
2130 struct obd_ioobj ioo;
2131 struct lustre_handle lockh = { 0 };
2134 struct dt_object *dob;
2140 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2141 body = tsi->tsi_ost_body;
2143 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2144 RETURN(err_serious(-EPROTO));
2146 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2147 &RMF_OST_LADVISE_HDR);
2148 if (ladvise_hdr == NULL)
2149 RETURN(err_serious(-EPROTO));
2151 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2152 ladvise_hdr->lah_count < 1)
2153 RETURN(err_serious(-EPROTO));
2155 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2156 RETURN(err_serious(-EPROTO));
2158 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2159 if (ladvise == NULL)
2160 RETURN(err_serious(-EPROTO));
2162 num_advise = req_capsule_get_size(&req->rq_pill,
2163 &RMF_OST_LADVISE, RCL_CLIENT) /
2165 if (num_advise < ladvise_hdr->lah_count)
2166 RETURN(err_serious(-EPROTO));
2168 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2169 repbody->oa = body->oa;
2171 info = ofd_info_init(env, exp);
2173 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2174 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2178 fo = ofd_object_find(env, ofd, &info->fti_fid);
2183 LASSERT(fo != NULL);
2184 dob = ofd_object_child(fo);
2186 if (ptlrpc_connection_is_local(exp->exp_connection))
2187 dbt |= DT_BUFS_TYPE_LOCAL;
2189 for (i = 0; i < num_advise; i++, ladvise++) {
2190 start = ladvise->lla_start;
2191 end = ladvise->lla_end;
2193 rc = err_serious(-EPROTO);
2197 /* Handle different advice types */
2198 switch (ladvise->lla_advice) {
2202 case LU_LADVISE_WILLREAD:
2206 ioo.ioo_oid = body->oa.o_oi;
2208 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2209 &tsi->tsi_resid, start, end - 1,
2210 &lockh, LCK_PR, &flags);
2214 req->rq_status = ofd_ladvise_prefetch(env, fo,
2217 tgt_extent_unlock(&lockh, LCK_PR);
2219 case LU_LADVISE_DONTNEED:
2220 rc = dt_ladvise(env, dob, ladvise->lla_start,
2221 ladvise->lla_end, LU_LADVISE_DONTNEED);
2228 ofd_object_put(env, fo);
2229 req->rq_status = rc;
2234 * OFD request handler for OST_QUOTACTL RPC.
2236 * This is part of request processing to validate incoming request fields,
2237 * get the requested data from OSD and pack reply.
2239 * \param[in] tsi target session environment for this request
2241 * \retval 0 if successful
2242 * \retval negative value on error
2244 static int ofd_quotactl(struct tgt_session_info *tsi)
2246 struct obd_quotactl *oqctl, *repoqc;
2247 struct lu_nodemap *nodemap;
2253 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2255 RETURN(err_serious(-EPROTO));
2257 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2259 RETURN(err_serious(-ENOMEM));
2263 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2264 if (IS_ERR(nodemap))
2265 RETURN(PTR_ERR(nodemap));
2268 if (oqctl->qc_type == USRQUOTA)
2269 id = nodemap_map_id(nodemap, NODEMAP_UID,
2270 NODEMAP_CLIENT_TO_FS,
2272 else if (oqctl->qc_type == GRPQUOTA)
2273 id = nodemap_map_id(nodemap, NODEMAP_GID,
2274 NODEMAP_CLIENT_TO_FS,
2277 nodemap_putref(nodemap);
2279 if (repoqc->qc_id != id)
2280 swap(repoqc->qc_id, id);
2282 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2284 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2287 if (repoqc->qc_id != id)
2288 swap(repoqc->qc_id, id);
2294 * Calculate the amount of time for lock prolongation.
2296 * This is helper for ofd_prolong_extent_locks() function to get
2297 * the timeout extra time.
2299 * \param[in] req current request
2301 * \retval amount of time to extend the timeout with
2303 static inline time64_t prolong_timeout(struct ptlrpc_request *req)
2305 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2306 time64_t req_timeout;
2309 return obd_timeout / 2;
2311 req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec;
2312 return max_t(time64_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
2317 * Prolong lock timeout for the given extent.
2319 * This function finds all locks related with incoming request and
2320 * prolongs their timeout.
2322 * If a client is holding a lock for a long time while it sends
2323 * read or write RPCs to the OST for the object under this lock,
2324 * then we don't want the OST to evict the client. Otherwise,
2325 * if the network or disk is very busy then the client may not
2326 * be able to make any progress to clear out dirty pages under
2327 * the lock and the application will fail.
2329 * Every time a Bulk Read/Write (BRW) request arrives for the object
2330 * covered by the lock, extend the timeout on that lock. The RPC should
2331 * contain a lock handle for the lock it is using, but this
2332 * isn't handled correctly by all client versions, and the
2333 * request may cover multiple locks.
2335 * \param[in] tsi target session environment for this request
2336 * \param[in] data struct of data to prolong locks
2339 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2340 struct ldlm_prolong_args *data)
2342 struct obdo *oa = &tsi->tsi_ost_body->oa;
2343 struct ldlm_lock *lock;
2347 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2348 data->lpa_export = tsi->tsi_exp;
2349 data->lpa_resid = tsi->tsi_resid;
2351 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2352 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2353 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2354 data->lpa_extent.end);
2356 if (oa->o_valid & OBD_MD_FLHANDLE) {
2357 /* mostly a request should be covered by only one lock, try
2359 lock = ldlm_handle2lock(&oa->o_handle);
2361 /* Fast path to check if the lock covers the whole IO
2362 * region exclusively. */
2363 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2364 &data->lpa_extent)) {
2366 LASSERT(lock->l_export == data->lpa_export);
2367 ldlm_lock_prolong_one(lock, data);
2368 LDLM_LOCK_PUT(lock);
2369 if (data->lpa_locks_cnt > 0)
2371 /* The lock was destroyed probably lets try
2374 lock->l_last_used = ktime_get();
2375 LDLM_LOCK_PUT(lock);
2380 ldlm_resource_prolong(data);
2385 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2387 * Determine if \a lock and the lock from request \a req are equivalent
2388 * by comparing their resource names, modes, and extents.
2390 * It is used to give priority to read and write RPCs being done
2391 * under this lock so that the client can drop the contended
2392 * lock more quickly and let other clients use it. This improves
2393 * overall performance in the case where the first client gets a
2394 * very large lock extent that prevents other clients from
2395 * submitting their writes.
2397 * \param[in] req ptlrpc_request being processed
2398 * \param[in] lock contended lock to match
2400 * \retval 1 if lock is matched
2401 * \retval 0 otherwise
2403 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2404 struct ldlm_lock *lock)
2406 struct niobuf_remote *rnb;
2407 struct obd_ioobj *ioo;
2408 enum ldlm_mode mode;
2409 struct ldlm_extent ext;
2410 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2414 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2415 LASSERT(ioo != NULL);
2417 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2418 LASSERT(rnb != NULL);
2420 ext.start = rnb->rnb_offset;
2421 rnb += ioo->ioo_bufcnt - 1;
2422 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2424 LASSERT(lock->l_resource != NULL);
2425 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2428 /* a bulk write can only hold a reference on a PW extent lock
2431 mode = LCK_PW | LCK_GROUP;
2432 if (opc == OST_READ)
2433 /* whereas a bulk read can be protected by either a PR or PW
2437 if (!(lock->l_granted_mode & mode))
2440 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2444 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2446 * Check for whether the given PTLRPC request (\a req) is blocking
2447 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2450 * \param[in] req the incoming request
2452 * \retval 1 if \a req is blocking an LDLM lock cancel
2453 * \retval 0 if it is not
2454 * \retval -ESTALE if lock is not found
2456 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2458 struct tgt_session_info *tsi;
2459 struct obd_ioobj *ioo;
2460 struct niobuf_remote *rnb;
2462 struct ldlm_prolong_args pa = { 0 };
2466 /* Don't use tgt_ses_info() to get session info, because lock_match()
2467 * can be called while request has no processing thread yet. */
2468 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2471 * Use LASSERT below because malformed RPCs should have
2472 * been filtered out in tgt_hpreq_handler().
2474 opc = lustre_msg_get_opc(req->rq_reqmsg);
2475 LASSERT(opc == OST_READ || opc == OST_WRITE);
2477 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2478 LASSERT(ioo != NULL);
2480 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2481 LASSERT(rnb != NULL);
2482 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2484 pa.lpa_mode = LCK_PW | LCK_GROUP;
2485 if (opc == OST_READ)
2486 pa.lpa_mode |= LCK_PR;
2488 pa.lpa_extent.start = rnb->rnb_offset;
2489 rnb += ioo->ioo_bufcnt - 1;
2490 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2492 DEBUG_REQ(D_RPCTRACE, req,
2493 "%s %s: refresh rw locks for "DFID" (%llu->%llu)",
2494 tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid),
2495 pa.lpa_extent.start, pa.lpa_extent.end);
2497 ofd_prolong_extent_locks(tsi, &pa);
2499 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n",
2500 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2502 if (pa.lpa_blocks_cnt > 0)
2505 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2509 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2511 * Called after the request has been handled. It refreshes lock timeout again
2512 * so that client has more time to send lock cancel RPC.
2514 * \param[in] req request which is being processed.
2516 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2518 ofd_rw_hpreq_check(req);
2522 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2524 * This function checks if the given lock is the same by its resname, mode
2525 * and extent as one taken from the request.
2526 * It is used to give priority to punch/truncate RPCs that might lead to
2527 * the fastest release of that lock when a lock is contended.
2529 * \param[in] req ptlrpc_request being processed
2530 * \param[in] lock contended lock to match
2532 * \retval 1 if lock is matched
2533 * \retval 0 otherwise
2535 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2536 struct ldlm_lock *lock)
2538 struct tgt_session_info *tsi;
2540 struct ldlm_extent ext;
2544 /* Don't use tgt_ses_info() to get session info, because lock_match()
2545 * can be called while request has no processing thread yet. */
2546 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2549 * Use LASSERT below because malformed RPCs should have
2550 * been filtered out in tgt_hpreq_handler().
2552 LASSERT(tsi->tsi_ost_body != NULL);
2553 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2554 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2557 oa = &tsi->tsi_ost_body->oa;
2558 ext.start = oa->o_size;
2559 ext.end = oa->o_blocks;
2561 LASSERT(lock->l_resource != NULL);
2562 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2565 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2568 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2572 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2574 * High-priority queue request check for whether the given punch request
2575 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2576 * covered by an LDLM lock.
2580 * \param[in] req the incoming request
2582 * \retval 1 if \a req is blocking an LDLM lock cancel
2583 * \retval 0 if it is not
2584 * \retval -ESTALE if lock is not found
2586 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2588 struct tgt_session_info *tsi;
2590 struct ldlm_prolong_args pa = { 0 };
2594 /* Don't use tgt_ses_info() to get session info, because lock_match()
2595 * can be called while request has no processing thread yet. */
2596 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2597 LASSERT(tsi != NULL);
2598 oa = &tsi->tsi_ost_body->oa;
2600 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2601 oa->o_flags & OBD_FL_SRVLOCK));
2603 pa.lpa_mode = LCK_PW | LCK_GROUP;
2604 pa.lpa_extent.start = oa->o_size;
2605 pa.lpa_extent.end = oa->o_blocks;
2608 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2609 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2610 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2612 ofd_prolong_extent_locks(tsi, &pa);
2614 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2615 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2617 if (pa.lpa_blocks_cnt > 0)
2620 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2624 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2626 * Called after the request has been handled. It refreshes lock timeout again
2627 * so that client has more time to send lock cancel RPC.
2629 * \param[in] req request which is being processed.
2631 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2633 ofd_punch_hpreq_check(req);
2636 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2637 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2638 .hpreq_check = ofd_rw_hpreq_check,
2639 .hpreq_fini = ofd_rw_hpreq_fini
2642 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2643 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2644 .hpreq_check = ofd_punch_hpreq_check,
2645 .hpreq_fini = ofd_punch_hpreq_fini
2649 * Assign high priority operations to an IO request.
2651 * Check if the incoming request is a candidate for
2652 * high-priority processing. If it is, assign it a high
2653 * priority operations table.
2655 * \param[in] tsi target session environment for this request
2657 static void ofd_hp_brw(struct tgt_session_info *tsi)
2659 struct niobuf_remote *rnb;
2660 struct obd_ioobj *ioo;
2664 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2665 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2666 if (ioo->ioo_bufcnt > 0) {
2667 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2668 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2670 /* no high priority if server lock is needed */
2671 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2672 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2676 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2680 * Assign high priority operations to an punch request.
2682 * Check if the incoming request is a candidate for
2683 * high-priority processing. If it is, assign it a high
2684 * priority operations table.
2686 * \param[in] tsi target session environment for this request
2688 static void ofd_hp_punch(struct tgt_session_info *tsi)
2690 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2691 /* no high-priority if server lock is needed */
2692 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2693 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2694 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2695 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2697 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2700 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2701 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2702 #define OST_BRW_READ OST_READ
2703 #define OST_BRW_WRITE OST_WRITE
2706 * Table of OFD-specific request handlers
2708 * This table contains all opcodes accepted by OFD and
2709 * specifies handlers for them. The tgt_request_handler()
2710 * uses such table from each target to process incoming
2713 static struct tgt_handler ofd_tgt_handlers[] = {
2714 TGT_RPC_HANDLER(OST_FIRST_OPC,
2715 0, OST_CONNECT, tgt_connect,
2716 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2717 TGT_RPC_HANDLER(OST_FIRST_OPC,
2718 0, OST_DISCONNECT, tgt_disconnect,
2719 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2720 TGT_RPC_HANDLER(OST_FIRST_OPC,
2721 0, OST_SET_INFO, ofd_set_info_hdl,
2722 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2723 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2724 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl),
2725 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2726 OST_SETATTR, ofd_setattr_hdl),
2727 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2728 OST_CREATE, ofd_create_hdl),
2729 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2730 OST_DESTROY, ofd_destroy_hdl),
2731 TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl),
2732 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
2734 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2735 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write,
2737 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2738 OST_PUNCH, ofd_punch_hdl,
2740 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl),
2741 TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl),
2742 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl),
2745 static struct tgt_opc_slice ofd_common_slice[] = {
2747 .tos_opc_start = OST_FIRST_OPC,
2748 .tos_opc_end = OST_LAST_OPC,
2749 .tos_hs = ofd_tgt_handlers
2752 .tos_opc_start = OBD_FIRST_OPC,
2753 .tos_opc_end = OBD_LAST_OPC,
2754 .tos_hs = tgt_obd_handlers
2757 .tos_opc_start = LDLM_FIRST_OPC,
2758 .tos_opc_end = LDLM_LAST_OPC,
2759 .tos_hs = tgt_dlm_handlers
2762 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2763 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2764 .tos_hs = tgt_out_handlers
2767 .tos_opc_start = SEQ_FIRST_OPC,
2768 .tos_opc_end = SEQ_LAST_OPC,
2769 .tos_hs = seq_handlers
2772 .tos_opc_start = LFSCK_FIRST_OPC,
2773 .tos_opc_end = LFSCK_LAST_OPC,
2774 .tos_hs = tgt_lfsck_handlers
2777 .tos_opc_start = SEC_FIRST_OPC,
2778 .tos_opc_end = SEC_LAST_OPC,
2779 .tos_hs = tgt_sec_ctx_handlers
2786 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2787 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2790 * Implementation of lu_context_key::lct_key_exit.
2792 * Optional method called on lu_context_exit() for all allocated
2794 * It is used in OFD to sanitize context values which may be re-used
2795 * during another request processing by the same thread.
2797 * \param[in] ctx execution context
2798 * \param[in] key context key
2799 * \param[in] data ofd_thread_info
2801 static void ofd_key_exit(const struct lu_context *ctx,
2802 struct lu_context_key *key, void *data)
2804 struct ofd_thread_info *info = data;
2806 info->fti_env = NULL;
2807 info->fti_exp = NULL;
2810 info->fti_pre_version = 0;
2812 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2815 struct lu_context_key ofd_thread_key = {
2816 .lct_tags = LCT_DT_THREAD,
2817 .lct_init = ofd_key_init,
2818 .lct_fini = ofd_key_fini,
2819 .lct_exit = ofd_key_exit
2823 * Initialize OFD device according to parameters in the config log \a cfg.
2825 * This is the main starting point of OFD initialization. It fills all OFD
2826 * parameters with their initial values and calls other initializing functions
2827 * to set up all OFD subsystems.
2829 * \param[in] env execution environment
2830 * \param[in] m OFD device
2831 * \param[in] ldt LU device type of OFD
2832 * \param[in] cfg configuration log
2834 * \retval 0 if successful
2835 * \retval negative value on error
2837 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2838 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2840 const char *dev = lustre_cfg_string(cfg, 0);
2841 struct ofd_thread_info *info = NULL;
2842 struct obd_device *obd;
2843 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2845 struct nm_config_file *nodemap_config;
2846 struct obd_device_target *obt;
2852 obd = class_name2obd(dev);
2854 CERROR("Cannot find obd with name %s\n", dev);
2858 rc = lu_env_refill((struct lu_env *)env);
2863 obt->obt_magic = OBT_MAGIC;
2865 spin_lock_init(&m->ofd_flags_lock);
2866 m->ofd_raid_degraded = 0;
2867 m->ofd_checksum_t10pi_enforce = 0;
2868 m->ofd_sync_journal = 0;
2870 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2872 m->ofd_seq_count = 0;
2873 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2874 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2875 spin_lock_init(&m->ofd_inconsistency_lock);
2877 spin_lock_init(&m->ofd_batch_lock);
2878 init_rwsem(&m->ofd_lastid_rwsem);
2880 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2881 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2882 /* set this lu_device to obd, because error handling need it */
2883 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2885 /* No connection accepted until configurations will finish */
2886 spin_lock(&obd->obd_dev_lock);
2887 obd->obd_no_conn = 1;
2888 spin_unlock(&obd->obd_dev_lock);
2889 obd->obd_replayable = 1;
2890 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2891 char *str = lustre_cfg_string(cfg, 4);
2893 if (strchr(str, 'n')) {
2894 CWARN("%s: recovery disabled\n", obd->obd_name);
2895 obd->obd_replayable = 0;
2899 info = ofd_info_init(env, NULL);
2903 rc = ofd_stack_init(env, m, cfg, &lmd_flags);
2905 CERROR("%s: can't init device stack, rc %d\n",
2910 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
2911 ofd_procfs_add_brw_stats_symlink(m);
2914 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2915 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2916 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2917 LDLM_NAMESPACE_SERVER,
2918 LDLM_NAMESPACE_GREEDY,
2920 if (m->ofd_namespace == NULL)
2921 GOTO(err_fini_stack, rc = -ENOMEM);
2922 /* set obd_namespace for compatibility with old code */
2923 obd->obd_namespace = m->ofd_namespace;
2924 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2925 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2926 m->ofd_namespace->ns_lvbp = m;
2928 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2929 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2931 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2932 OBD_FAIL_OST_ALL_REQUEST_NET,
2933 OBD_FAIL_OST_ALL_REPLY_NET);
2935 GOTO(err_free_ns, rc);
2937 if (lmd_flags & LMD_FLG_SKIP_LFSCK)
2938 m->ofd_skip_lfsck = 1;
2939 if (lmd_flags & LMD_FLG_LOCAL_RECOV)
2940 m->ofd_lut.lut_local_recovery = 1;
2942 rc = ofd_tunables_init(m);
2944 GOTO(err_fini_lut, rc);
2946 tgd->tgd_reserved_pcnt = 0;
2948 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
2949 m->ofd_cksum_types_supported =
2950 obd_cksum_types_supported_server(obd->obd_name);
2951 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2952 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
2953 OFD_PRECREATE_SMALL_FS)
2954 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2956 rc = ofd_fs_setup(env, m, obd);
2958 GOTO(err_fini_proc, rc);
2960 fid.f_seq = FID_SEQ_LOCAL_NAME;
2963 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
2966 GOTO(err_fini_fs, rc);
2968 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
2970 if (IS_ERR(nodemap_config)) {
2971 rc = PTR_ERR(nodemap_config);
2973 GOTO(err_fini_los, rc);
2975 obt->obt_nodemap_config_file = nodemap_config;
2978 rc = ofd_start_inconsistency_verification_thread(m);
2980 GOTO(err_fini_nm, rc);
2982 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
2987 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
2988 obt->obt_nodemap_config_file = NULL;
2990 local_oid_storage_fini(env, m->ofd_los);
2993 ofd_fs_cleanup(env, m);
2997 tgt_fini(env, &m->ofd_lut);
2999 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3000 obd->obd_namespace = m->ofd_namespace = NULL;
3002 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3007 * Stop the OFD device
3009 * This function stops the OFD device and all its subsystems.
3010 * This is the end of OFD lifecycle.
3012 * \param[in] env execution environment
3013 * \param[in] m OFD device
3015 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3017 struct obd_device *obd = ofd_obd(m);
3018 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3019 struct lfsck_stop stop;
3021 stop.ls_status = LS_PAUSED;
3023 lfsck_stop(env, m->ofd_osd, &stop);
3024 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3025 target_recovery_fini(obd);
3026 if (m->ofd_namespace != NULL)
3027 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3028 d->ld_obd->obd_force);
3030 obd_exports_barrier(obd);
3031 obd_zombie_barrier();
3034 tgt_fini(env, &m->ofd_lut);
3035 ofd_stop_inconsistency_verification_thread(m);
3036 lfsck_degister(env, m->ofd_osd);
3037 ofd_fs_cleanup(env, m);
3038 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3039 obd->u.obt.obt_nodemap_config_file = NULL;
3041 if (m->ofd_namespace != NULL) {
3042 ldlm_namespace_free_post(m->ofd_namespace);
3043 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3046 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3048 LASSERT(atomic_read(&d->ld_ref) == 0);
3049 server_put_mount(obd->obd_name, true);
3054 * Implementation of lu_device_type_operations::ldto_device_fini.
3056 * Finalize device. Dual to ofd_device_init(). It is called from
3057 * obd_precleanup() and stops the current device.
3059 * \param[in] env execution environment
3060 * \param[in] d LU device of OFD
3064 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3065 struct lu_device *d)
3068 ofd_fini(env, ofd_dev(d));
3073 * Implementation of lu_device_type_operations::ldto_device_free.
3075 * Free OFD device. Dual to ofd_device_alloc().
3077 * \param[in] env execution environment
3078 * \param[in] d LU device of OFD
3082 static struct lu_device *ofd_device_free(const struct lu_env *env,
3083 struct lu_device *d)
3085 struct ofd_device *m = ofd_dev(d);
3087 dt_device_fini(&m->ofd_dt_dev);
3093 * Implementation of lu_device_type_operations::ldto_device_alloc.
3095 * This function allocates the new OFD device. It is called from
3096 * obd_setup() if OBD device had lu_device_type defined.
3098 * \param[in] env execution environment
3099 * \param[in] t lu_device_type of OFD device
3100 * \param[in] cfg configuration log
3102 * \retval pointer to the lu_device of just allocated OFD
3103 * \retval ERR_PTR of return value on error
3105 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3106 struct lu_device_type *t,
3107 struct lustre_cfg *cfg)
3109 struct ofd_device *m;
3110 struct lu_device *l;
3115 return ERR_PTR(-ENOMEM);
3117 l = &m->ofd_dt_dev.dd_lu_dev;
3118 dt_device_init(&m->ofd_dt_dev, t);
3119 rc = ofd_init0(env, m, t, cfg);
3121 ofd_device_free(env, l);
3128 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3129 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3131 static struct lu_device_type_operations ofd_device_type_ops = {
3132 .ldto_init = ofd_type_init,
3133 .ldto_fini = ofd_type_fini,
3135 .ldto_start = ofd_type_start,
3136 .ldto_stop = ofd_type_stop,
3138 .ldto_device_alloc = ofd_device_alloc,
3139 .ldto_device_free = ofd_device_free,
3140 .ldto_device_fini = ofd_device_fini
3143 static struct lu_device_type ofd_device_type = {
3144 .ldt_tags = LU_DEVICE_DT,
3145 .ldt_name = LUSTRE_OST_NAME,
3146 .ldt_ops = &ofd_device_type_ops,
3147 .ldt_ctx_tags = LCT_DT_THREAD
3151 * Initialize OFD module.
3153 * This function is called upon module loading. It registers OFD device type
3154 * and prepares all in-memory structures used by all OFD devices.
3156 * \retval 0 if successful
3157 * \retval negative value on error
3159 static int __init ofd_init(void)
3163 rc = lu_kmem_init(ofd_caches);
3166 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3167 LUSTRE_OST_NAME, &ofd_device_type);
3174 * This function is called upon OFD module unloading.
3175 * It frees all related structures and unregisters OFD device type.
3177 static void __exit ofd_exit(void)
3179 lu_kmem_fini(ofd_caches);
3180 class_unregister_type(LUSTRE_OST_NAME);
3183 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3184 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3185 MODULE_VERSION(LUSTRE_VERSION_STRING);
3186 MODULE_LICENSE("GPL");
3188 module_init(ofd_init);
3189 module_exit(ofd_exit);