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);
458 static void ofd_object_free_rcu(struct rcu_head *head)
460 struct ofd_object *of = container_of(head, struct ofd_object,
463 kmem_cache_free(ofd_object_kmem, of);
467 * Implementation of lu_object_operations::loo_object_free.
469 * Finish OFD object lifecycle and free its memory.
471 * \param[in] env execution environment
472 * \param[in] o LU object of OFD object
474 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
476 struct ofd_object *of = ofd_obj(o);
477 struct lu_object_header *h;
482 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
483 PFID(lu_object_fid(o)));
486 lu_object_header_fini(h);
487 OBD_FREE_PRE(of, sizeof(*of), "slab-freed");
488 call_rcu(&of->ofo_header.loh_rcu, ofd_object_free_rcu);
493 * Implementation of lu_object_operations::loo_object_print.
495 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
496 * LU_OBJECT_DEBUG() for more details about the compound object printing.
498 * \param[in] env execution environment
499 * \param[in] cookie opaque data passed to the printer function
500 * \param[in] p printer function to use
501 * \param[in] o LU object of OFD object
503 * \retval 0 if successful
504 * \retval negative value on error
506 static int ofd_object_print(const struct lu_env *env, void *cookie,
507 lu_printer_t p, const struct lu_object *o)
509 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
512 static struct lu_object_operations ofd_obj_ops = {
513 .loo_object_init = ofd_object_init,
514 .loo_object_free = ofd_object_free,
515 .loo_object_print = ofd_object_print
519 * Implementation of lu_device_operations::lod_object_alloc.
521 * This function allocates OFD part of compound OFD-OSD object and
522 * initializes its header, because OFD is the top device in stack
524 * \param[in] env execution environment
525 * \param[in] hdr object header, NULL for OFD
526 * \param[in] d lu_device
528 * \retval allocated object if successful
529 * \retval NULL value on failed allocation
531 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
532 const struct lu_object_header *hdr,
535 struct ofd_object *of;
539 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
542 struct lu_object_header *h;
544 o = &of->ofo_obj.do_lu;
546 lu_object_header_init(h);
547 lu_object_init(o, h, d);
548 lu_object_add_top(h, o);
549 o->lo_ops = &ofd_obj_ops;
557 * Return the result of LFSCK run to the OFD.
559 * Notify OFD about result of LFSCK run. That may block the new object
560 * creation until problem is fixed by LFSCK.
562 * \param[in] env execution environment
563 * \param[in] data pointer to the OFD device
564 * \param[in] event LFSCK event type
566 * \retval 0 if successful
567 * \retval negative value on unknown event
569 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
570 enum lfsck_events event)
572 struct ofd_device *ofd = data;
573 struct obd_device *obd = ofd_obd(ofd);
576 case LE_LASTID_REBUILDING:
577 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
578 "on the device until the LAST_ID rebuilt successfully.\n",
580 down_write(&ofd->ofd_lastid_rwsem);
581 ofd->ofd_lastid_rebuilding = 1;
582 up_write(&ofd->ofd_lastid_rwsem);
584 case LE_LASTID_REBUILT: {
585 down_write(&ofd->ofd_lastid_rwsem);
586 ofd_seqs_free(env, ofd);
587 ofd->ofd_lastid_rebuilding = 0;
588 ofd->ofd_lastid_gen++;
589 up_write(&ofd->ofd_lastid_rwsem);
590 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
595 CERROR("%s: unknown lfsck event: rc = %d\n",
596 ofd_name(ofd), event);
604 * Implementation of lu_device_operations::ldo_prepare.
606 * This method is called after layer has been initialized and before it starts
607 * serving user requests. In OFD it starts lfsk check routines and initializes
610 * \param[in] env execution environment
611 * \param[in] pdev higher device in stack, NULL for OFD
612 * \param[in] dev lu_device of OFD device
614 * \retval 0 if successful
615 * \retval negative value on error
617 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
618 struct lu_device *dev)
620 struct ofd_thread_info *info;
621 struct ofd_device *ofd = ofd_dev(dev);
622 struct obd_device *obd = ofd_obd(ofd);
623 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
628 info = ofd_info_init(env, NULL);
632 /* initialize lower device */
633 rc = next->ld_ops->ldo_prepare(env, dev, next);
637 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
638 ofd_lfsck_out_notify, ofd, false);
640 CERROR("%s: failed to initialize lfsck: rc = %d\n",
645 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
646 /* The LFSCK instance is registered just now, so it must be there when
647 * register the namespace to such instance. */
648 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
650 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
651 OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_OST_PREPARE_DELAY, OBD_FAIL_ONCE,
652 (OBD_TIMEOUT_DEFAULT + 1) / 4);
653 LASSERT(obd->obd_no_conn);
654 spin_lock(&obd->obd_dev_lock);
655 obd->obd_no_conn = 0;
656 spin_unlock(&obd->obd_dev_lock);
658 if (obd->obd_recovering == 0)
659 ofd_postrecov(env, ofd);
665 * Implementation of lu_device_operations::ldo_recovery_complete.
667 * This method notifies all layers about 'recovery complete' event. That means
668 * device is in full state and consistent. An OFD calculates available grant
669 * space upon this event.
671 * \param[in] env execution environment
672 * \param[in] dev lu_device of OFD device
674 * \retval 0 if successful
675 * \retval negative value on error
677 static int ofd_recovery_complete(const struct lu_env *env,
678 struct lu_device *dev)
680 struct ofd_thread_info *oti = ofd_info(env);
681 struct ofd_device *ofd = ofd_dev(dev);
682 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
688 * Grant space for object precreation on the self export.
689 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
690 * is enough to create 10k objects. More space is then acquired for
691 * precreation in tgt_grant_create().
693 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
694 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
695 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
696 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
697 OBD_CONNECT_GRANT_PARAM;
698 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
700 rc = next->ld_ops->ldo_recovery_complete(env, next);
705 * lu_device_operations matrix for OFD device.
707 static struct lu_device_operations ofd_lu_ops = {
708 .ldo_object_alloc = ofd_object_alloc,
709 .ldo_process_config = ofd_process_config,
710 .ldo_recovery_complete = ofd_recovery_complete,
711 .ldo_prepare = ofd_prepare,
714 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
716 * Expose OSD statistics to OFD layer.
718 * The osd interfaces to the backend file system exposes useful data
719 * such as brw_stats and read or write cache states. This same data
720 * needs to be exposed into the obdfilter (ofd) layer to maintain
721 * backwards compatibility. This function creates the symlinks in the
722 * proc layer to enable this.
724 * \param[in] ofd OFD device
726 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
728 struct obd_device *obd = ofd_obd(ofd);
729 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
730 struct kobj_type *osd_type;
733 osd_type = get_ktype(&ofd->ofd_osd->dd_kobj);
734 for (i = 0; osd_type->default_attrs[i]; i++) {
735 if (strcmp(osd_type->default_attrs[i]->name,
736 "read_cache_enable") == 0) {
737 ofd->ofd_read_cache_enable =
738 osd_type->default_attrs[i];
741 if (strcmp(osd_type->default_attrs[i]->name,
742 "readcache_max_filesize") == 0) {
743 ofd->ofd_read_cache_max_filesize =
744 osd_type->default_attrs[i];
747 if (strcmp(osd_type->default_attrs[i]->name,
748 "writethrough_cache_enable") == 0) {
749 ofd->ofd_write_cache_enable =
750 osd_type->default_attrs[i];
754 if (obd->obd_proc_entry == NULL)
757 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
758 "../../%s/%s/brw_stats",
759 osd_obd->obd_type->typ_name, obd->obd_name);
764 * Cleanup all procfs entries in OFD.
766 * \param[in] ofd OFD device
768 static void ofd_procfs_fini(struct ofd_device *ofd)
770 struct obd_device *obd = ofd_obd(ofd);
772 tgt_tunables_fini(&ofd->ofd_lut);
773 lprocfs_free_per_client_stats(obd);
774 lprocfs_obd_cleanup(obd);
775 lprocfs_free_obd_stats(obd);
776 lprocfs_job_stats_fini(obd);
780 * Stop SEQ/FID server on OFD.
782 * \param[in] env execution environment
783 * \param[in] ofd OFD device
785 * \retval 0 if successful
786 * \retval negative value on error
788 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
790 return seq_site_fini(env, &ofd->ofd_seq_site);
794 * Start SEQ/FID server on OFD.
796 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
797 * It also connects to the master server to get own FID sequence (SEQ) range
798 * to this particular OFD. Typically that happens when the OST is first
799 * formatted or in the rare case that it exhausts the local sequence range.
801 * The sequence range is allocated out to the MDTs for OST object allocations,
802 * and not directly to the clients.
804 * \param[in] env execution environment
805 * \param[in] ofd OFD device
807 * \retval 0 if successful
808 * \retval negative value on error
810 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
812 struct seq_server_site *ss = &ofd->ofd_seq_site;
813 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
814 char *obd_name = ofd_name(ofd);
816 int len = strlen(obd_name) + 7;
819 ss = &ofd->ofd_seq_site;
820 lu->ld_site->ld_seq_site = ss;
821 ss->ss_lu = lu->ld_site;
822 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
824 OBD_ALLOC(name, len);
828 OBD_ALLOC_PTR(ss->ss_server_seq);
829 if (ss->ss_server_seq == NULL)
830 GOTO(out_name, rc = -ENOMEM);
832 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
833 LUSTRE_SEQ_SERVER, ss);
835 CERROR("%s: seq server init error: rc = %d\n", obd_name, rc);
836 GOTO(out_server, rc);
838 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
840 OBD_ALLOC_PTR(ss->ss_client_seq);
841 if (ss->ss_client_seq == NULL)
842 GOTO(out_server, rc = -ENOMEM);
844 snprintf(name, len, "%s-super", obd_name);
845 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
848 CERROR("%s: seq client init error: rc = %d\n", obd_name, rc);
849 GOTO(out_client, rc);
852 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
856 seq_client_fini(ss->ss_client_seq);
857 OBD_FREE_PTR(ss->ss_client_seq);
858 ss->ss_client_seq = NULL;
860 seq_server_fini(ss->ss_server_seq, env);
861 OBD_FREE_PTR(ss->ss_server_seq);
862 ss->ss_server_seq = NULL;
871 * OFD request handler for OST_SET_INFO RPC.
873 * This is OFD-specific part of request handling
875 * \param[in] tsi target session environment for this request
877 * \retval 0 if successful
878 * \retval negative value on error
880 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
882 struct ptlrpc_request *req = tgt_ses_req(tsi);
883 struct ost_body *body = NULL, *repbody;
884 void *key, *val = NULL;
885 int keylen, vallen, rc = 0;
886 bool is_grant_shrink;
890 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
892 DEBUG_REQ(D_HA, req, "no set_info key");
893 RETURN(err_serious(-EFAULT));
895 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
898 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
900 DEBUG_REQ(D_HA, req, "no set_info val");
901 RETURN(err_serious(-EFAULT));
903 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
906 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
908 /* In this case the value is actually an RMF_OST_BODY, so we
909 * transmutate the type of this PTLRPC */
910 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
912 rc = req_capsule_server_pack(tsi->tsi_pill);
916 if (is_grant_shrink) {
917 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
919 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
922 /** handle grant shrink, similar to a read request */
923 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
925 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
927 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
930 CERROR("%s: Unsupported key %s\n",
931 tgt_name(tsi->tsi_tgt), (char *)key);
934 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
941 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
943 * This function returns a list of extents which describes how a file's
944 * blocks are laid out on the disk.
946 * \param[in] env execution environment
947 * \param[in] ofd OFD device
948 * \param[in] fid FID of object
949 * \param[in] fiemap fiemap structure to fill with data
951 * \retval 0 if \a fiemap is filled with data successfully
952 * \retval negative value on error
954 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
955 struct lu_fid *fid, struct fiemap *fiemap)
957 struct ofd_object *fo;
960 fo = ofd_object_find(env, ofd, fid);
962 CERROR("%s: error finding object "DFID"\n",
963 ofd_name(ofd), PFID(fid));
967 ofd_read_lock(env, fo);
968 if (ofd_object_exists(fo))
969 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
972 ofd_read_unlock(env, fo);
973 ofd_object_put(env, fo);
978 static int ofd_lock_unlock_region(const struct lu_env *env,
979 struct ldlm_namespace *ns,
980 struct ldlm_res_id *res_id,
981 unsigned long long begin,
982 unsigned long long end)
986 struct lustre_handle lh = { 0 };
988 LASSERT(begin <= end);
990 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
994 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
995 tgt_extent_unlock(&lh, LCK_PR);
1001 * Lock the sparse areas of given resource.
1003 * The locking of sparse areas will cause dirty data to be flushed back from
1004 * clients. This is used when getting the FIEMAP of an object to make sure
1005 * there is no unaccounted cached data on clients.
1007 * This function goes through \a fiemap list of extents and locks only sparse
1008 * areas between extents.
1010 * \param[in] ns LDLM namespace
1011 * \param[in] res_id resource ID
1012 * \param[in] fiemap file extents mapping on disk
1013 * \param[in] locked list head of regions list
1015 * \retval 0 if successful
1016 * \retval negative value on error
1018 static int lock_zero_regions(const struct lu_env *env,
1019 struct ldlm_namespace *ns,
1020 struct ldlm_res_id *res_id,
1021 struct fiemap *fiemap)
1023 __u64 begin = fiemap->fm_start;
1026 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1030 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1031 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1032 if (fiemap_start[i].fe_logical > begin) {
1033 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1034 begin, fiemap_start[i].fe_logical);
1035 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1036 fiemap_start[i].fe_logical);
1041 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1044 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1045 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1046 begin, fiemap->fm_start + fiemap->fm_length);
1047 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1048 fiemap->fm_start + fiemap->fm_length);
1056 * OFD request handler for OST_GET_INFO RPC.
1058 * This is OFD-specific part of request handling. The OFD-specific keys are:
1059 * - KEY_LAST_ID (obsolete)
1063 * This function reads needed data from storage and fills reply with it.
1065 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1066 * and is kept for compatibility.
1068 * \param[in] tsi target session environment for this request
1070 * \retval 0 if successful
1071 * \retval negative value on error
1073 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1075 struct obd_export *exp = tsi->tsi_exp;
1076 struct ofd_device *ofd = ofd_exp(exp);
1077 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1080 int replylen, rc = 0;
1084 /* this common part for get_info rpc */
1085 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1087 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1088 RETURN(err_serious(-EPROTO));
1090 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1093 if (KEY_IS(KEY_LAST_ID)) {
1095 struct ofd_seq *oseq;
1097 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1098 rc = req_capsule_server_pack(tsi->tsi_pill);
1100 RETURN(err_serious(rc));
1102 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1104 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1105 (u64)exp->exp_filter_data.fed_group);
1109 *last_id = ofd_seq_last_oid(oseq);
1110 ofd_seq_put(tsi->tsi_env, oseq);
1111 } else if (KEY_IS(KEY_FIEMAP)) {
1112 struct ll_fiemap_info_key *fm_key;
1113 struct fiemap *fiemap;
1116 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1118 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1119 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1121 RETURN(err_serious(rc));
1123 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1125 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1127 replylen = fiemap_count_to_size(
1128 fm_key->lfik_fiemap.fm_extent_count);
1129 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1130 RCL_SERVER, replylen);
1132 rc = req_capsule_server_pack(tsi->tsi_pill);
1134 RETURN(err_serious(rc));
1136 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1140 *fiemap = fm_key->lfik_fiemap;
1141 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1143 /* LU-3219: Lock the sparse areas to make sure dirty
1144 * flushed back from client, then call fiemap again. */
1145 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1146 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1147 ost_fid_build_resid(fid, &fti->fti_resid);
1148 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1149 &fti->fti_resid, fiemap);
1151 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1154 } else if (KEY_IS(KEY_LAST_FID)) {
1155 struct ofd_device *ofd = ofd_exp(exp);
1156 struct ofd_seq *oseq;
1160 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1161 rc = req_capsule_server_pack(tsi->tsi_pill);
1163 RETURN(err_serious(rc));
1165 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1167 RETURN(err_serious(-EPROTO));
1169 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1171 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1175 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1176 ostid_seq(&fti->fti_ostid));
1178 RETURN(PTR_ERR(oseq));
1180 rc = ostid_to_fid(fid, &oseq->os_oi,
1181 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1185 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1188 ofd_seq_put(tsi->tsi_env, oseq);
1190 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1194 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1201 * OFD request handler for OST_GETATTR RPC.
1203 * This is OFD-specific part of request handling. It finds the OFD object
1204 * by its FID, gets attributes from storage and packs result to the reply.
1206 * \param[in] tsi target session environment for this request
1208 * \retval 0 if successful
1209 * \retval negative value on error
1211 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1213 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1214 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1215 struct ost_body *repbody;
1216 struct lustre_handle lh = { 0 };
1217 struct ofd_object *fo;
1219 enum ldlm_mode lock_mode = LCK_PR;
1224 LASSERT(tsi->tsi_ost_body != NULL);
1226 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1227 if (repbody == NULL)
1230 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1231 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1233 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1234 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1237 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1240 rc = tgt_extent_lock(tsi->tsi_env,
1241 tsi->tsi_tgt->lut_obd->obd_namespace,
1242 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1248 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1250 GOTO(out, rc = PTR_ERR(fo));
1252 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1256 obdo_from_la(&repbody->oa, &fti->fti_attr,
1257 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1259 /* Store object version in reply */
1260 curr_version = dt_version_get(tsi->tsi_env,
1261 ofd_object_child(fo));
1262 if ((__s64)curr_version != -EOPNOTSUPP) {
1263 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1264 repbody->oa.o_data_version = curr_version;
1267 if (fo->ofo_ff.ff_layout_version > 0) {
1268 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1269 repbody->oa.o_layout_version =
1270 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1272 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1273 PFID(&tsi->tsi_fid),
1274 repbody->oa.o_layout_version);
1278 ofd_object_put(tsi->tsi_env, fo);
1281 tgt_extent_unlock(&lh, lock_mode);
1283 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1286 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1287 repbody->oa.o_flags = OBD_FL_FLUSH;
1293 * OFD request handler for OST_SETATTR RPC.
1295 * This is OFD-specific part of request handling. It finds the OFD object
1296 * by its FID, sets attributes from request and packs result to the reply.
1298 * \param[in] tsi target session environment for this request
1300 * \retval 0 if successful
1301 * \retval negative value on error
1303 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1305 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1306 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1307 struct ost_body *body = tsi->tsi_ost_body;
1308 struct ost_body *repbody;
1309 struct ldlm_resource *res;
1310 struct ofd_object *fo;
1315 LASSERT(body != NULL);
1317 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1318 if (repbody == NULL)
1321 repbody->oa.o_oi = body->oa.o_oi;
1322 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1324 /* This would be very bad - accidentally truncating a file when
1325 * changing the time or similar - bug 12203. */
1326 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1327 body->oa.o_size != OBD_OBJECT_EOF) {
1328 static char mdsinum[48];
1330 if (body->oa.o_valid & OBD_MD_FLFID)
1331 snprintf(mdsinum, sizeof(mdsinum) - 1,
1332 "of parent "DFID, body->oa.o_parent_seq,
1333 body->oa.o_parent_oid, 0);
1337 CERROR("%s: setattr from %s is trying to truncate object "DFID
1338 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1339 PFID(&tsi->tsi_fid), mdsinum);
1343 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1345 GOTO(out, rc = PTR_ERR(fo));
1347 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1348 fti->fti_attr.la_valid &= ~LA_TYPE;
1350 /* setting objects attributes (including owner/group) */
1351 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1355 obdo_from_la(&repbody->oa, &fti->fti_attr,
1356 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1358 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1362 ofd_object_put(tsi->tsi_env, fo);
1365 /* we do not call this before to avoid lu_object_find() in
1366 * ->lvbo_update() holding another reference on the object.
1367 * otherwise concurrent destroy can make the object unavailable
1368 * for 2nd lu_object_find() waiting for the first reference
1369 * to go... deadlock! */
1370 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1371 &tsi->tsi_resid, LDLM_EXTENT, 0);
1373 ldlm_res_lvbo_update(res, NULL, 0);
1374 ldlm_resource_putref(res);
1381 * Destroy OST orphans.
1383 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1384 * set then we must destroy possible orphaned objects.
1386 * \param[in] env execution environment
1387 * \param[in] exp OBD export
1388 * \param[in] ofd OFD device
1389 * \param[in] oa obdo structure for reply
1391 * \retval 0 if successful
1392 * \retval negative value on error
1394 static int ofd_orphans_destroy(const struct lu_env *env,
1395 struct obd_export *exp,
1396 struct ofd_device *ofd, struct obdo *oa)
1398 struct ofd_thread_info *info = ofd_info(env);
1399 struct lu_fid *fid = &info->fti_fid;
1400 struct ost_id *oi = &oa->o_oi;
1401 struct ofd_seq *oseq;
1402 u64 seq = ostid_seq(oi);
1403 u64 end_id = ostid_id(oi);
1411 oseq = ofd_seq_get(ofd, seq);
1413 CERROR("%s: Can not find seq for "DOSTID"\n",
1414 ofd_name(ofd), POSTID(oi));
1419 last = ofd_seq_last_oid(oseq);
1422 LASSERT(exp != NULL);
1423 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1425 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1428 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1429 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1431 while (oid > end_id) {
1432 rc = fid_set_id(fid, oid);
1433 if (unlikely(rc != 0))
1436 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1437 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1438 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1439 /* this is pretty fatal... */
1440 CEMERG("%s: error destroying precreated id "
1442 ofd_name(ofd), PFID(fid), rc);
1446 ofd_seq_last_oid_set(oseq, oid);
1447 /* update last_id on disk periodically so that if we
1448 * restart * we don't need to re-scan all of the just
1449 * deleted objects. */
1450 if ((oid & 511) == 0)
1451 ofd_seq_last_oid_write(env, ofd, oseq);
1455 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1456 ofd_name(ofd), seq, oid);
1460 ofd_seq_last_oid_set(oseq, oid);
1461 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1463 /* don't reuse orphan object, return last used objid */
1464 rc = ostid_set_id(oi, last);
1470 ofd_seq_put(env, oseq);
1475 * OFD request handler for OST_CREATE RPC.
1477 * This is OFD-specific part of request handling. Its main purpose is to
1478 * create new data objects on OST, but it also used to destroy orphans.
1480 * \param[in] tsi target session environment for this request
1482 * \retval 0 if successful
1483 * \retval negative value on error
1485 static int ofd_create_hdl(struct tgt_session_info *tsi)
1487 struct ptlrpc_request *req = tgt_ses_req(tsi);
1488 struct ost_body *repbody;
1489 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1490 struct obdo *rep_oa;
1491 struct obd_export *exp = tsi->tsi_exp;
1492 struct ofd_device *ofd = ofd_exp(exp);
1493 u64 seq = ostid_seq(&oa->o_oi);
1494 u64 oid = ostid_id(&oa->o_oi);
1495 struct ofd_seq *oseq;
1503 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1506 if (ofd->ofd_no_precreate)
1509 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1510 if (repbody == NULL)
1513 down_read(&ofd->ofd_lastid_rwsem);
1514 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1515 * we may do that in the future.
1516 * Return -ENOSPC until the LAST_ID rebuilt. */
1517 if (unlikely(ofd->ofd_lastid_rebuilding))
1518 GOTO(out_sem, rc = -ENOSPC);
1520 rep_oa = &repbody->oa;
1521 rep_oa->o_oi = oa->o_oi;
1523 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1525 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1527 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1529 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1530 ofd_name(ofd), seq, PTR_ERR(oseq));
1531 GOTO(out_sem, rc = -EINVAL);
1534 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1535 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1536 if (!ofd_obd(ofd)->obd_recovering ||
1537 oid > ofd_seq_last_oid(oseq)) {
1538 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1539 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1540 ofd_seq_last_oid(oseq));
1541 GOTO(out_nolock, rc = -EINVAL);
1543 /* Do nothing here, we re-create objects during recovery
1544 * upon write replay, see ofd_preprw_write() */
1545 GOTO(out_nolock, rc = 0);
1547 /* former ofd_handle_precreate */
1548 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1549 (oa->o_flags & OBD_FL_DELORPHAN)) {
1550 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1552 /* destroy orphans */
1553 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1554 exp->exp_conn_cnt) {
1555 CERROR("%s: dropping old orphan cleanup request\n",
1557 GOTO(out_nolock, rc = 0);
1559 /* This causes inflight precreates to abort and drop lock */
1560 oseq->os_destroys_in_progress = 1;
1561 mutex_lock(&oseq->os_create_lock);
1562 if (!oseq->os_destroys_in_progress) {
1564 "%s:[%llu] destroys_in_progress already cleared\n",
1565 ofd_name(ofd), seq);
1566 rc = ostid_set_id(&rep_oa->o_oi,
1567 ofd_seq_last_oid(oseq));
1570 diff = oid - ofd_seq_last_oid(oseq);
1571 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1572 ofd_seq_last_oid(oseq), diff);
1573 if (-diff > OST_MAX_PRECREATE) {
1574 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1575 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1576 "(%llu), trust the OST\n",
1577 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1578 PFID(&oseq->os_oi.oi_fid),
1579 ofd_seq_last_oid(oseq));
1581 /* Let MDS know that we are so far ahead. */
1582 rc = ostid_set_id(&rep_oa->o_oi,
1583 ofd_seq_last_oid(oseq) + 1);
1584 } else if (diff < 0) {
1585 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1587 oseq->os_destroys_in_progress = 0;
1589 /* XXX: Used by MDS for the first time! */
1590 oseq->os_destroys_in_progress = 0;
1593 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1594 ofd->ofd_lastid_gen)) {
1595 /* Keep the export ref so we can send the reply. */
1596 ofd_obd_disconnect(class_export_get(exp));
1597 GOTO(out_nolock, rc = -ENOTCONN);
1600 mutex_lock(&oseq->os_create_lock);
1601 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1602 exp->exp_conn_cnt) {
1603 CERROR("%s: dropping old precreate request\n",
1607 /* only precreate if seq is 0, IDIF or normal and also o_id
1608 * must be specfied */
1609 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1610 !fid_seq_is_idif(seq)) || oid == 0) {
1611 diff = 1; /* shouldn't we create this right now? */
1613 diff = oid - ofd_seq_last_oid(oseq);
1614 /* Do sync create if the seq is about to used up */
1615 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1616 if (unlikely(oid >= IDIF_MAX_OID - 1))
1618 } else if (fid_seq_is_norm(seq)) {
1620 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1623 CERROR("%s : invalid o_seq "DOSTID"\n",
1624 ofd_name(ofd), POSTID(&oa->o_oi));
1625 GOTO(out, rc = -EINVAL);
1630 CERROR("%s: invalid precreate request for "
1631 DOSTID", last_id %llu. "
1632 "Likely MDS last_id corruption\n",
1633 ofd_name(ofd), POSTID(&oa->o_oi),
1634 ofd_seq_last_oid(oseq));
1635 GOTO(out, rc = -EINVAL);
1640 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1646 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1647 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1648 /* don't enforce grant during orphan recovery */
1649 granted = tgt_grant_create(tsi->tsi_env,
1650 ofd_obd(ofd)->obd_self_export,
1655 CDEBUG(D_HA, "%s: failed to acquire grant "
1656 "space for precreate (%lld): rc = %d\n",
1657 ofd_name(ofd), diff, rc);
1662 /* This can happen if a new OST is formatted and installed
1663 * in place of an old one at the same index. Instead of
1664 * precreating potentially millions of deleted old objects
1665 * (possibly filling the OST), only precreate the last batch.
1666 * LFSCK will eventually clean up any orphans. LU-14 */
1667 if (diff > 5 * OST_MAX_PRECREATE) {
1668 diff = OST_MAX_PRECREATE / 2;
1669 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1670 "OST replaced or reformatted: "
1671 "LFSCK will clean up",
1674 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1675 "%u larger than the LAST_ID "DOSTID", only "
1676 "precreating the last %lld objects.\n",
1677 ofd_name(ofd), POSTID(&oa->o_oi),
1678 5 * OST_MAX_PRECREATE,
1679 POSTID(&oseq->os_oi), diff);
1680 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1684 next_id = ofd_seq_last_oid(oseq) + 1;
1685 count = ofd_precreate_batch(ofd, (int)diff);
1687 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1688 " at %llu\n", ofd_name(ofd),
1689 count, seq, next_id);
1691 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1692 && ktime_get_seconds() > enough_time) {
1693 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1694 " created at a rate of %d/s\n",
1695 ofd_name(ofd), created, diff + created,
1696 created / DISK_TIMEOUT);
1700 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1701 oseq, count, sync_trans);
1705 } else if (rc < 0) {
1711 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1712 LCONSOLE_WARN("%s: can't create the same count of"
1713 " objects when replaying the request"
1714 " (diff is %lld). see LU-4621\n",
1715 ofd_name(ofd), diff);
1718 /* some objects got created, we can return
1719 * them, even if last creation failed */
1722 CERROR("%s: unable to precreate: rc = %d\n",
1725 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1726 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1727 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1732 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1736 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1738 if (unlikely(!oseq->os_last_id_synced))
1739 oseq->os_last_id_synced = 1;
1741 mutex_unlock(&oseq->os_create_lock);
1744 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1745 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1746 struct lu_fid *fid = &info->fti_fid;
1748 /* For compatible purpose, it needs to convert back to
1749 * OST ID before put it on wire. */
1750 *fid = rep_oa->o_oi.oi_fid;
1751 fid_to_ostid(fid, &rep_oa->o_oi);
1753 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1755 ofd_seq_put(tsi->tsi_env, oseq);
1758 up_read(&ofd->ofd_lastid_rwsem);
1763 * OFD request handler for OST_DESTROY RPC.
1765 * This is OFD-specific part of request handling. It destroys data objects
1766 * related to destroyed object on MDT.
1768 * \param[in] tsi target session environment for this request
1770 * \retval 0 if successful
1771 * \retval negative value on error
1773 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1775 const struct ost_body *body = tsi->tsi_ost_body;
1776 struct ost_body *repbody;
1777 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1778 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1779 struct lu_fid *fid = &fti->fti_fid;
1786 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1789 /* This is old case for clients before Lustre 2.4 */
1790 /* If there's a DLM request, cancel the locks mentioned in it */
1791 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1793 struct ldlm_request *dlm;
1795 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1798 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1801 *fid = body->oa.o_oi.oi_fid;
1802 oid = ostid_id(&body->oa.o_oi);
1805 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1807 /* check that o_misc makes sense */
1808 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1809 count = body->oa.o_misc;
1811 count = 1; /* default case - single destroy */
1813 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1814 POSTID(&body->oa.o_oi), count);
1819 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1820 if (lrc == -ENOENT) {
1822 "%s: destroying non-existent object "DFID"\n",
1823 ofd_name(ofd), PFID(fid));
1824 /* rewrite rc with -ENOENT only if it is 0 */
1827 } else if (lrc != 0) {
1828 CERROR("%s: error destroying object "DFID": %d\n",
1829 ofd_name(ofd), PFID(fid), lrc);
1835 lrc = fid_set_id(fid, oid);
1836 if (unlikely(lrc != 0 && count > 0))
1837 GOTO(out, rc = lrc);
1840 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1846 fid_to_ostid(fid, &repbody->oa.o_oi);
1851 * OFD request handler for OST_STATFS RPC.
1853 * This function gets statfs data from storage as part of request
1856 * \param[in] tsi target session environment for this request
1858 * \retval 0 if successful
1859 * \retval negative value on error
1861 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1863 struct obd_statfs *osfs;
1868 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1870 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1872 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1873 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1875 CERROR("%s: statfs failed: rc = %d\n",
1876 tgt_name(tsi->tsi_tgt), rc);
1878 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1881 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1888 * OFD request handler for OST_SYNC RPC.
1890 * Sync object data or all filesystem data to the disk and pack the
1893 * \param[in] tsi target session environment for this request
1895 * \retval 0 if successful
1896 * \retval negative value on error
1898 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1900 struct ost_body *body = tsi->tsi_ost_body;
1901 struct ost_body *repbody;
1902 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1903 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1904 struct ofd_object *fo = NULL;
1909 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1911 /* if no objid is specified, it means "sync whole filesystem" */
1912 if (!fid_is_zero(&tsi->tsi_fid)) {
1913 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1915 RETURN(PTR_ERR(fo));
1918 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1919 fo != NULL ? ofd_object_child(fo) : NULL,
1920 repbody->oa.o_size, repbody->oa.o_blocks);
1924 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1929 repbody->oa.o_oi = body->oa.o_oi;
1930 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1932 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1934 obdo_from_la(&repbody->oa, &fti->fti_attr,
1937 /* don't return rc from getattr */
1942 ofd_object_put(tsi->tsi_env, fo);
1947 * OFD request handler for OST_PUNCH RPC.
1949 * This is part of request processing. Validate request fields,
1950 * punch (truncate) the given OFD object and pack reply.
1952 * \param[in] tsi target session environment for this request
1954 * \retval 0 if successful
1955 * \retval negative value on error
1957 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1959 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1960 struct ost_body *repbody;
1961 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1962 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1963 struct ldlm_resource *res;
1964 struct ofd_object *fo;
1966 struct lustre_handle lh = { 0, };
1973 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
1975 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1976 BUILD_BUG_ON(!(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK));
1978 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1979 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1980 RETURN(err_serious(-EPROTO));
1982 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1983 if (repbody == NULL)
1984 RETURN(err_serious(-ENOMEM));
1986 /* punch start,end are passed in o_size,o_blocks throught wire */
1990 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1993 /* standard truncate optimization: if file body is completely
1994 * destroyed, don't send data back to the server. */
1996 flags |= LDLM_FL_AST_DISCARD_DATA;
1998 repbody->oa.o_oi = oa->o_oi;
1999 repbody->oa.o_valid = OBD_MD_FLID;
2001 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2002 oa->o_flags & OBD_FL_SRVLOCK;
2005 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
2006 end, &lh, LCK_PW, &flags);
2011 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2012 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2013 oa->o_valid, start, end);
2015 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2018 GOTO(out, rc = PTR_ERR(fo));
2020 la_from_obdo(&info->fti_attr, oa,
2021 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2022 info->fti_attr.la_size = start;
2023 info->fti_attr.la_valid |= LA_SIZE;
2025 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2030 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2034 ofd_object_put(tsi->tsi_env, fo);
2037 tgt_extent_unlock(&lh, LCK_PW);
2039 /* we do not call this before to avoid lu_object_find() in
2040 * ->lvbo_update() holding another reference on the object.
2041 * otherwise concurrent destroy can make the object unavailable
2042 * for 2nd lu_object_find() waiting for the first reference
2043 * to go... deadlock! */
2044 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2047 struct ost_lvb *res_lvb;
2049 ldlm_res_lvbo_update(res, NULL, 0);
2050 res_lvb = res->lr_lvb_data;
2051 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2052 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2053 ldlm_resource_putref(res);
2059 static int ofd_ladvise_prefetch(const struct lu_env *env,
2060 struct ofd_object *fo,
2061 struct niobuf_local *lnb,
2062 __u64 start, __u64 end, enum dt_bufs_type dbt)
2064 struct ofd_thread_info *info = ofd_info(env);
2065 pgoff_t start_index, end_index, pages;
2066 struct niobuf_remote rnb;
2067 unsigned long nr_local;
2073 ofd_read_lock(env, fo);
2074 if (!ofd_object_exists(fo))
2075 GOTO(out_unlock, rc = -ENOENT);
2077 rc = ofd_attr_get(env, fo, &info->fti_attr);
2079 GOTO(out_unlock, rc);
2081 if (end > info->fti_attr.la_size)
2082 end = info->fti_attr.la_size;
2085 GOTO(out_unlock, rc);
2087 /* We need page aligned offset and length */
2088 start_index = start >> PAGE_SHIFT;
2089 end_index = (end - 1) >> PAGE_SHIFT;
2090 pages = end_index - start_index + 1;
2092 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2093 PTLRPC_MAX_BRW_PAGES;
2094 rnb.rnb_offset = start_index << PAGE_SHIFT;
2095 rnb.rnb_len = nr_local << PAGE_SHIFT;
2096 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb,
2097 PTLRPC_MAX_BRW_PAGES, dbt);
2098 if (unlikely(rc < 0))
2101 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2102 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2105 start_index += nr_local;
2110 ofd_read_unlock(env, fo);
2115 * OFD request handler for OST_LADVISE RPC.
2117 * Tune cache or perfetch policies according to advices.
2119 * \param[in] tsi target session environment for this request
2121 * \retval 0 if successful
2122 * \retval negative errno on error
2124 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2126 struct ptlrpc_request *req = tgt_ses_req(tsi);
2127 struct obd_export *exp = tsi->tsi_exp;
2128 struct ofd_device *ofd = ofd_exp(exp);
2129 struct ost_body *body, *repbody;
2130 struct ofd_thread_info *info;
2131 struct ofd_object *fo;
2132 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2133 const struct lu_env *env = svc_thread->t_env;
2134 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2135 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2136 struct lu_ladvise *ladvise;
2138 struct ladvise_hdr *ladvise_hdr;
2139 struct obd_ioobj ioo;
2140 struct lustre_handle lockh = { 0 };
2143 struct dt_object *dob;
2149 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2150 body = tsi->tsi_ost_body;
2152 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2153 RETURN(err_serious(-EPROTO));
2155 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2156 &RMF_OST_LADVISE_HDR);
2157 if (ladvise_hdr == NULL)
2158 RETURN(err_serious(-EPROTO));
2160 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2161 ladvise_hdr->lah_count < 1)
2162 RETURN(err_serious(-EPROTO));
2164 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2165 RETURN(err_serious(-EPROTO));
2167 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2168 if (ladvise == NULL)
2169 RETURN(err_serious(-EPROTO));
2171 num_advise = req_capsule_get_size(&req->rq_pill,
2172 &RMF_OST_LADVISE, RCL_CLIENT) /
2174 if (num_advise < ladvise_hdr->lah_count)
2175 RETURN(err_serious(-EPROTO));
2177 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2178 repbody->oa = body->oa;
2180 info = ofd_info_init(env, exp);
2182 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2183 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2187 fo = ofd_object_find(env, ofd, &info->fti_fid);
2192 LASSERT(fo != NULL);
2193 dob = ofd_object_child(fo);
2195 if (ptlrpc_connection_is_local(exp->exp_connection))
2196 dbt |= DT_BUFS_TYPE_LOCAL;
2198 for (i = 0; i < num_advise; i++, ladvise++) {
2199 start = ladvise->lla_start;
2200 end = ladvise->lla_end;
2202 rc = err_serious(-EPROTO);
2206 /* Handle different advice types */
2207 switch (ladvise->lla_advice) {
2211 case LU_LADVISE_WILLREAD:
2215 ioo.ioo_oid = body->oa.o_oi;
2217 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2218 &tsi->tsi_resid, start, end - 1,
2219 &lockh, LCK_PR, &flags);
2223 req->rq_status = ofd_ladvise_prefetch(env, fo,
2226 tgt_extent_unlock(&lockh, LCK_PR);
2228 case LU_LADVISE_DONTNEED:
2229 rc = dt_ladvise(env, dob, ladvise->lla_start,
2230 ladvise->lla_end, LU_LADVISE_DONTNEED);
2237 ofd_object_put(env, fo);
2238 req->rq_status = rc;
2243 * OFD request handler for OST_QUOTACTL RPC.
2245 * This is part of request processing to validate incoming request fields,
2246 * get the requested data from OSD and pack reply.
2248 * \param[in] tsi target session environment for this request
2250 * \retval 0 if successful
2251 * \retval negative value on error
2253 static int ofd_quotactl(struct tgt_session_info *tsi)
2255 struct obd_quotactl *oqctl, *repoqc;
2256 struct lu_nodemap *nodemap;
2262 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2264 RETURN(err_serious(-EPROTO));
2266 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2268 RETURN(err_serious(-ENOMEM));
2272 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2273 if (IS_ERR(nodemap))
2274 RETURN(PTR_ERR(nodemap));
2277 if (oqctl->qc_type == USRQUOTA)
2278 id = nodemap_map_id(nodemap, NODEMAP_UID,
2279 NODEMAP_CLIENT_TO_FS,
2281 else if (oqctl->qc_type == GRPQUOTA)
2282 id = nodemap_map_id(nodemap, NODEMAP_GID,
2283 NODEMAP_CLIENT_TO_FS,
2286 nodemap_putref(nodemap);
2288 if (repoqc->qc_id != id)
2289 swap(repoqc->qc_id, id);
2291 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2293 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2296 if (repoqc->qc_id != id)
2297 swap(repoqc->qc_id, id);
2303 * Calculate the amount of time for lock prolongation.
2305 * This is helper for ofd_prolong_extent_locks() function to get
2306 * the timeout extra time.
2308 * \param[in] req current request
2310 * \retval amount of time to extend the timeout with
2312 static inline time64_t prolong_timeout(struct ptlrpc_request *req)
2314 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2315 time64_t req_timeout;
2318 return obd_timeout / 2;
2320 req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec;
2321 return max_t(time64_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
2326 * Prolong lock timeout for the given extent.
2328 * This function finds all locks related with incoming request and
2329 * prolongs their timeout.
2331 * If a client is holding a lock for a long time while it sends
2332 * read or write RPCs to the OST for the object under this lock,
2333 * then we don't want the OST to evict the client. Otherwise,
2334 * if the network or disk is very busy then the client may not
2335 * be able to make any progress to clear out dirty pages under
2336 * the lock and the application will fail.
2338 * Every time a Bulk Read/Write (BRW) request arrives for the object
2339 * covered by the lock, extend the timeout on that lock. The RPC should
2340 * contain a lock handle for the lock it is using, but this
2341 * isn't handled correctly by all client versions, and the
2342 * request may cover multiple locks.
2344 * \param[in] tsi target session environment for this request
2345 * \param[in] data struct of data to prolong locks
2348 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2349 struct ldlm_prolong_args *data)
2351 struct obdo *oa = &tsi->tsi_ost_body->oa;
2352 struct ldlm_lock *lock;
2356 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2357 data->lpa_export = tsi->tsi_exp;
2358 data->lpa_resid = tsi->tsi_resid;
2360 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2361 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2362 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2363 data->lpa_extent.end);
2365 if (oa->o_valid & OBD_MD_FLHANDLE) {
2366 /* mostly a request should be covered by only one lock, try
2368 lock = ldlm_handle2lock(&oa->o_handle);
2370 /* Fast path to check if the lock covers the whole IO
2371 * region exclusively. */
2372 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2373 &data->lpa_extent)) {
2375 LASSERT(lock->l_export == data->lpa_export);
2376 ldlm_lock_prolong_one(lock, data);
2377 LDLM_LOCK_PUT(lock);
2378 if (data->lpa_locks_cnt > 0)
2380 /* The lock was destroyed probably lets try
2383 lock->l_last_used = ktime_get();
2384 LDLM_LOCK_PUT(lock);
2389 ldlm_resource_prolong(data);
2394 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2396 * Determine if \a lock and the lock from request \a req are equivalent
2397 * by comparing their resource names, modes, and extents.
2399 * It is used to give priority to read and write RPCs being done
2400 * under this lock so that the client can drop the contended
2401 * lock more quickly and let other clients use it. This improves
2402 * overall performance in the case where the first client gets a
2403 * very large lock extent that prevents other clients from
2404 * submitting their writes.
2406 * \param[in] req ptlrpc_request being processed
2407 * \param[in] lock contended lock to match
2409 * \retval 1 if lock is matched
2410 * \retval 0 otherwise
2412 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2413 struct ldlm_lock *lock)
2415 struct niobuf_remote *rnb;
2416 struct obd_ioobj *ioo;
2417 enum ldlm_mode mode;
2418 struct ldlm_extent ext;
2419 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2423 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2424 LASSERT(ioo != NULL);
2426 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2427 LASSERT(rnb != NULL);
2429 ext.start = rnb->rnb_offset;
2430 rnb += ioo->ioo_bufcnt - 1;
2431 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2433 LASSERT(lock->l_resource != NULL);
2434 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2437 /* a bulk write can only hold a reference on a PW extent lock
2440 mode = LCK_PW | LCK_GROUP;
2441 if (opc == OST_READ)
2442 /* whereas a bulk read can be protected by either a PR or PW
2446 if (!(lock->l_granted_mode & mode))
2449 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2453 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2455 * Check for whether the given PTLRPC request (\a req) is blocking
2456 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2459 * \param[in] req the incoming request
2461 * \retval 1 if \a req is blocking an LDLM lock cancel
2462 * \retval 0 if it is not
2463 * \retval -ESTALE if lock is not found
2465 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2467 struct tgt_session_info *tsi;
2468 struct obd_ioobj *ioo;
2469 struct niobuf_remote *rnb;
2471 struct ldlm_prolong_args pa = { 0 };
2475 /* Don't use tgt_ses_info() to get session info, because lock_match()
2476 * can be called while request has no processing thread yet. */
2477 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2480 * Use LASSERT below because malformed RPCs should have
2481 * been filtered out in tgt_hpreq_handler().
2483 opc = lustre_msg_get_opc(req->rq_reqmsg);
2484 LASSERT(opc == OST_READ || opc == OST_WRITE);
2486 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2487 LASSERT(ioo != NULL);
2489 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2490 LASSERT(rnb != NULL);
2491 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2493 pa.lpa_mode = LCK_PW | LCK_GROUP;
2494 if (opc == OST_READ)
2495 pa.lpa_mode |= LCK_PR;
2497 pa.lpa_extent.start = rnb->rnb_offset;
2498 rnb += ioo->ioo_bufcnt - 1;
2499 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2501 DEBUG_REQ(D_RPCTRACE, req,
2502 "%s %s: refresh rw locks for "DFID" (%llu->%llu)",
2503 tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid),
2504 pa.lpa_extent.start, pa.lpa_extent.end);
2506 ofd_prolong_extent_locks(tsi, &pa);
2508 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n",
2509 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2511 if (pa.lpa_blocks_cnt > 0)
2514 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2518 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2520 * Called after the request has been handled. It refreshes lock timeout again
2521 * so that client has more time to send lock cancel RPC.
2523 * \param[in] req request which is being processed.
2525 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2527 ofd_rw_hpreq_check(req);
2531 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2533 * This function checks if the given lock is the same by its resname, mode
2534 * and extent as one taken from the request.
2535 * It is used to give priority to punch/truncate RPCs that might lead to
2536 * the fastest release of that lock when a lock is contended.
2538 * \param[in] req ptlrpc_request being processed
2539 * \param[in] lock contended lock to match
2541 * \retval 1 if lock is matched
2542 * \retval 0 otherwise
2544 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2545 struct ldlm_lock *lock)
2547 struct tgt_session_info *tsi;
2549 struct ldlm_extent ext;
2553 /* Don't use tgt_ses_info() to get session info, because lock_match()
2554 * can be called while request has no processing thread yet. */
2555 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2558 * Use LASSERT below because malformed RPCs should have
2559 * been filtered out in tgt_hpreq_handler().
2561 LASSERT(tsi->tsi_ost_body != NULL);
2562 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2563 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2566 oa = &tsi->tsi_ost_body->oa;
2567 ext.start = oa->o_size;
2568 ext.end = oa->o_blocks;
2570 LASSERT(lock->l_resource != NULL);
2571 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2574 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2577 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2581 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2583 * High-priority queue request check for whether the given punch request
2584 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2585 * covered by an LDLM lock.
2589 * \param[in] req the incoming request
2591 * \retval 1 if \a req is blocking an LDLM lock cancel
2592 * \retval 0 if it is not
2593 * \retval -ESTALE if lock is not found
2595 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2597 struct tgt_session_info *tsi;
2599 struct ldlm_prolong_args pa = { 0 };
2603 /* Don't use tgt_ses_info() to get session info, because lock_match()
2604 * can be called while request has no processing thread yet. */
2605 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2606 LASSERT(tsi != NULL);
2607 oa = &tsi->tsi_ost_body->oa;
2609 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2610 oa->o_flags & OBD_FL_SRVLOCK));
2612 pa.lpa_mode = LCK_PW | LCK_GROUP;
2613 pa.lpa_extent.start = oa->o_size;
2614 pa.lpa_extent.end = oa->o_blocks;
2617 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2618 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2619 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2621 ofd_prolong_extent_locks(tsi, &pa);
2623 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2624 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2626 if (pa.lpa_blocks_cnt > 0)
2629 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2633 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2635 * Called after the request has been handled. It refreshes lock timeout again
2636 * so that client has more time to send lock cancel RPC.
2638 * \param[in] req request which is being processed.
2640 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2642 ofd_punch_hpreq_check(req);
2645 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2646 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2647 .hpreq_check = ofd_rw_hpreq_check,
2648 .hpreq_fini = ofd_rw_hpreq_fini
2651 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2652 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2653 .hpreq_check = ofd_punch_hpreq_check,
2654 .hpreq_fini = ofd_punch_hpreq_fini
2658 * Assign high priority operations to an IO request.
2660 * Check if the incoming request is a candidate for
2661 * high-priority processing. If it is, assign it a high
2662 * priority operations table.
2664 * \param[in] tsi target session environment for this request
2666 static void ofd_hp_brw(struct tgt_session_info *tsi)
2668 struct niobuf_remote *rnb;
2669 struct obd_ioobj *ioo;
2673 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2674 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2675 if (ioo->ioo_bufcnt > 0) {
2676 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2677 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2679 /* no high priority if server lock is needed */
2680 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2681 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2685 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2689 * Assign high priority operations to an punch request.
2691 * Check if the incoming request is a candidate for
2692 * high-priority processing. If it is, assign it a high
2693 * priority operations table.
2695 * \param[in] tsi target session environment for this request
2697 static void ofd_hp_punch(struct tgt_session_info *tsi)
2699 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2700 /* no high-priority if server lock is needed */
2701 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2702 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2703 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2704 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2706 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2709 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2710 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2711 #define OST_BRW_READ OST_READ
2712 #define OST_BRW_WRITE OST_WRITE
2715 * Table of OFD-specific request handlers
2717 * This table contains all opcodes accepted by OFD and
2718 * specifies handlers for them. The tgt_request_handler()
2719 * uses such table from each target to process incoming
2722 static struct tgt_handler ofd_tgt_handlers[] = {
2723 TGT_RPC_HANDLER(OST_FIRST_OPC,
2724 0, OST_CONNECT, tgt_connect,
2725 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2726 TGT_RPC_HANDLER(OST_FIRST_OPC,
2727 0, OST_DISCONNECT, tgt_disconnect,
2728 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2729 TGT_RPC_HANDLER(OST_FIRST_OPC,
2730 0, OST_SET_INFO, ofd_set_info_hdl,
2731 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2732 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2733 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl),
2734 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2735 OST_SETATTR, ofd_setattr_hdl),
2736 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2737 OST_CREATE, ofd_create_hdl),
2738 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2739 OST_DESTROY, ofd_destroy_hdl),
2740 TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl),
2741 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
2743 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2744 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write,
2746 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2747 OST_PUNCH, ofd_punch_hdl,
2749 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl),
2750 TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl),
2751 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl),
2754 static struct tgt_opc_slice ofd_common_slice[] = {
2756 .tos_opc_start = OST_FIRST_OPC,
2757 .tos_opc_end = OST_LAST_OPC,
2758 .tos_hs = ofd_tgt_handlers
2761 .tos_opc_start = OBD_FIRST_OPC,
2762 .tos_opc_end = OBD_LAST_OPC,
2763 .tos_hs = tgt_obd_handlers
2766 .tos_opc_start = LDLM_FIRST_OPC,
2767 .tos_opc_end = LDLM_LAST_OPC,
2768 .tos_hs = tgt_dlm_handlers
2771 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2772 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2773 .tos_hs = tgt_out_handlers
2776 .tos_opc_start = SEQ_FIRST_OPC,
2777 .tos_opc_end = SEQ_LAST_OPC,
2778 .tos_hs = seq_handlers
2781 .tos_opc_start = LFSCK_FIRST_OPC,
2782 .tos_opc_end = LFSCK_LAST_OPC,
2783 .tos_hs = tgt_lfsck_handlers
2786 .tos_opc_start = SEC_FIRST_OPC,
2787 .tos_opc_end = SEC_LAST_OPC,
2788 .tos_hs = tgt_sec_ctx_handlers
2795 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2796 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2799 * Implementation of lu_context_key::lct_key_exit.
2801 * Optional method called on lu_context_exit() for all allocated
2803 * It is used in OFD to sanitize context values which may be re-used
2804 * during another request processing by the same thread.
2806 * \param[in] ctx execution context
2807 * \param[in] key context key
2808 * \param[in] data ofd_thread_info
2810 static void ofd_key_exit(const struct lu_context *ctx,
2811 struct lu_context_key *key, void *data)
2813 struct ofd_thread_info *info = data;
2815 info->fti_env = NULL;
2816 info->fti_exp = NULL;
2819 info->fti_pre_version = 0;
2821 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2824 struct lu_context_key ofd_thread_key = {
2825 .lct_tags = LCT_DT_THREAD,
2826 .lct_init = ofd_key_init,
2827 .lct_fini = ofd_key_fini,
2828 .lct_exit = ofd_key_exit
2832 * Initialize OFD device according to parameters in the config log \a cfg.
2834 * This is the main starting point of OFD initialization. It fills all OFD
2835 * parameters with their initial values and calls other initializing functions
2836 * to set up all OFD subsystems.
2838 * \param[in] env execution environment
2839 * \param[in] m OFD device
2840 * \param[in] ldt LU device type of OFD
2841 * \param[in] cfg configuration log
2843 * \retval 0 if successful
2844 * \retval negative value on error
2846 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2847 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2849 const char *dev = lustre_cfg_string(cfg, 0);
2850 struct ofd_thread_info *info = NULL;
2851 struct obd_device *obd;
2852 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2854 struct nm_config_file *nodemap_config;
2855 struct obd_device_target *obt;
2861 obd = class_name2obd(dev);
2863 CERROR("Cannot find obd with name %s\n", dev);
2867 rc = lu_env_refill((struct lu_env *)env);
2872 obt->obt_magic = OBT_MAGIC;
2874 spin_lock_init(&m->ofd_flags_lock);
2875 m->ofd_raid_degraded = 0;
2876 m->ofd_checksum_t10pi_enforce = 0;
2877 m->ofd_sync_journal = 0;
2879 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2881 m->ofd_seq_count = 0;
2882 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2883 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2884 spin_lock_init(&m->ofd_inconsistency_lock);
2886 spin_lock_init(&m->ofd_batch_lock);
2887 init_rwsem(&m->ofd_lastid_rwsem);
2889 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2890 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2891 /* set this lu_device to obd, because error handling need it */
2892 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2894 /* No connection accepted until configurations will finish */
2895 spin_lock(&obd->obd_dev_lock);
2896 obd->obd_no_conn = 1;
2897 spin_unlock(&obd->obd_dev_lock);
2898 obd->obd_replayable = 1;
2899 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2900 char *str = lustre_cfg_string(cfg, 4);
2902 if (strchr(str, 'n')) {
2903 CWARN("%s: recovery disabled\n", obd->obd_name);
2904 obd->obd_replayable = 0;
2908 info = ofd_info_init(env, NULL);
2912 rc = ofd_stack_init(env, m, cfg, &lmd_flags);
2914 CERROR("%s: can't init device stack, rc %d\n",
2919 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
2920 ofd_procfs_add_brw_stats_symlink(m);
2923 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2924 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2925 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2926 LDLM_NAMESPACE_SERVER,
2927 LDLM_NAMESPACE_GREEDY,
2929 if (m->ofd_namespace == NULL)
2930 GOTO(err_fini_stack, rc = -ENOMEM);
2931 /* set obd_namespace for compatibility with old code */
2932 obd->obd_namespace = m->ofd_namespace;
2933 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2934 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2935 m->ofd_namespace->ns_lvbp = m;
2937 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2938 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2940 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2941 OBD_FAIL_OST_ALL_REQUEST_NET,
2942 OBD_FAIL_OST_ALL_REPLY_NET);
2944 GOTO(err_free_ns, rc);
2946 if (lmd_flags & LMD_FLG_SKIP_LFSCK)
2947 m->ofd_skip_lfsck = 1;
2948 if (lmd_flags & LMD_FLG_LOCAL_RECOV)
2949 m->ofd_lut.lut_local_recovery = 1;
2951 rc = ofd_tunables_init(m);
2953 GOTO(err_fini_lut, rc);
2955 tgd->tgd_reserved_pcnt = 0;
2957 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
2958 m->ofd_cksum_types_supported =
2959 obd_cksum_types_supported_server(obd->obd_name);
2960 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2961 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
2962 OFD_PRECREATE_SMALL_FS)
2963 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2965 rc = ofd_fs_setup(env, m, obd);
2967 GOTO(err_fini_proc, rc);
2969 fid.f_seq = FID_SEQ_LOCAL_NAME;
2972 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
2975 GOTO(err_fini_fs, rc);
2977 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
2979 if (IS_ERR(nodemap_config)) {
2980 rc = PTR_ERR(nodemap_config);
2982 GOTO(err_fini_los, rc);
2984 obt->obt_nodemap_config_file = nodemap_config;
2987 rc = ofd_start_inconsistency_verification_thread(m);
2989 GOTO(err_fini_nm, rc);
2991 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
2996 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
2997 obt->obt_nodemap_config_file = NULL;
2999 local_oid_storage_fini(env, m->ofd_los);
3002 ofd_fs_cleanup(env, m);
3006 tgt_fini(env, &m->ofd_lut);
3008 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3009 obd->obd_namespace = m->ofd_namespace = NULL;
3011 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3016 * Stop the OFD device
3018 * This function stops the OFD device and all its subsystems.
3019 * This is the end of OFD lifecycle.
3021 * \param[in] env execution environment
3022 * \param[in] m OFD device
3024 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3026 struct obd_device *obd = ofd_obd(m);
3027 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3028 struct lfsck_stop stop;
3030 stop.ls_status = LS_PAUSED;
3032 lfsck_stop(env, m->ofd_osd, &stop);
3033 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3034 target_recovery_fini(obd);
3035 if (m->ofd_namespace != NULL)
3036 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3037 d->ld_obd->obd_force);
3039 obd_exports_barrier(obd);
3040 obd_zombie_barrier();
3043 tgt_fini(env, &m->ofd_lut);
3044 ofd_stop_inconsistency_verification_thread(m);
3045 lfsck_degister(env, m->ofd_osd);
3046 ofd_fs_cleanup(env, m);
3047 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3048 obd->u.obt.obt_nodemap_config_file = NULL;
3050 if (m->ofd_namespace != NULL) {
3051 ldlm_namespace_free_post(m->ofd_namespace);
3052 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3055 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3057 LASSERT(atomic_read(&d->ld_ref) == 0);
3058 server_put_mount(obd->obd_name, true);
3063 * Implementation of lu_device_type_operations::ldto_device_fini.
3065 * Finalize device. Dual to ofd_device_init(). It is called from
3066 * obd_precleanup() and stops the current device.
3068 * \param[in] env execution environment
3069 * \param[in] d LU device of OFD
3073 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3074 struct lu_device *d)
3077 ofd_fini(env, ofd_dev(d));
3082 * Implementation of lu_device_type_operations::ldto_device_free.
3084 * Free OFD device. Dual to ofd_device_alloc().
3086 * \param[in] env execution environment
3087 * \param[in] d LU device of OFD
3091 static struct lu_device *ofd_device_free(const struct lu_env *env,
3092 struct lu_device *d)
3094 struct ofd_device *m = ofd_dev(d);
3096 dt_device_fini(&m->ofd_dt_dev);
3102 * Implementation of lu_device_type_operations::ldto_device_alloc.
3104 * This function allocates the new OFD device. It is called from
3105 * obd_setup() if OBD device had lu_device_type defined.
3107 * \param[in] env execution environment
3108 * \param[in] t lu_device_type of OFD device
3109 * \param[in] cfg configuration log
3111 * \retval pointer to the lu_device of just allocated OFD
3112 * \retval ERR_PTR of return value on error
3114 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3115 struct lu_device_type *t,
3116 struct lustre_cfg *cfg)
3118 struct ofd_device *m;
3119 struct lu_device *l;
3124 return ERR_PTR(-ENOMEM);
3126 l = &m->ofd_dt_dev.dd_lu_dev;
3127 dt_device_init(&m->ofd_dt_dev, t);
3128 rc = ofd_init0(env, m, t, cfg);
3130 ofd_device_free(env, l);
3137 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3138 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3140 static struct lu_device_type_operations ofd_device_type_ops = {
3141 .ldto_init = ofd_type_init,
3142 .ldto_fini = ofd_type_fini,
3144 .ldto_start = ofd_type_start,
3145 .ldto_stop = ofd_type_stop,
3147 .ldto_device_alloc = ofd_device_alloc,
3148 .ldto_device_free = ofd_device_free,
3149 .ldto_device_fini = ofd_device_fini
3152 static struct lu_device_type ofd_device_type = {
3153 .ldt_tags = LU_DEVICE_DT,
3154 .ldt_name = LUSTRE_OST_NAME,
3155 .ldt_ops = &ofd_device_type_ops,
3156 .ldt_ctx_tags = LCT_DT_THREAD
3160 * Initialize OFD module.
3162 * This function is called upon module loading. It registers OFD device type
3163 * and prepares all in-memory structures used by all OFD devices.
3165 * \retval 0 if successful
3166 * \retval negative value on error
3168 static int __init ofd_init(void)
3172 rc = lu_kmem_init(ofd_caches);
3175 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3176 LUSTRE_OST_NAME, &ofd_device_type);
3183 * This function is called upon OFD module unloading.
3184 * It frees all related structures and unregisters OFD device type.
3186 static void __exit ofd_exit(void)
3188 lu_kmem_fini(ofd_caches);
3189 class_unregister_type(LUSTRE_OST_NAME);
3192 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3193 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3194 MODULE_VERSION(LUSTRE_VERSION_STRING);
3195 MODULE_LICENSE("GPL");
3197 module_init(ofd_init);
3198 module_exit(ofd_exit);