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)
166 const char *dev = lustre_cfg_string(cfg, 0);
168 struct ofd_thread_info *info = ofd_info(env);
169 struct lustre_mount_info *lmi;
170 struct lustre_mount_data *lmd;
176 lmi = server_get_mount(dev);
178 CERROR("Cannot get mount info for %s!\n", dev);
182 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
183 if (lmd && lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
184 m->ofd_skip_lfsck = 1;
186 if (lmd && lmd->lmd_flags & LMD_FLG_NO_PRECREATE)
187 m->ofd_no_precreate = 1;
189 /* find bottom osd */
190 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
194 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
195 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
196 OBD_FREE(osdname, MTI_NAME_MAXLEN);
200 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
202 m->ofd_osd = lu2dt_dev(d);
204 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
205 "%s-osd", lustre_cfg_string(cfg, 0));
211 * Finalize the device stack OFD-OSD.
213 * This function cleans OFD-OSD device stack and
214 * disconnects OFD from the OSD.
216 * \param[in] env execution environment
217 * \param[in] m OFD device
218 * \param[in] top top device of stack
220 * \retval 0 if successful
221 * \retval negative value on error
223 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
224 struct lu_device *top)
226 struct obd_device *obd = ofd_obd(m);
227 struct lustre_cfg_bufs bufs;
228 struct lustre_cfg *lcfg;
233 lu_site_purge(env, top->ld_site, ~0);
234 /* process cleanup, pass mdt obd name to get obd umount flags */
235 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
240 lustre_cfg_bufs_set_string(&bufs, 1, flags);
241 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
244 lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs);
247 top->ld_ops->ldo_process_config(env, top, lcfg);
248 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
250 if (m->ofd_los != NULL) {
251 local_oid_storage_fini(env, m->ofd_los);
255 lu_site_purge(env, top->ld_site, ~0);
256 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
257 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
258 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
261 LASSERT(m->ofd_osd_exp);
262 obd_disconnect(m->ofd_osd_exp);
267 static void ofd_stack_pre_fini(const struct lu_env *env, struct ofd_device *m,
268 struct lu_device *top)
270 struct lustre_cfg_bufs bufs;
271 struct lustre_cfg *lcfg;
276 lustre_cfg_bufs_reset(&bufs, ofd_name(m));
277 lustre_cfg_bufs_set_string(&bufs, 1, NULL);
278 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
280 CERROR("%s: failed to trigger LCFG_PRE_CLEANUP\n", ofd_name(m));
282 lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, &bufs);
283 top->ld_ops->ldo_process_config(env, top, lcfg);
284 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
285 lcfg->lcfg_buflens));
291 /* For interoperability, see mdt_interop_param[]. */
292 static struct cfg_interop_param ofd_interop_param[] = {
293 { "ost.quota_type", NULL },
298 * Check if parameters are symlinks to the OSD.
300 * Some parameters were moved from ofd to osd and only their
301 * symlinks were kept in ofd by LU-3106. They are:
302 * -writehthrough_cache_enable
303 * -readcache_max_filesize
307 * Since they are not included by the static lprocfs var list, a pre-check
308 * is added for them to avoid "unknown param" errors. If they are matched
309 * in this check, they will be passed to the OSD directly.
311 * \param[in] param parameters to check
313 * \retval true if param is symlink to OSD param
316 static bool match_symlink_param(char *param)
321 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
322 sval = strchr(param, '=');
324 paramlen = sval - param;
325 if (strncmp(param, "brw_stats", paramlen) == 0)
334 * Process various configuration parameters.
336 * This function is used by MGS to process specific configurations and
337 * pass them through to the next device in server stack, i.e. the OSD.
339 * \param[in] env execution environment
340 * \param[in] d LU device of OFD
341 * \param[in] cfg parameters to process
343 * \retval 0 if successful
344 * \retval negative value on error
346 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
347 struct lustre_cfg *cfg)
349 struct ofd_device *m = ofd_dev(d);
350 struct dt_device *dt_next = m->ofd_osd;
351 struct lu_device *next = &dt_next->dd_lu_dev;
356 switch (cfg->lcfg_command) {
358 /* For interoperability */
359 struct cfg_interop_param *ptr = NULL;
360 struct lustre_cfg *old_cfg = NULL;
364 param = lustre_cfg_string(cfg, 1);
366 CERROR("param is empty\n");
371 ptr = class_find_old_param(param, ofd_interop_param);
373 if (ptr->new_param == NULL) {
375 CWARN("For interoperability, skip this %s."
376 " It is obsolete.\n", ptr->old_param);
380 CWARN("Found old param %s, changed it to %s.\n",
381 ptr->old_param, ptr->new_param);
384 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
391 if (match_symlink_param(param)) {
392 rc = next->ld_ops->ldo_process_config(env, next, cfg);
396 count = class_modify_config(cfg, PARAM_OST,
397 &d->ld_obd->obd_kset.kobj);
402 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
404 /* we don't understand; pass it on */
405 rc = next->ld_ops->ldo_process_config(env, next, cfg);
408 case LCFG_SPTLRPC_CONF: {
413 /* others are passed further */
414 rc = next->ld_ops->ldo_process_config(env, next, cfg);
421 * Implementation of lu_object_operations::loo_object_init for OFD
423 * Allocate just the next object (OSD) in stack.
425 * \param[in] env execution environment
426 * \param[in] o lu_object of OFD object
427 * \param[in] conf additional configuration parameters, not used here
429 * \retval 0 if successful
430 * \retval negative value on error
432 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
433 const struct lu_object_conf *conf)
435 struct ofd_device *d = ofd_dev(o->lo_dev);
436 struct lu_device *under;
437 struct lu_object *below;
442 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
443 PFID(lu_object_fid(o)));
445 under = &d->ofd_osd->dd_lu_dev;
446 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
448 lu_object_add(o, below);
456 * Implementation of lu_object_operations::loo_object_free.
458 * Finish OFD object lifecycle and free its memory.
460 * \param[in] env execution environment
461 * \param[in] o LU object of OFD object
463 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
465 struct ofd_object *of = ofd_obj(o);
466 struct lu_object_header *h;
471 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
472 PFID(lu_object_fid(o)));
475 lu_object_header_fini(h);
476 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
481 * Implementation of lu_object_operations::loo_object_print.
483 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
484 * LU_OBJECT_DEBUG() for more details about the compound object printing.
486 * \param[in] env execution environment
487 * \param[in] cookie opaque data passed to the printer function
488 * \param[in] p printer function to use
489 * \param[in] o LU object of OFD object
491 * \retval 0 if successful
492 * \retval negative value on error
494 static int ofd_object_print(const struct lu_env *env, void *cookie,
495 lu_printer_t p, const struct lu_object *o)
497 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
500 static struct lu_object_operations ofd_obj_ops = {
501 .loo_object_init = ofd_object_init,
502 .loo_object_free = ofd_object_free,
503 .loo_object_print = ofd_object_print
507 * Implementation of lu_device_operations::lod_object_alloc.
509 * This function allocates OFD part of compound OFD-OSD object and
510 * initializes its header, because OFD is the top device in stack
512 * \param[in] env execution environment
513 * \param[in] hdr object header, NULL for OFD
514 * \param[in] d lu_device
516 * \retval allocated object if successful
517 * \retval NULL value on failed allocation
519 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
520 const struct lu_object_header *hdr,
523 struct ofd_object *of;
527 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
530 struct lu_object_header *h;
532 o = &of->ofo_obj.do_lu;
534 lu_object_header_init(h);
535 lu_object_init(o, h, d);
536 lu_object_add_top(h, o);
537 o->lo_ops = &ofd_obj_ops;
545 * Return the result of LFSCK run to the OFD.
547 * Notify OFD about result of LFSCK run. That may block the new object
548 * creation until problem is fixed by LFSCK.
550 * \param[in] env execution environment
551 * \param[in] data pointer to the OFD device
552 * \param[in] event LFSCK event type
554 * \retval 0 if successful
555 * \retval negative value on unknown event
557 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
558 enum lfsck_events event)
560 struct ofd_device *ofd = data;
561 struct obd_device *obd = ofd_obd(ofd);
564 case LE_LASTID_REBUILDING:
565 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
566 "on the device until the LAST_ID rebuilt successfully.\n",
568 down_write(&ofd->ofd_lastid_rwsem);
569 ofd->ofd_lastid_rebuilding = 1;
570 up_write(&ofd->ofd_lastid_rwsem);
572 case LE_LASTID_REBUILT: {
573 down_write(&ofd->ofd_lastid_rwsem);
574 ofd_seqs_free(env, ofd);
575 ofd->ofd_lastid_rebuilding = 0;
576 ofd->ofd_lastid_gen++;
577 up_write(&ofd->ofd_lastid_rwsem);
578 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
583 CERROR("%s: unknown lfsck event: rc = %d\n",
584 ofd_name(ofd), event);
592 * Implementation of lu_device_operations::ldo_prepare.
594 * This method is called after layer has been initialized and before it starts
595 * serving user requests. In OFD it starts lfsk check routines and initializes
598 * \param[in] env execution environment
599 * \param[in] pdev higher device in stack, NULL for OFD
600 * \param[in] dev lu_device of OFD device
602 * \retval 0 if successful
603 * \retval negative value on error
605 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
606 struct lu_device *dev)
608 struct ofd_thread_info *info;
609 struct ofd_device *ofd = ofd_dev(dev);
610 struct obd_device *obd = ofd_obd(ofd);
611 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
616 info = ofd_info_init(env, NULL);
620 /* initialize lower device */
621 rc = next->ld_ops->ldo_prepare(env, dev, next);
625 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
626 ofd_lfsck_out_notify, ofd, false);
628 CERROR("%s: failed to initialize lfsck: rc = %d\n",
633 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
634 /* The LFSCK instance is registered just now, so it must be there when
635 * register the namespace to such instance. */
636 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
638 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
639 OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_OST_PREPARE_DELAY, OBD_FAIL_ONCE,
640 (OBD_TIMEOUT_DEFAULT + 1) / 4);
641 LASSERT(obd->obd_no_conn);
642 spin_lock(&obd->obd_dev_lock);
643 obd->obd_no_conn = 0;
644 spin_unlock(&obd->obd_dev_lock);
646 if (obd->obd_recovering == 0)
647 ofd_postrecov(env, ofd);
653 * Implementation of lu_device_operations::ldo_recovery_complete.
655 * This method notifies all layers about 'recovery complete' event. That means
656 * device is in full state and consistent. An OFD calculates available grant
657 * space upon this event.
659 * \param[in] env execution environment
660 * \param[in] dev lu_device of OFD device
662 * \retval 0 if successful
663 * \retval negative value on error
665 static int ofd_recovery_complete(const struct lu_env *env,
666 struct lu_device *dev)
668 struct ofd_thread_info *oti = ofd_info(env);
669 struct ofd_device *ofd = ofd_dev(dev);
670 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
676 * Grant space for object precreation on the self export.
677 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
678 * is enough to create 10k objects. More space is then acquired for
679 * precreation in tgt_grant_create().
681 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
682 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
683 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
684 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
685 OBD_CONNECT_GRANT_PARAM;
686 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
688 rc = next->ld_ops->ldo_recovery_complete(env, next);
693 * lu_device_operations matrix for OFD device.
695 static struct lu_device_operations ofd_lu_ops = {
696 .ldo_object_alloc = ofd_object_alloc,
697 .ldo_process_config = ofd_process_config,
698 .ldo_recovery_complete = ofd_recovery_complete,
699 .ldo_prepare = ofd_prepare,
702 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
704 * Expose OSD statistics to OFD layer.
706 * The osd interfaces to the backend file system exposes useful data
707 * such as brw_stats and read or write cache states. This same data
708 * needs to be exposed into the obdfilter (ofd) layer to maintain
709 * backwards compatibility. This function creates the symlinks in the
710 * proc layer to enable this.
712 * \param[in] ofd OFD device
714 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
716 struct obd_device *obd = ofd_obd(ofd);
717 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
718 struct kobj_type *osd_type;
721 osd_type = get_ktype(&ofd->ofd_osd->dd_kobj);
722 for (i = 0; osd_type->default_attrs[i]; i++) {
723 if (strcmp(osd_type->default_attrs[i]->name,
724 "read_cache_enable") == 0) {
725 ofd->ofd_read_cache_enable =
726 osd_type->default_attrs[i];
729 if (strcmp(osd_type->default_attrs[i]->name,
730 "readcache_max_filesize") == 0) {
731 ofd->ofd_read_cache_max_filesize =
732 osd_type->default_attrs[i];
735 if (strcmp(osd_type->default_attrs[i]->name,
736 "writethrough_cache_enable") == 0) {
737 ofd->ofd_write_cache_enable =
738 osd_type->default_attrs[i];
742 if (obd->obd_proc_entry == NULL)
745 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
746 "../../%s/%s/brw_stats",
747 osd_obd->obd_type->typ_name, obd->obd_name);
752 * Cleanup all procfs entries in OFD.
754 * \param[in] ofd OFD device
756 static void ofd_procfs_fini(struct ofd_device *ofd)
758 struct obd_device *obd = ofd_obd(ofd);
760 tgt_tunables_fini(&ofd->ofd_lut);
761 lprocfs_free_per_client_stats(obd);
762 lprocfs_obd_cleanup(obd);
763 lprocfs_free_obd_stats(obd);
764 lprocfs_job_stats_fini(obd);
768 * Stop SEQ/FID server on OFD.
770 * \param[in] env execution environment
771 * \param[in] ofd OFD device
773 * \retval 0 if successful
774 * \retval negative value on error
776 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
778 return seq_site_fini(env, &ofd->ofd_seq_site);
782 * Start SEQ/FID server on OFD.
784 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
785 * It also connects to the master server to get own FID sequence (SEQ) range
786 * to this particular OFD. Typically that happens when the OST is first
787 * formatted or in the rare case that it exhausts the local sequence range.
789 * The sequence range is allocated out to the MDTs for OST object allocations,
790 * and not directly to the clients.
792 * \param[in] env execution environment
793 * \param[in] ofd OFD device
795 * \retval 0 if successful
796 * \retval negative value on error
798 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
800 struct seq_server_site *ss = &ofd->ofd_seq_site;
801 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
802 char *obd_name = ofd_name(ofd);
804 int len = strlen(obd_name) + 7;
807 ss = &ofd->ofd_seq_site;
808 lu->ld_site->ld_seq_site = ss;
809 ss->ss_lu = lu->ld_site;
810 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
812 OBD_ALLOC(name, len);
816 OBD_ALLOC_PTR(ss->ss_server_seq);
817 if (ss->ss_server_seq == NULL)
818 GOTO(out_name, rc = -ENOMEM);
820 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
821 LUSTRE_SEQ_SERVER, ss);
823 CERROR("%s: seq server init error: rc = %d\n", obd_name, rc);
824 GOTO(out_server, rc);
826 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
828 OBD_ALLOC_PTR(ss->ss_client_seq);
829 if (ss->ss_client_seq == NULL)
830 GOTO(out_server, rc = -ENOMEM);
832 snprintf(name, len, "%s-super", obd_name);
833 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
836 CERROR("%s: seq client init error: rc = %d\n", obd_name, rc);
837 GOTO(out_client, rc);
840 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
844 seq_client_fini(ss->ss_client_seq);
845 OBD_FREE_PTR(ss->ss_client_seq);
846 ss->ss_client_seq = NULL;
848 seq_server_fini(ss->ss_server_seq, env);
849 OBD_FREE_PTR(ss->ss_server_seq);
850 ss->ss_server_seq = NULL;
859 * OFD request handler for OST_SET_INFO RPC.
861 * This is OFD-specific part of request handling
863 * \param[in] tsi target session environment for this request
865 * \retval 0 if successful
866 * \retval negative value on error
868 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
870 struct ptlrpc_request *req = tgt_ses_req(tsi);
871 struct ost_body *body = NULL, *repbody;
872 void *key, *val = NULL;
873 int keylen, vallen, rc = 0;
874 bool is_grant_shrink;
878 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
880 DEBUG_REQ(D_HA, req, "no set_info key");
881 RETURN(err_serious(-EFAULT));
883 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
886 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
888 DEBUG_REQ(D_HA, req, "no set_info val");
889 RETURN(err_serious(-EFAULT));
891 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
894 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
896 /* In this case the value is actually an RMF_OST_BODY, so we
897 * transmutate the type of this PTLRPC */
898 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
900 rc = req_capsule_server_pack(tsi->tsi_pill);
904 if (is_grant_shrink) {
905 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
907 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
910 /** handle grant shrink, similar to a read request */
911 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
913 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
915 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
918 CERROR("%s: Unsupported key %s\n",
919 tgt_name(tsi->tsi_tgt), (char *)key);
922 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
929 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
931 * This function returns a list of extents which describes how a file's
932 * blocks are laid out on the disk.
934 * \param[in] env execution environment
935 * \param[in] ofd OFD device
936 * \param[in] fid FID of object
937 * \param[in] fiemap fiemap structure to fill with data
939 * \retval 0 if \a fiemap is filled with data successfully
940 * \retval negative value on error
942 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
943 struct lu_fid *fid, struct fiemap *fiemap)
945 struct ofd_object *fo;
948 fo = ofd_object_find(env, ofd, fid);
950 CERROR("%s: error finding object "DFID"\n",
951 ofd_name(ofd), PFID(fid));
955 ofd_read_lock(env, fo);
956 if (ofd_object_exists(fo))
957 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
960 ofd_read_unlock(env, fo);
961 ofd_object_put(env, fo);
966 static int ofd_lock_unlock_region(const struct lu_env *env,
967 struct ldlm_namespace *ns,
968 struct ldlm_res_id *res_id,
969 unsigned long long begin,
970 unsigned long long end)
974 struct lustre_handle lh = { 0 };
976 LASSERT(begin <= end);
978 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
982 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
983 tgt_extent_unlock(&lh, LCK_PR);
989 * Lock the sparse areas of given resource.
991 * The locking of sparse areas will cause dirty data to be flushed back from
992 * clients. This is used when getting the FIEMAP of an object to make sure
993 * there is no unaccounted cached data on clients.
995 * This function goes through \a fiemap list of extents and locks only sparse
996 * areas between extents.
998 * \param[in] ns LDLM namespace
999 * \param[in] res_id resource ID
1000 * \param[in] fiemap file extents mapping on disk
1001 * \param[in] locked list head of regions list
1003 * \retval 0 if successful
1004 * \retval negative value on error
1006 static int lock_zero_regions(const struct lu_env *env,
1007 struct ldlm_namespace *ns,
1008 struct ldlm_res_id *res_id,
1009 struct fiemap *fiemap)
1011 __u64 begin = fiemap->fm_start;
1014 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1018 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1019 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1020 if (fiemap_start[i].fe_logical > begin) {
1021 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1022 begin, fiemap_start[i].fe_logical);
1023 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1024 fiemap_start[i].fe_logical);
1029 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1032 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1033 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1034 begin, fiemap->fm_start + fiemap->fm_length);
1035 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1036 fiemap->fm_start + fiemap->fm_length);
1044 * OFD request handler for OST_GET_INFO RPC.
1046 * This is OFD-specific part of request handling. The OFD-specific keys are:
1047 * - KEY_LAST_ID (obsolete)
1051 * This function reads needed data from storage and fills reply with it.
1053 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1054 * and is kept for compatibility.
1056 * \param[in] tsi target session environment for this request
1058 * \retval 0 if successful
1059 * \retval negative value on error
1061 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1063 struct obd_export *exp = tsi->tsi_exp;
1064 struct ofd_device *ofd = ofd_exp(exp);
1065 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1068 int replylen, rc = 0;
1072 /* this common part for get_info rpc */
1073 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1075 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1076 RETURN(err_serious(-EPROTO));
1078 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1081 if (KEY_IS(KEY_LAST_ID)) {
1083 struct ofd_seq *oseq;
1085 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1086 rc = req_capsule_server_pack(tsi->tsi_pill);
1088 RETURN(err_serious(rc));
1090 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1092 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1093 (u64)exp->exp_filter_data.fed_group);
1097 *last_id = ofd_seq_last_oid(oseq);
1098 ofd_seq_put(tsi->tsi_env, oseq);
1099 } else if (KEY_IS(KEY_FIEMAP)) {
1100 struct ll_fiemap_info_key *fm_key;
1101 struct fiemap *fiemap;
1104 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1106 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1107 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1109 RETURN(err_serious(rc));
1111 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1113 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1115 replylen = fiemap_count_to_size(
1116 fm_key->lfik_fiemap.fm_extent_count);
1117 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1118 RCL_SERVER, replylen);
1120 rc = req_capsule_server_pack(tsi->tsi_pill);
1122 RETURN(err_serious(rc));
1124 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1128 *fiemap = fm_key->lfik_fiemap;
1129 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1131 /* LU-3219: Lock the sparse areas to make sure dirty
1132 * flushed back from client, then call fiemap again. */
1133 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1134 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1135 ost_fid_build_resid(fid, &fti->fti_resid);
1136 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1137 &fti->fti_resid, fiemap);
1139 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1142 } else if (KEY_IS(KEY_LAST_FID)) {
1143 struct ofd_device *ofd = ofd_exp(exp);
1144 struct ofd_seq *oseq;
1148 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1149 rc = req_capsule_server_pack(tsi->tsi_pill);
1151 RETURN(err_serious(rc));
1153 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1155 RETURN(err_serious(-EPROTO));
1157 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1159 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1163 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1164 ostid_seq(&fti->fti_ostid));
1166 RETURN(PTR_ERR(oseq));
1168 rc = ostid_to_fid(fid, &oseq->os_oi,
1169 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1173 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1176 ofd_seq_put(tsi->tsi_env, oseq);
1178 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1182 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1189 * OFD request handler for OST_GETATTR RPC.
1191 * This is OFD-specific part of request handling. It finds the OFD object
1192 * by its FID, gets attributes from storage and packs result to the reply.
1194 * \param[in] tsi target session environment for this request
1196 * \retval 0 if successful
1197 * \retval negative value on error
1199 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1201 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1202 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1203 struct ost_body *repbody;
1204 struct lustre_handle lh = { 0 };
1205 struct ofd_object *fo;
1207 enum ldlm_mode lock_mode = LCK_PR;
1212 LASSERT(tsi->tsi_ost_body != NULL);
1214 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1215 if (repbody == NULL)
1218 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1219 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1221 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1222 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1225 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1228 rc = tgt_extent_lock(tsi->tsi_env,
1229 tsi->tsi_tgt->lut_obd->obd_namespace,
1230 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1236 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1238 GOTO(out, rc = PTR_ERR(fo));
1240 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1244 obdo_from_la(&repbody->oa, &fti->fti_attr,
1245 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1247 /* Store object version in reply */
1248 curr_version = dt_version_get(tsi->tsi_env,
1249 ofd_object_child(fo));
1250 if ((__s64)curr_version != -EOPNOTSUPP) {
1251 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1252 repbody->oa.o_data_version = curr_version;
1255 if (fo->ofo_ff.ff_layout_version > 0) {
1256 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1257 repbody->oa.o_layout_version =
1258 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1260 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1261 PFID(&tsi->tsi_fid),
1262 repbody->oa.o_layout_version);
1266 ofd_object_put(tsi->tsi_env, fo);
1269 tgt_extent_unlock(&lh, lock_mode);
1271 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1274 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1275 repbody->oa.o_flags = OBD_FL_FLUSH;
1281 * OFD request handler for OST_SETATTR RPC.
1283 * This is OFD-specific part of request handling. It finds the OFD object
1284 * by its FID, sets attributes from request and packs result to the reply.
1286 * \param[in] tsi target session environment for this request
1288 * \retval 0 if successful
1289 * \retval negative value on error
1291 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1293 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1294 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1295 struct ost_body *body = tsi->tsi_ost_body;
1296 struct ost_body *repbody;
1297 struct ldlm_resource *res;
1298 struct ofd_object *fo;
1303 LASSERT(body != NULL);
1305 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1306 if (repbody == NULL)
1309 repbody->oa.o_oi = body->oa.o_oi;
1310 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1312 /* This would be very bad - accidentally truncating a file when
1313 * changing the time or similar - bug 12203. */
1314 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1315 body->oa.o_size != OBD_OBJECT_EOF) {
1316 static char mdsinum[48];
1318 if (body->oa.o_valid & OBD_MD_FLFID)
1319 snprintf(mdsinum, sizeof(mdsinum) - 1,
1320 "of parent "DFID, body->oa.o_parent_seq,
1321 body->oa.o_parent_oid, 0);
1325 CERROR("%s: setattr from %s is trying to truncate object "DFID
1326 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1327 PFID(&tsi->tsi_fid), mdsinum);
1331 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1333 GOTO(out, rc = PTR_ERR(fo));
1335 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1336 fti->fti_attr.la_valid &= ~LA_TYPE;
1338 /* setting objects attributes (including owner/group) */
1339 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1343 obdo_from_la(&repbody->oa, &fti->fti_attr,
1344 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1346 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1350 ofd_object_put(tsi->tsi_env, fo);
1353 /* we do not call this before to avoid lu_object_find() in
1354 * ->lvbo_update() holding another reference on the object.
1355 * otherwise concurrent destroy can make the object unavailable
1356 * for 2nd lu_object_find() waiting for the first reference
1357 * to go... deadlock! */
1358 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1359 &tsi->tsi_resid, LDLM_EXTENT, 0);
1361 ldlm_res_lvbo_update(res, NULL, 0);
1362 ldlm_resource_putref(res);
1369 * Destroy OST orphans.
1371 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1372 * set then we must destroy possible orphaned objects.
1374 * \param[in] env execution environment
1375 * \param[in] exp OBD export
1376 * \param[in] ofd OFD device
1377 * \param[in] oa obdo structure for reply
1379 * \retval 0 if successful
1380 * \retval negative value on error
1382 static int ofd_orphans_destroy(const struct lu_env *env,
1383 struct obd_export *exp,
1384 struct ofd_device *ofd, struct obdo *oa)
1386 struct ofd_thread_info *info = ofd_info(env);
1387 struct lu_fid *fid = &info->fti_fid;
1388 struct ost_id *oi = &oa->o_oi;
1389 struct ofd_seq *oseq;
1390 u64 seq = ostid_seq(oi);
1391 u64 end_id = ostid_id(oi);
1399 oseq = ofd_seq_get(ofd, seq);
1401 CERROR("%s: Can not find seq for "DOSTID"\n",
1402 ofd_name(ofd), POSTID(oi));
1407 last = ofd_seq_last_oid(oseq);
1410 LASSERT(exp != NULL);
1411 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1413 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1416 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1417 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1419 while (oid > end_id) {
1420 rc = fid_set_id(fid, oid);
1421 if (unlikely(rc != 0))
1424 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1425 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1426 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1427 /* this is pretty fatal... */
1428 CEMERG("%s: error destroying precreated id "
1430 ofd_name(ofd), PFID(fid), rc);
1434 ofd_seq_last_oid_set(oseq, oid);
1435 /* update last_id on disk periodically so that if we
1436 * restart * we don't need to re-scan all of the just
1437 * deleted objects. */
1438 if ((oid & 511) == 0)
1439 ofd_seq_last_oid_write(env, ofd, oseq);
1443 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1444 ofd_name(ofd), seq, oid);
1448 ofd_seq_last_oid_set(oseq, oid);
1449 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1451 /* don't reuse orphan object, return last used objid */
1452 rc = ostid_set_id(oi, last);
1458 ofd_seq_put(env, oseq);
1463 * OFD request handler for OST_CREATE RPC.
1465 * This is OFD-specific part of request handling. Its main purpose is to
1466 * create new data objects on OST, but it also used to destroy orphans.
1468 * \param[in] tsi target session environment for this request
1470 * \retval 0 if successful
1471 * \retval negative value on error
1473 static int ofd_create_hdl(struct tgt_session_info *tsi)
1475 struct ptlrpc_request *req = tgt_ses_req(tsi);
1476 struct ost_body *repbody;
1477 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1478 struct obdo *rep_oa;
1479 struct obd_export *exp = tsi->tsi_exp;
1480 struct ofd_device *ofd = ofd_exp(exp);
1481 u64 seq = ostid_seq(&oa->o_oi);
1482 u64 oid = ostid_id(&oa->o_oi);
1483 struct ofd_seq *oseq;
1491 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1494 if (ofd->ofd_no_precreate)
1497 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1498 if (repbody == NULL)
1501 down_read(&ofd->ofd_lastid_rwsem);
1502 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1503 * we may do that in the future.
1504 * Return -ENOSPC until the LAST_ID rebuilt. */
1505 if (unlikely(ofd->ofd_lastid_rebuilding))
1506 GOTO(out_sem, rc = -ENOSPC);
1508 rep_oa = &repbody->oa;
1509 rep_oa->o_oi = oa->o_oi;
1511 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1513 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1515 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1517 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1518 ofd_name(ofd), seq, PTR_ERR(oseq));
1519 GOTO(out_sem, rc = -EINVAL);
1522 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1523 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1524 if (!ofd_obd(ofd)->obd_recovering ||
1525 oid > ofd_seq_last_oid(oseq)) {
1526 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1527 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1528 ofd_seq_last_oid(oseq));
1529 GOTO(out_nolock, rc = -EINVAL);
1531 /* Do nothing here, we re-create objects during recovery
1532 * upon write replay, see ofd_preprw_write() */
1533 GOTO(out_nolock, rc = 0);
1535 /* former ofd_handle_precreate */
1536 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1537 (oa->o_flags & OBD_FL_DELORPHAN)) {
1538 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1540 /* destroy orphans */
1541 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1542 exp->exp_conn_cnt) {
1543 CERROR("%s: dropping old orphan cleanup request\n",
1545 GOTO(out_nolock, rc = 0);
1547 /* This causes inflight precreates to abort and drop lock */
1548 oseq->os_destroys_in_progress = 1;
1549 mutex_lock(&oseq->os_create_lock);
1550 if (!oseq->os_destroys_in_progress) {
1552 "%s:[%llu] destroys_in_progress already cleared\n",
1553 ofd_name(ofd), seq);
1554 rc = ostid_set_id(&rep_oa->o_oi,
1555 ofd_seq_last_oid(oseq));
1558 diff = oid - ofd_seq_last_oid(oseq);
1559 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1560 ofd_seq_last_oid(oseq), diff);
1561 if (-diff > OST_MAX_PRECREATE) {
1562 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1563 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1564 "(%llu), trust the OST\n",
1565 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1566 PFID(&oseq->os_oi.oi_fid),
1567 ofd_seq_last_oid(oseq));
1569 /* Let MDS know that we are so far ahead. */
1570 rc = ostid_set_id(&rep_oa->o_oi,
1571 ofd_seq_last_oid(oseq) + 1);
1572 } else if (diff < 0) {
1573 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1575 oseq->os_destroys_in_progress = 0;
1577 /* XXX: Used by MDS for the first time! */
1578 oseq->os_destroys_in_progress = 0;
1581 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1582 ofd->ofd_lastid_gen)) {
1583 /* Keep the export ref so we can send the reply. */
1584 ofd_obd_disconnect(class_export_get(exp));
1585 GOTO(out_nolock, rc = -ENOTCONN);
1588 mutex_lock(&oseq->os_create_lock);
1589 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1590 exp->exp_conn_cnt) {
1591 CERROR("%s: dropping old precreate request\n",
1595 /* only precreate if seq is 0, IDIF or normal and also o_id
1596 * must be specfied */
1597 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1598 !fid_seq_is_idif(seq)) || oid == 0) {
1599 diff = 1; /* shouldn't we create this right now? */
1601 diff = oid - ofd_seq_last_oid(oseq);
1602 /* Do sync create if the seq is about to used up */
1603 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1604 if (unlikely(oid >= IDIF_MAX_OID - 1))
1606 } else if (fid_seq_is_norm(seq)) {
1608 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1611 CERROR("%s : invalid o_seq "DOSTID"\n",
1612 ofd_name(ofd), POSTID(&oa->o_oi));
1613 GOTO(out, rc = -EINVAL);
1618 CERROR("%s: invalid precreate request for "
1619 DOSTID", last_id %llu. "
1620 "Likely MDS last_id corruption\n",
1621 ofd_name(ofd), POSTID(&oa->o_oi),
1622 ofd_seq_last_oid(oseq));
1623 GOTO(out, rc = -EINVAL);
1628 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1634 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1635 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1636 /* don't enforce grant during orphan recovery */
1637 granted = tgt_grant_create(tsi->tsi_env,
1638 ofd_obd(ofd)->obd_self_export,
1643 CDEBUG(D_HA, "%s: failed to acquire grant "
1644 "space for precreate (%lld): rc = %d\n",
1645 ofd_name(ofd), diff, rc);
1650 /* This can happen if a new OST is formatted and installed
1651 * in place of an old one at the same index. Instead of
1652 * precreating potentially millions of deleted old objects
1653 * (possibly filling the OST), only precreate the last batch.
1654 * LFSCK will eventually clean up any orphans. LU-14 */
1655 if (diff > 5 * OST_MAX_PRECREATE) {
1656 diff = OST_MAX_PRECREATE / 2;
1657 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1658 "OST replaced or reformatted: "
1659 "LFSCK will clean up",
1662 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1663 "%u larger than the LAST_ID "DOSTID", only "
1664 "precreating the last %lld objects.\n",
1665 ofd_name(ofd), POSTID(&oa->o_oi),
1666 5 * OST_MAX_PRECREATE,
1667 POSTID(&oseq->os_oi), diff);
1668 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1672 next_id = ofd_seq_last_oid(oseq) + 1;
1673 count = ofd_precreate_batch(ofd, (int)diff);
1675 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1676 " at %llu\n", ofd_name(ofd),
1677 count, seq, next_id);
1679 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1680 && ktime_get_seconds() > enough_time) {
1681 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1682 " created at a rate of %d/s\n",
1683 ofd_name(ofd), created, diff + created,
1684 created / DISK_TIMEOUT);
1688 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1689 oseq, count, sync_trans);
1693 } else if (rc < 0) {
1699 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1700 LCONSOLE_WARN("%s: can't create the same count of"
1701 " objects when replaying the request"
1702 " (diff is %lld). see LU-4621\n",
1703 ofd_name(ofd), diff);
1706 /* some objects got created, we can return
1707 * them, even if last creation failed */
1710 CERROR("%s: unable to precreate: rc = %d\n",
1713 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1714 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1715 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1720 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1724 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1726 if (unlikely(!oseq->os_last_id_synced))
1727 oseq->os_last_id_synced = 1;
1729 mutex_unlock(&oseq->os_create_lock);
1732 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1733 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1734 struct lu_fid *fid = &info->fti_fid;
1736 /* For compatible purpose, it needs to convert back to
1737 * OST ID before put it on wire. */
1738 *fid = rep_oa->o_oi.oi_fid;
1739 fid_to_ostid(fid, &rep_oa->o_oi);
1741 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1743 ofd_seq_put(tsi->tsi_env, oseq);
1746 up_read(&ofd->ofd_lastid_rwsem);
1751 * OFD request handler for OST_DESTROY RPC.
1753 * This is OFD-specific part of request handling. It destroys data objects
1754 * related to destroyed object on MDT.
1756 * \param[in] tsi target session environment for this request
1758 * \retval 0 if successful
1759 * \retval negative value on error
1761 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1763 const struct ost_body *body = tsi->tsi_ost_body;
1764 struct ost_body *repbody;
1765 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1766 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1767 struct lu_fid *fid = &fti->fti_fid;
1774 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1777 /* This is old case for clients before Lustre 2.4 */
1778 /* If there's a DLM request, cancel the locks mentioned in it */
1779 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1781 struct ldlm_request *dlm;
1783 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1786 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1789 *fid = body->oa.o_oi.oi_fid;
1790 oid = ostid_id(&body->oa.o_oi);
1793 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1795 /* check that o_misc makes sense */
1796 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1797 count = body->oa.o_misc;
1799 count = 1; /* default case - single destroy */
1801 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1802 POSTID(&body->oa.o_oi), count);
1807 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1808 if (lrc == -ENOENT) {
1810 "%s: destroying non-existent object "DFID"\n",
1811 ofd_name(ofd), PFID(fid));
1812 /* rewrite rc with -ENOENT only if it is 0 */
1815 } else if (lrc != 0) {
1816 CERROR("%s: error destroying object "DFID": %d\n",
1817 ofd_name(ofd), PFID(fid), lrc);
1823 lrc = fid_set_id(fid, oid);
1824 if (unlikely(lrc != 0 && count > 0))
1825 GOTO(out, rc = lrc);
1828 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1834 fid_to_ostid(fid, &repbody->oa.o_oi);
1839 * OFD request handler for OST_STATFS RPC.
1841 * This function gets statfs data from storage as part of request
1844 * \param[in] tsi target session environment for this request
1846 * \retval 0 if successful
1847 * \retval negative value on error
1849 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1851 struct obd_statfs *osfs;
1856 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1858 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1860 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1861 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1863 CERROR("%s: statfs failed: rc = %d\n",
1864 tgt_name(tsi->tsi_tgt), rc);
1866 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1869 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1876 * OFD request handler for OST_SYNC RPC.
1878 * Sync object data or all filesystem data to the disk and pack the
1881 * \param[in] tsi target session environment for this request
1883 * \retval 0 if successful
1884 * \retval negative value on error
1886 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1888 struct ost_body *body = tsi->tsi_ost_body;
1889 struct ost_body *repbody;
1890 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1891 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1892 struct ofd_object *fo = NULL;
1897 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1899 /* if no objid is specified, it means "sync whole filesystem" */
1900 if (!fid_is_zero(&tsi->tsi_fid)) {
1901 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1903 RETURN(PTR_ERR(fo));
1906 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1907 fo != NULL ? ofd_object_child(fo) : NULL,
1908 repbody->oa.o_size, repbody->oa.o_blocks);
1912 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1917 repbody->oa.o_oi = body->oa.o_oi;
1918 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1920 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1922 obdo_from_la(&repbody->oa, &fti->fti_attr,
1925 /* don't return rc from getattr */
1930 ofd_object_put(tsi->tsi_env, fo);
1935 * OFD request handler for OST_PUNCH RPC.
1937 * This is part of request processing. Validate request fields,
1938 * punch (truncate) the given OFD object and pack reply.
1940 * \param[in] tsi target session environment for this request
1942 * \retval 0 if successful
1943 * \retval negative value on error
1945 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1947 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1948 struct ost_body *repbody;
1949 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1950 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1951 struct ldlm_resource *res;
1952 struct ofd_object *fo;
1954 struct lustre_handle lh = { 0, };
1961 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
1963 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1964 BUILD_BUG_ON(!(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK));
1966 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1967 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1968 RETURN(err_serious(-EPROTO));
1970 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1971 if (repbody == NULL)
1972 RETURN(err_serious(-ENOMEM));
1974 /* punch start,end are passed in o_size,o_blocks throught wire */
1978 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1981 /* standard truncate optimization: if file body is completely
1982 * destroyed, don't send data back to the server. */
1984 flags |= LDLM_FL_AST_DISCARD_DATA;
1986 repbody->oa.o_oi = oa->o_oi;
1987 repbody->oa.o_valid = OBD_MD_FLID;
1989 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1990 oa->o_flags & OBD_FL_SRVLOCK;
1993 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
1994 end, &lh, LCK_PW, &flags);
1999 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2000 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2001 oa->o_valid, start, end);
2003 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2006 GOTO(out, rc = PTR_ERR(fo));
2008 la_from_obdo(&info->fti_attr, oa,
2009 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2010 info->fti_attr.la_size = start;
2011 info->fti_attr.la_valid |= LA_SIZE;
2013 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2018 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2022 ofd_object_put(tsi->tsi_env, fo);
2025 tgt_extent_unlock(&lh, LCK_PW);
2027 /* we do not call this before to avoid lu_object_find() in
2028 * ->lvbo_update() holding another reference on the object.
2029 * otherwise concurrent destroy can make the object unavailable
2030 * for 2nd lu_object_find() waiting for the first reference
2031 * to go... deadlock! */
2032 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2035 struct ost_lvb *res_lvb;
2037 ldlm_res_lvbo_update(res, NULL, 0);
2038 res_lvb = res->lr_lvb_data;
2039 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2040 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2041 ldlm_resource_putref(res);
2047 static int ofd_ladvise_prefetch(const struct lu_env *env,
2048 struct ofd_object *fo,
2049 struct niobuf_local *lnb,
2050 __u64 start, __u64 end, enum dt_bufs_type dbt)
2052 struct ofd_thread_info *info = ofd_info(env);
2053 pgoff_t start_index, end_index, pages;
2054 struct niobuf_remote rnb;
2055 unsigned long nr_local;
2061 ofd_read_lock(env, fo);
2062 if (!ofd_object_exists(fo))
2063 GOTO(out_unlock, rc = -ENOENT);
2065 rc = ofd_attr_get(env, fo, &info->fti_attr);
2067 GOTO(out_unlock, rc);
2069 if (end > info->fti_attr.la_size)
2070 end = info->fti_attr.la_size;
2073 GOTO(out_unlock, rc);
2075 /* We need page aligned offset and length */
2076 start_index = start >> PAGE_SHIFT;
2077 end_index = (end - 1) >> PAGE_SHIFT;
2078 pages = end_index - start_index + 1;
2080 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2081 PTLRPC_MAX_BRW_PAGES;
2082 rnb.rnb_offset = start_index << PAGE_SHIFT;
2083 rnb.rnb_len = nr_local << PAGE_SHIFT;
2084 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb,
2085 PTLRPC_MAX_BRW_PAGES, dbt);
2086 if (unlikely(rc < 0))
2089 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2090 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2093 start_index += nr_local;
2098 ofd_read_unlock(env, fo);
2103 * OFD request handler for OST_LADVISE RPC.
2105 * Tune cache or perfetch policies according to advices.
2107 * \param[in] tsi target session environment for this request
2109 * \retval 0 if successful
2110 * \retval negative errno on error
2112 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2114 struct ptlrpc_request *req = tgt_ses_req(tsi);
2115 struct obd_export *exp = tsi->tsi_exp;
2116 struct ofd_device *ofd = ofd_exp(exp);
2117 struct ost_body *body, *repbody;
2118 struct ofd_thread_info *info;
2119 struct ofd_object *fo;
2120 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2121 const struct lu_env *env = svc_thread->t_env;
2122 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2123 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2124 struct lu_ladvise *ladvise;
2126 struct ladvise_hdr *ladvise_hdr;
2127 struct obd_ioobj ioo;
2128 struct lustre_handle lockh = { 0 };
2131 struct dt_object *dob;
2137 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2138 body = tsi->tsi_ost_body;
2140 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2141 RETURN(err_serious(-EPROTO));
2143 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2144 &RMF_OST_LADVISE_HDR);
2145 if (ladvise_hdr == NULL)
2146 RETURN(err_serious(-EPROTO));
2148 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2149 ladvise_hdr->lah_count < 1)
2150 RETURN(err_serious(-EPROTO));
2152 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2153 RETURN(err_serious(-EPROTO));
2155 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2156 if (ladvise == NULL)
2157 RETURN(err_serious(-EPROTO));
2159 num_advise = req_capsule_get_size(&req->rq_pill,
2160 &RMF_OST_LADVISE, RCL_CLIENT) /
2162 if (num_advise < ladvise_hdr->lah_count)
2163 RETURN(err_serious(-EPROTO));
2165 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2166 repbody->oa = body->oa;
2168 info = ofd_info_init(env, exp);
2170 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2171 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2175 fo = ofd_object_find(env, ofd, &info->fti_fid);
2180 LASSERT(fo != NULL);
2181 dob = ofd_object_child(fo);
2183 if (ptlrpc_connection_is_local(exp->exp_connection))
2184 dbt |= DT_BUFS_TYPE_LOCAL;
2186 for (i = 0; i < num_advise; i++, ladvise++) {
2187 start = ladvise->lla_start;
2188 end = ladvise->lla_end;
2190 rc = err_serious(-EPROTO);
2194 /* Handle different advice types */
2195 switch (ladvise->lla_advice) {
2199 case LU_LADVISE_WILLREAD:
2203 ioo.ioo_oid = body->oa.o_oi;
2205 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2206 &tsi->tsi_resid, start, end - 1,
2207 &lockh, LCK_PR, &flags);
2211 req->rq_status = ofd_ladvise_prefetch(env, fo,
2214 tgt_extent_unlock(&lockh, LCK_PR);
2216 case LU_LADVISE_DONTNEED:
2217 rc = dt_ladvise(env, dob, ladvise->lla_start,
2218 ladvise->lla_end, LU_LADVISE_DONTNEED);
2225 ofd_object_put(env, fo);
2226 req->rq_status = rc;
2231 * OFD request handler for OST_QUOTACTL RPC.
2233 * This is part of request processing to validate incoming request fields,
2234 * get the requested data from OSD and pack reply.
2236 * \param[in] tsi target session environment for this request
2238 * \retval 0 if successful
2239 * \retval negative value on error
2241 static int ofd_quotactl(struct tgt_session_info *tsi)
2243 struct obd_quotactl *oqctl, *repoqc;
2244 struct lu_nodemap *nodemap;
2250 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2252 RETURN(err_serious(-EPROTO));
2254 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2256 RETURN(err_serious(-ENOMEM));
2260 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2261 if (IS_ERR(nodemap))
2262 RETURN(PTR_ERR(nodemap));
2265 if (oqctl->qc_type == USRQUOTA)
2266 id = nodemap_map_id(nodemap, NODEMAP_UID,
2267 NODEMAP_CLIENT_TO_FS,
2269 else if (oqctl->qc_type == GRPQUOTA)
2270 id = nodemap_map_id(nodemap, NODEMAP_GID,
2271 NODEMAP_CLIENT_TO_FS,
2274 nodemap_putref(nodemap);
2276 if (repoqc->qc_id != id)
2277 swap(repoqc->qc_id, id);
2279 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2281 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2284 if (repoqc->qc_id != id)
2285 swap(repoqc->qc_id, id);
2291 * Calculate the amount of time for lock prolongation.
2293 * This is helper for ofd_prolong_extent_locks() function to get
2294 * the timeout extra time.
2296 * \param[in] req current request
2298 * \retval amount of time to extend the timeout with
2300 static inline time64_t prolong_timeout(struct ptlrpc_request *req)
2302 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2303 time64_t req_timeout;
2306 return obd_timeout / 2;
2308 req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec;
2309 return max_t(time64_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
2314 * Prolong lock timeout for the given extent.
2316 * This function finds all locks related with incoming request and
2317 * prolongs their timeout.
2319 * If a client is holding a lock for a long time while it sends
2320 * read or write RPCs to the OST for the object under this lock,
2321 * then we don't want the OST to evict the client. Otherwise,
2322 * if the network or disk is very busy then the client may not
2323 * be able to make any progress to clear out dirty pages under
2324 * the lock and the application will fail.
2326 * Every time a Bulk Read/Write (BRW) request arrives for the object
2327 * covered by the lock, extend the timeout on that lock. The RPC should
2328 * contain a lock handle for the lock it is using, but this
2329 * isn't handled correctly by all client versions, and the
2330 * request may cover multiple locks.
2332 * \param[in] tsi target session environment for this request
2333 * \param[in] data struct of data to prolong locks
2336 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2337 struct ldlm_prolong_args *data)
2339 struct obdo *oa = &tsi->tsi_ost_body->oa;
2340 struct ldlm_lock *lock;
2344 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2345 data->lpa_export = tsi->tsi_exp;
2346 data->lpa_resid = tsi->tsi_resid;
2348 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2349 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2350 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2351 data->lpa_extent.end);
2353 if (oa->o_valid & OBD_MD_FLHANDLE) {
2354 /* mostly a request should be covered by only one lock, try
2356 lock = ldlm_handle2lock(&oa->o_handle);
2358 /* Fast path to check if the lock covers the whole IO
2359 * region exclusively. */
2360 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2361 &data->lpa_extent)) {
2363 LASSERT(lock->l_export == data->lpa_export);
2364 ldlm_lock_prolong_one(lock, data);
2365 LDLM_LOCK_PUT(lock);
2366 if (data->lpa_locks_cnt > 0)
2368 /* The lock was destroyed probably lets try
2371 lock->l_last_used = ktime_get();
2372 LDLM_LOCK_PUT(lock);
2377 ldlm_resource_prolong(data);
2382 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2384 * Determine if \a lock and the lock from request \a req are equivalent
2385 * by comparing their resource names, modes, and extents.
2387 * It is used to give priority to read and write RPCs being done
2388 * under this lock so that the client can drop the contended
2389 * lock more quickly and let other clients use it. This improves
2390 * overall performance in the case where the first client gets a
2391 * very large lock extent that prevents other clients from
2392 * submitting their writes.
2394 * \param[in] req ptlrpc_request being processed
2395 * \param[in] lock contended lock to match
2397 * \retval 1 if lock is matched
2398 * \retval 0 otherwise
2400 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2401 struct ldlm_lock *lock)
2403 struct niobuf_remote *rnb;
2404 struct obd_ioobj *ioo;
2405 enum ldlm_mode mode;
2406 struct ldlm_extent ext;
2407 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2411 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2412 LASSERT(ioo != NULL);
2414 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2415 LASSERT(rnb != NULL);
2417 ext.start = rnb->rnb_offset;
2418 rnb += ioo->ioo_bufcnt - 1;
2419 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2421 LASSERT(lock->l_resource != NULL);
2422 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2425 /* a bulk write can only hold a reference on a PW extent lock
2428 mode = LCK_PW | LCK_GROUP;
2429 if (opc == OST_READ)
2430 /* whereas a bulk read can be protected by either a PR or PW
2434 if (!(lock->l_granted_mode & mode))
2437 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2441 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2443 * Check for whether the given PTLRPC request (\a req) is blocking
2444 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2447 * \param[in] req the incoming request
2449 * \retval 1 if \a req is blocking an LDLM lock cancel
2450 * \retval 0 if it is not
2451 * \retval -ESTALE if lock is not found
2453 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2455 struct tgt_session_info *tsi;
2456 struct obd_ioobj *ioo;
2457 struct niobuf_remote *rnb;
2459 struct ldlm_prolong_args pa = { 0 };
2463 /* Don't use tgt_ses_info() to get session info, because lock_match()
2464 * can be called while request has no processing thread yet. */
2465 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2468 * Use LASSERT below because malformed RPCs should have
2469 * been filtered out in tgt_hpreq_handler().
2471 opc = lustre_msg_get_opc(req->rq_reqmsg);
2472 LASSERT(opc == OST_READ || opc == OST_WRITE);
2474 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2475 LASSERT(ioo != NULL);
2477 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2478 LASSERT(rnb != NULL);
2479 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2481 pa.lpa_mode = LCK_PW | LCK_GROUP;
2482 if (opc == OST_READ)
2483 pa.lpa_mode |= LCK_PR;
2485 pa.lpa_extent.start = rnb->rnb_offset;
2486 rnb += ioo->ioo_bufcnt - 1;
2487 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2489 DEBUG_REQ(D_RPCTRACE, req,
2490 "%s %s: refresh rw locks for "DFID" (%llu->%llu)",
2491 tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid),
2492 pa.lpa_extent.start, pa.lpa_extent.end);
2494 ofd_prolong_extent_locks(tsi, &pa);
2496 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n",
2497 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2499 if (pa.lpa_blocks_cnt > 0)
2502 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2506 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2508 * Called after the request has been handled. It refreshes lock timeout again
2509 * so that client has more time to send lock cancel RPC.
2511 * \param[in] req request which is being processed.
2513 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2515 ofd_rw_hpreq_check(req);
2519 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2521 * This function checks if the given lock is the same by its resname, mode
2522 * and extent as one taken from the request.
2523 * It is used to give priority to punch/truncate RPCs that might lead to
2524 * the fastest release of that lock when a lock is contended.
2526 * \param[in] req ptlrpc_request being processed
2527 * \param[in] lock contended lock to match
2529 * \retval 1 if lock is matched
2530 * \retval 0 otherwise
2532 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2533 struct ldlm_lock *lock)
2535 struct tgt_session_info *tsi;
2537 struct ldlm_extent ext;
2541 /* Don't use tgt_ses_info() to get session info, because lock_match()
2542 * can be called while request has no processing thread yet. */
2543 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2546 * Use LASSERT below because malformed RPCs should have
2547 * been filtered out in tgt_hpreq_handler().
2549 LASSERT(tsi->tsi_ost_body != NULL);
2550 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2551 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2554 oa = &tsi->tsi_ost_body->oa;
2555 ext.start = oa->o_size;
2556 ext.end = oa->o_blocks;
2558 LASSERT(lock->l_resource != NULL);
2559 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2562 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2565 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2569 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2571 * High-priority queue request check for whether the given punch request
2572 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2573 * covered by an LDLM lock.
2577 * \param[in] req the incoming request
2579 * \retval 1 if \a req is blocking an LDLM lock cancel
2580 * \retval 0 if it is not
2581 * \retval -ESTALE if lock is not found
2583 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2585 struct tgt_session_info *tsi;
2587 struct ldlm_prolong_args pa = { 0 };
2591 /* Don't use tgt_ses_info() to get session info, because lock_match()
2592 * can be called while request has no processing thread yet. */
2593 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2594 LASSERT(tsi != NULL);
2595 oa = &tsi->tsi_ost_body->oa;
2597 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2598 oa->o_flags & OBD_FL_SRVLOCK));
2600 pa.lpa_mode = LCK_PW | LCK_GROUP;
2601 pa.lpa_extent.start = oa->o_size;
2602 pa.lpa_extent.end = oa->o_blocks;
2605 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2606 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2607 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2609 ofd_prolong_extent_locks(tsi, &pa);
2611 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2612 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2614 if (pa.lpa_blocks_cnt > 0)
2617 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2621 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2623 * Called after the request has been handled. It refreshes lock timeout again
2624 * so that client has more time to send lock cancel RPC.
2626 * \param[in] req request which is being processed.
2628 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2630 ofd_punch_hpreq_check(req);
2633 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2634 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2635 .hpreq_check = ofd_rw_hpreq_check,
2636 .hpreq_fini = ofd_rw_hpreq_fini
2639 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2640 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2641 .hpreq_check = ofd_punch_hpreq_check,
2642 .hpreq_fini = ofd_punch_hpreq_fini
2646 * Assign high priority operations to an IO request.
2648 * Check if the incoming request is a candidate for
2649 * high-priority processing. If it is, assign it a high
2650 * priority operations table.
2652 * \param[in] tsi target session environment for this request
2654 static void ofd_hp_brw(struct tgt_session_info *tsi)
2656 struct niobuf_remote *rnb;
2657 struct obd_ioobj *ioo;
2661 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2662 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2663 if (ioo->ioo_bufcnt > 0) {
2664 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2665 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2667 /* no high priority if server lock is needed */
2668 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2669 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2673 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2677 * Assign high priority operations to an punch request.
2679 * Check if the incoming request is a candidate for
2680 * high-priority processing. If it is, assign it a high
2681 * priority operations table.
2683 * \param[in] tsi target session environment for this request
2685 static void ofd_hp_punch(struct tgt_session_info *tsi)
2687 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2688 /* no high-priority if server lock is needed */
2689 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2690 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2691 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2692 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2694 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2697 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2698 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2699 #define OST_BRW_READ OST_READ
2700 #define OST_BRW_WRITE OST_WRITE
2703 * Table of OFD-specific request handlers
2705 * This table contains all opcodes accepted by OFD and
2706 * specifies handlers for them. The tgt_request_handler()
2707 * uses such table from each target to process incoming
2710 static struct tgt_handler ofd_tgt_handlers[] = {
2711 TGT_RPC_HANDLER(OST_FIRST_OPC,
2712 0, OST_CONNECT, tgt_connect,
2713 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2714 TGT_RPC_HANDLER(OST_FIRST_OPC,
2715 0, OST_DISCONNECT, tgt_disconnect,
2716 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2717 TGT_RPC_HANDLER(OST_FIRST_OPC,
2718 0, OST_SET_INFO, ofd_set_info_hdl,
2719 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2720 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2721 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl),
2722 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2723 OST_SETATTR, ofd_setattr_hdl),
2724 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2725 OST_CREATE, ofd_create_hdl),
2726 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2727 OST_DESTROY, ofd_destroy_hdl),
2728 TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl),
2729 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
2731 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2732 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write,
2734 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2735 OST_PUNCH, ofd_punch_hdl,
2737 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl),
2738 TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl),
2739 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl),
2742 static struct tgt_opc_slice ofd_common_slice[] = {
2744 .tos_opc_start = OST_FIRST_OPC,
2745 .tos_opc_end = OST_LAST_OPC,
2746 .tos_hs = ofd_tgt_handlers
2749 .tos_opc_start = OBD_FIRST_OPC,
2750 .tos_opc_end = OBD_LAST_OPC,
2751 .tos_hs = tgt_obd_handlers
2754 .tos_opc_start = LDLM_FIRST_OPC,
2755 .tos_opc_end = LDLM_LAST_OPC,
2756 .tos_hs = tgt_dlm_handlers
2759 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2760 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2761 .tos_hs = tgt_out_handlers
2764 .tos_opc_start = SEQ_FIRST_OPC,
2765 .tos_opc_end = SEQ_LAST_OPC,
2766 .tos_hs = seq_handlers
2769 .tos_opc_start = LFSCK_FIRST_OPC,
2770 .tos_opc_end = LFSCK_LAST_OPC,
2771 .tos_hs = tgt_lfsck_handlers
2774 .tos_opc_start = SEC_FIRST_OPC,
2775 .tos_opc_end = SEC_LAST_OPC,
2776 .tos_hs = tgt_sec_ctx_handlers
2783 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2784 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2787 * Implementation of lu_context_key::lct_key_exit.
2789 * Optional method called on lu_context_exit() for all allocated
2791 * It is used in OFD to sanitize context values which may be re-used
2792 * during another request processing by the same thread.
2794 * \param[in] ctx execution context
2795 * \param[in] key context key
2796 * \param[in] data ofd_thread_info
2798 static void ofd_key_exit(const struct lu_context *ctx,
2799 struct lu_context_key *key, void *data)
2801 struct ofd_thread_info *info = data;
2803 info->fti_env = NULL;
2804 info->fti_exp = NULL;
2807 info->fti_pre_version = 0;
2809 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2812 struct lu_context_key ofd_thread_key = {
2813 .lct_tags = LCT_DT_THREAD,
2814 .lct_init = ofd_key_init,
2815 .lct_fini = ofd_key_fini,
2816 .lct_exit = ofd_key_exit
2820 * Initialize OFD device according to parameters in the config log \a cfg.
2822 * This is the main starting point of OFD initialization. It fills all OFD
2823 * parameters with their initial values and calls other initializing functions
2824 * to set up all OFD subsystems.
2826 * \param[in] env execution environment
2827 * \param[in] m OFD device
2828 * \param[in] ldt LU device type of OFD
2829 * \param[in] cfg configuration log
2831 * \retval 0 if successful
2832 * \retval negative value on error
2834 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2835 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2837 const char *dev = lustre_cfg_string(cfg, 0);
2838 struct ofd_thread_info *info = NULL;
2839 struct obd_device *obd;
2840 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2842 struct nm_config_file *nodemap_config;
2843 struct obd_device_target *obt;
2848 obd = class_name2obd(dev);
2850 CERROR("Cannot find obd with name %s\n", dev);
2854 rc = lu_env_refill((struct lu_env *)env);
2859 obt->obt_magic = OBT_MAGIC;
2861 spin_lock_init(&m->ofd_flags_lock);
2862 m->ofd_raid_degraded = 0;
2863 m->ofd_checksum_t10pi_enforce = 0;
2864 m->ofd_sync_journal = 0;
2866 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2868 m->ofd_seq_count = 0;
2869 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2870 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2871 spin_lock_init(&m->ofd_inconsistency_lock);
2873 spin_lock_init(&m->ofd_batch_lock);
2874 init_rwsem(&m->ofd_lastid_rwsem);
2876 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2877 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2878 /* set this lu_device to obd, because error handling need it */
2879 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2881 /* No connection accepted until configurations will finish */
2882 spin_lock(&obd->obd_dev_lock);
2883 obd->obd_no_conn = 1;
2884 spin_unlock(&obd->obd_dev_lock);
2885 obd->obd_replayable = 1;
2886 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2887 char *str = lustre_cfg_string(cfg, 4);
2889 if (strchr(str, 'n')) {
2890 CWARN("%s: recovery disabled\n", obd->obd_name);
2891 obd->obd_replayable = 0;
2895 info = ofd_info_init(env, NULL);
2899 rc = ofd_stack_init(env, m, cfg);
2901 CERROR("%s: can't init device stack, rc %d\n",
2906 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
2907 ofd_procfs_add_brw_stats_symlink(m);
2910 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2911 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2912 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2913 LDLM_NAMESPACE_SERVER,
2914 LDLM_NAMESPACE_GREEDY,
2916 if (m->ofd_namespace == NULL)
2917 GOTO(err_fini_stack, rc = -ENOMEM);
2918 /* set obd_namespace for compatibility with old code */
2919 obd->obd_namespace = m->ofd_namespace;
2920 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2921 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2922 m->ofd_namespace->ns_lvbp = m;
2924 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2925 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2927 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2928 OBD_FAIL_OST_ALL_REQUEST_NET,
2929 OBD_FAIL_OST_ALL_REPLY_NET);
2931 GOTO(err_free_ns, rc);
2933 rc = ofd_tunables_init(m);
2935 GOTO(err_fini_lut, rc);
2937 tgd->tgd_reserved_pcnt = 0;
2939 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
2940 m->ofd_cksum_types_supported =
2941 obd_cksum_types_supported_server(obd->obd_name);
2942 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2943 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
2944 OFD_PRECREATE_SMALL_FS)
2945 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2947 rc = ofd_fs_setup(env, m, obd);
2949 GOTO(err_fini_proc, rc);
2951 fid.f_seq = FID_SEQ_LOCAL_NAME;
2954 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
2957 GOTO(err_fini_fs, rc);
2959 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
2961 if (IS_ERR(nodemap_config)) {
2962 rc = PTR_ERR(nodemap_config);
2964 GOTO(err_fini_los, rc);
2966 obt->obt_nodemap_config_file = nodemap_config;
2969 rc = ofd_start_inconsistency_verification_thread(m);
2971 GOTO(err_fini_nm, rc);
2973 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
2978 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
2979 obt->obt_nodemap_config_file = NULL;
2981 local_oid_storage_fini(env, m->ofd_los);
2984 ofd_fs_cleanup(env, m);
2988 tgt_fini(env, &m->ofd_lut);
2990 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
2991 obd->obd_namespace = m->ofd_namespace = NULL;
2993 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
2998 * Stop the OFD device
3000 * This function stops the OFD device and all its subsystems.
3001 * This is the end of OFD lifecycle.
3003 * \param[in] env execution environment
3004 * \param[in] m OFD device
3006 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3008 struct obd_device *obd = ofd_obd(m);
3009 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3010 struct lfsck_stop stop;
3012 stop.ls_status = LS_PAUSED;
3014 lfsck_stop(env, m->ofd_osd, &stop);
3015 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3016 target_recovery_fini(obd);
3017 if (m->ofd_namespace != NULL)
3018 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3019 d->ld_obd->obd_force);
3021 obd_exports_barrier(obd);
3022 obd_zombie_barrier();
3025 tgt_fini(env, &m->ofd_lut);
3026 ofd_stop_inconsistency_verification_thread(m);
3027 lfsck_degister(env, m->ofd_osd);
3028 ofd_fs_cleanup(env, m);
3029 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3030 obd->u.obt.obt_nodemap_config_file = NULL;
3032 if (m->ofd_namespace != NULL) {
3033 ldlm_namespace_free_post(m->ofd_namespace);
3034 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3037 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3039 LASSERT(atomic_read(&d->ld_ref) == 0);
3040 server_put_mount(obd->obd_name, true);
3045 * Implementation of lu_device_type_operations::ldto_device_fini.
3047 * Finalize device. Dual to ofd_device_init(). It is called from
3048 * obd_precleanup() and stops the current device.
3050 * \param[in] env execution environment
3051 * \param[in] d LU device of OFD
3055 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3056 struct lu_device *d)
3059 ofd_fini(env, ofd_dev(d));
3064 * Implementation of lu_device_type_operations::ldto_device_free.
3066 * Free OFD device. Dual to ofd_device_alloc().
3068 * \param[in] env execution environment
3069 * \param[in] d LU device of OFD
3073 static struct lu_device *ofd_device_free(const struct lu_env *env,
3074 struct lu_device *d)
3076 struct ofd_device *m = ofd_dev(d);
3078 dt_device_fini(&m->ofd_dt_dev);
3084 * Implementation of lu_device_type_operations::ldto_device_alloc.
3086 * This function allocates the new OFD device. It is called from
3087 * obd_setup() if OBD device had lu_device_type defined.
3089 * \param[in] env execution environment
3090 * \param[in] t lu_device_type of OFD device
3091 * \param[in] cfg configuration log
3093 * \retval pointer to the lu_device of just allocated OFD
3094 * \retval ERR_PTR of return value on error
3096 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3097 struct lu_device_type *t,
3098 struct lustre_cfg *cfg)
3100 struct ofd_device *m;
3101 struct lu_device *l;
3106 return ERR_PTR(-ENOMEM);
3108 l = &m->ofd_dt_dev.dd_lu_dev;
3109 dt_device_init(&m->ofd_dt_dev, t);
3110 rc = ofd_init0(env, m, t, cfg);
3112 ofd_device_free(env, l);
3119 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3120 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3122 static struct lu_device_type_operations ofd_device_type_ops = {
3123 .ldto_init = ofd_type_init,
3124 .ldto_fini = ofd_type_fini,
3126 .ldto_start = ofd_type_start,
3127 .ldto_stop = ofd_type_stop,
3129 .ldto_device_alloc = ofd_device_alloc,
3130 .ldto_device_free = ofd_device_free,
3131 .ldto_device_fini = ofd_device_fini
3134 static struct lu_device_type ofd_device_type = {
3135 .ldt_tags = LU_DEVICE_DT,
3136 .ldt_name = LUSTRE_OST_NAME,
3137 .ldt_ops = &ofd_device_type_ops,
3138 .ldt_ctx_tags = LCT_DT_THREAD
3142 * Initialize OFD module.
3144 * This function is called upon module loading. It registers OFD device type
3145 * and prepares all in-memory structures used by all OFD devices.
3147 * \retval 0 if successful
3148 * \retval negative value on error
3150 static int __init ofd_init(void)
3154 rc = lu_kmem_init(ofd_caches);
3157 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3158 LUSTRE_OST_NAME, &ofd_device_type);
3165 * This function is called upon OFD module unloading.
3166 * It frees all related structures and unregisters OFD device type.
3168 static void __exit ofd_exit(void)
3170 lu_kmem_fini(ofd_caches);
3171 class_unregister_type(LUSTRE_OST_NAME);
3174 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3175 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3176 MODULE_VERSION(LUSTRE_VERSION_STRING);
3177 MODULE_LICENSE("GPL");
3179 module_init(ofd_init);
3180 module_exit(ofd_exit);