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 struct kmem_cache *tgt_fmd_kmem;
87 static struct lu_kmem_descr ofd_caches[] = {
89 .ckd_cache = &ofd_object_kmem,
90 .ckd_name = "ofd_obj",
91 .ckd_size = sizeof(struct ofd_object)
94 .ckd_cache = &tgt_fmd_kmem,
95 .ckd_name = "ll_fmd_cache",
96 .ckd_size = sizeof(struct tgt_fmd_data)
104 * Connect OFD to the next device in the stack.
106 * This function is used for device stack configuration and links OFD
107 * device with bottom OSD device.
109 * \param[in] env execution environment
110 * \param[in] m OFD device
111 * \param[in] next name of next device in the stack
112 * \param[out] exp export to return
114 * \retval 0 and export in \a exp if successful
115 * \retval negative value on error
117 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
118 const char *next, struct obd_export **exp)
120 struct obd_connect_data *data = NULL;
121 struct obd_device *obd;
127 GOTO(out, rc = -ENOMEM);
129 obd = class_name2obd(next);
131 CERROR("%s: can't locate next device: %s\n",
133 GOTO(out, rc = -ENOTCONN);
136 data->ocd_connect_flags = OBD_CONNECT_VERSION;
137 data->ocd_version = LUSTRE_VERSION_CODE;
139 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
141 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
142 ofd_name(m), next, rc);
146 m->ofd_dt_dev.dd_lu_dev.ld_site =
147 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
148 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
149 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
150 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
159 * Initialize stack of devices.
161 * This function initializes OFD-OSD device stack to serve OST requests
163 * \param[in] env execution environment
164 * \param[in] m OFD device
165 * \param[in] cfg Lustre config for this server
167 * \retval 0 if successful
168 * \retval negative value on error
170 static int ofd_stack_init(const struct lu_env *env,
171 struct ofd_device *m, struct lustre_cfg *cfg)
173 const char *dev = lustre_cfg_string(cfg, 0);
175 struct ofd_thread_info *info = ofd_info(env);
176 struct lustre_mount_info *lmi;
177 struct lustre_mount_data *lmd;
183 lmi = server_get_mount(dev);
185 CERROR("Cannot get mount info for %s!\n", dev);
189 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
190 if (lmd != NULL && lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
191 m->ofd_skip_lfsck = 1;
193 /* find bottom osd */
194 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
198 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
199 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
200 OBD_FREE(osdname, MTI_NAME_MAXLEN);
204 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
206 m->ofd_osd = lu2dt_dev(d);
208 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
209 "%s-osd", lustre_cfg_string(cfg, 0));
215 * Finalize the device stack OFD-OSD.
217 * This function cleans OFD-OSD device stack and
218 * disconnects OFD from the OSD.
220 * \param[in] env execution environment
221 * \param[in] m OFD device
222 * \param[in] top top device of stack
224 * \retval 0 if successful
225 * \retval negative value on error
227 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
228 struct lu_device *top)
230 struct obd_device *obd = ofd_obd(m);
231 struct lustre_cfg_bufs bufs;
232 struct lustre_cfg *lcfg;
237 lu_site_purge(env, top->ld_site, ~0);
238 /* process cleanup, pass mdt obd name to get obd umount flags */
239 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
244 lustre_cfg_bufs_set_string(&bufs, 1, flags);
245 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
248 lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs);
251 top->ld_ops->ldo_process_config(env, top, lcfg);
252 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
254 if (m->ofd_los != NULL) {
255 local_oid_storage_fini(env, m->ofd_los);
259 lu_site_purge(env, top->ld_site, ~0);
260 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
261 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
262 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
265 LASSERT(m->ofd_osd_exp);
266 obd_disconnect(m->ofd_osd_exp);
271 static void ofd_stack_pre_fini(const struct lu_env *env, struct ofd_device *m,
272 struct lu_device *top)
274 struct lustre_cfg_bufs bufs;
275 struct lustre_cfg *lcfg;
280 lustre_cfg_bufs_reset(&bufs, ofd_name(m));
281 lustre_cfg_bufs_set_string(&bufs, 1, NULL);
282 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
284 CERROR("%s: failed to trigger LCFG_PRE_CLEANUP\n", ofd_name(m));
286 lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, &bufs);
287 top->ld_ops->ldo_process_config(env, top, lcfg);
288 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
289 lcfg->lcfg_buflens));
295 /* For interoperability, see mdt_interop_param[]. */
296 static struct cfg_interop_param ofd_interop_param[] = {
297 { "ost.quota_type", NULL },
302 * Check if parameters are symlinks to the OSD.
304 * Some parameters were moved from ofd to osd and only their
305 * symlinks were kept in ofd by LU-3106. They are:
306 * -writehthrough_cache_enable
307 * -readcache_max_filesize
311 * Since they are not included by the static lprocfs var list, a pre-check
312 * is added for them to avoid "unknown param" errors. If they are matched
313 * in this check, they will be passed to the OSD directly.
315 * \param[in] param parameters to check
317 * \retval true if param is symlink to OSD param
320 static bool match_symlink_param(char *param)
325 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
326 sval = strchr(param, '=');
328 paramlen = sval - param;
329 if (strncmp(param, "brw_stats", paramlen) == 0)
338 * Process various configuration parameters.
340 * This function is used by MGS to process specific configurations and
341 * pass them through to the next device in server stack, i.e. the OSD.
343 * \param[in] env execution environment
344 * \param[in] d LU device of OFD
345 * \param[in] cfg parameters to process
347 * \retval 0 if successful
348 * \retval negative value on error
350 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
351 struct lustre_cfg *cfg)
353 struct ofd_device *m = ofd_dev(d);
354 struct dt_device *dt_next = m->ofd_osd;
355 struct lu_device *next = &dt_next->dd_lu_dev;
360 switch (cfg->lcfg_command) {
362 /* For interoperability */
363 struct cfg_interop_param *ptr = NULL;
364 struct lustre_cfg *old_cfg = NULL;
368 param = lustre_cfg_string(cfg, 1);
370 CERROR("param is empty\n");
375 ptr = class_find_old_param(param, ofd_interop_param);
377 if (ptr->new_param == NULL) {
379 CWARN("For interoperability, skip this %s."
380 " It is obsolete.\n", ptr->old_param);
384 CWARN("Found old param %s, changed it to %s.\n",
385 ptr->old_param, ptr->new_param);
388 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
395 if (match_symlink_param(param)) {
396 rc = next->ld_ops->ldo_process_config(env, next, cfg);
400 count = class_modify_config(cfg, PARAM_OST,
401 &d->ld_obd->obd_kset.kobj);
406 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
408 /* we don't understand; pass it on */
409 rc = next->ld_ops->ldo_process_config(env, next, cfg);
412 case LCFG_SPTLRPC_CONF: {
417 /* others are passed further */
418 rc = next->ld_ops->ldo_process_config(env, next, cfg);
425 * Implementation of lu_object_operations::loo_object_init for OFD
427 * Allocate just the next object (OSD) in stack.
429 * \param[in] env execution environment
430 * \param[in] o lu_object of OFD object
431 * \param[in] conf additional configuration parameters, not used here
433 * \retval 0 if successful
434 * \retval negative value on error
436 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
437 const struct lu_object_conf *conf)
439 struct ofd_device *d = ofd_dev(o->lo_dev);
440 struct lu_device *under;
441 struct lu_object *below;
446 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
447 PFID(lu_object_fid(o)));
449 under = &d->ofd_osd->dd_lu_dev;
450 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
452 lu_object_add(o, below);
460 * Implementation of lu_object_operations::loo_object_free.
462 * Finish OFD object lifecycle and free its memory.
464 * \param[in] env execution environment
465 * \param[in] o LU object of OFD object
467 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
469 struct ofd_object *of = ofd_obj(o);
470 struct lu_object_header *h;
475 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
476 PFID(lu_object_fid(o)));
479 lu_object_header_fini(h);
480 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
485 * Implementation of lu_object_operations::loo_object_print.
487 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
488 * LU_OBJECT_DEBUG() for more details about the compound object printing.
490 * \param[in] env execution environment
491 * \param[in] cookie opaque data passed to the printer function
492 * \param[in] p printer function to use
493 * \param[in] o LU object of OFD object
495 * \retval 0 if successful
496 * \retval negative value on error
498 static int ofd_object_print(const struct lu_env *env, void *cookie,
499 lu_printer_t p, const struct lu_object *o)
501 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
504 static struct lu_object_operations ofd_obj_ops = {
505 .loo_object_init = ofd_object_init,
506 .loo_object_free = ofd_object_free,
507 .loo_object_print = ofd_object_print
511 * Implementation of lu_device_operations::lod_object_alloc.
513 * This function allocates OFD part of compound OFD-OSD object and
514 * initializes its header, because OFD is the top device in stack
516 * \param[in] env execution environment
517 * \param[in] hdr object header, NULL for OFD
518 * \param[in] d lu_device
520 * \retval allocated object if successful
521 * \retval NULL value on failed allocation
523 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
524 const struct lu_object_header *hdr,
527 struct ofd_object *of;
531 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
534 struct lu_object_header *h;
536 o = &of->ofo_obj.do_lu;
538 lu_object_header_init(h);
539 lu_object_init(o, h, d);
540 lu_object_add_top(h, o);
541 o->lo_ops = &ofd_obj_ops;
549 * Return the result of LFSCK run to the OFD.
551 * Notify OFD about result of LFSCK run. That may block the new object
552 * creation until problem is fixed by LFSCK.
554 * \param[in] env execution environment
555 * \param[in] data pointer to the OFD device
556 * \param[in] event LFSCK event type
558 * \retval 0 if successful
559 * \retval negative value on unknown event
561 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
562 enum lfsck_events event)
564 struct ofd_device *ofd = data;
565 struct obd_device *obd = ofd_obd(ofd);
568 case LE_LASTID_REBUILDING:
569 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
570 "on the device until the LAST_ID rebuilt successfully.\n",
572 down_write(&ofd->ofd_lastid_rwsem);
573 ofd->ofd_lastid_rebuilding = 1;
574 up_write(&ofd->ofd_lastid_rwsem);
576 case LE_LASTID_REBUILT: {
577 down_write(&ofd->ofd_lastid_rwsem);
578 ofd_seqs_free(env, ofd);
579 ofd->ofd_lastid_rebuilding = 0;
580 ofd->ofd_lastid_gen++;
581 up_write(&ofd->ofd_lastid_rwsem);
582 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
587 CERROR("%s: unknown lfsck event: rc = %d\n",
588 ofd_name(ofd), event);
596 * Implementation of lu_device_operations::ldo_prepare.
598 * This method is called after layer has been initialized and before it starts
599 * serving user requests. In OFD it starts lfsk check routines and initializes
602 * \param[in] env execution environment
603 * \param[in] pdev higher device in stack, NULL for OFD
604 * \param[in] dev lu_device of OFD device
606 * \retval 0 if successful
607 * \retval negative value on error
609 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
610 struct lu_device *dev)
612 struct ofd_thread_info *info;
613 struct ofd_device *ofd = ofd_dev(dev);
614 struct obd_device *obd = ofd_obd(ofd);
615 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
620 info = ofd_info_init(env, NULL);
624 /* initialize lower device */
625 rc = next->ld_ops->ldo_prepare(env, dev, next);
629 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
630 ofd_lfsck_out_notify, ofd, false);
632 CERROR("%s: failed to initialize lfsck: rc = %d\n",
637 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
638 /* The LFSCK instance is registered just now, so it must be there when
639 * register the namespace to such instance. */
640 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
642 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
643 LASSERT(obd->obd_no_conn);
644 spin_lock(&obd->obd_dev_lock);
645 obd->obd_no_conn = 0;
646 spin_unlock(&obd->obd_dev_lock);
648 if (obd->obd_recovering == 0)
649 ofd_postrecov(env, ofd);
655 * Implementation of lu_device_operations::ldo_recovery_complete.
657 * This method notifies all layers about 'recovery complete' event. That means
658 * device is in full state and consistent. An OFD calculates available grant
659 * space upon this event.
661 * \param[in] env execution environment
662 * \param[in] dev lu_device of OFD device
664 * \retval 0 if successful
665 * \retval negative value on error
667 static int ofd_recovery_complete(const struct lu_env *env,
668 struct lu_device *dev)
670 struct ofd_thread_info *oti = ofd_info(env);
671 struct ofd_device *ofd = ofd_dev(dev);
672 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
678 * Grant space for object precreation on the self export.
679 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
680 * is enough to create 10k objects. More space is then acquired for
681 * precreation in tgt_grant_create().
683 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
684 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
685 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
686 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
687 OBD_CONNECT_GRANT_PARAM;
688 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
690 rc = next->ld_ops->ldo_recovery_complete(env, next);
695 * lu_device_operations matrix for OFD device.
697 static struct lu_device_operations ofd_lu_ops = {
698 .ldo_object_alloc = ofd_object_alloc,
699 .ldo_process_config = ofd_process_config,
700 .ldo_recovery_complete = ofd_recovery_complete,
701 .ldo_prepare = ofd_prepare,
704 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
706 * Expose OSD statistics to OFD layer.
708 * The osd interfaces to the backend file system exposes useful data
709 * such as brw_stats and read or write cache states. This same data
710 * needs to be exposed into the obdfilter (ofd) layer to maintain
711 * backwards compatibility. This function creates the symlinks in the
712 * proc layer to enable this.
714 * \param[in] ofd OFD device
716 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
718 struct obd_device *obd = ofd_obd(ofd);
719 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
720 struct kobj_type *osd_type;
723 osd_type = get_ktype(&ofd->ofd_osd->dd_kobj);
724 for (i = 0; osd_type->default_attrs[i]; i++) {
725 if (strcmp(osd_type->default_attrs[i]->name,
726 "read_cache_enable") == 0) {
727 ofd->ofd_read_cache_enable =
728 osd_type->default_attrs[i];
731 if (strcmp(osd_type->default_attrs[i]->name,
732 "readcache_max_filesize") == 0) {
733 ofd->ofd_read_cache_max_filesize =
734 osd_type->default_attrs[i];
737 if (strcmp(osd_type->default_attrs[i]->name,
738 "writethrough_cache_enable") == 0) {
739 ofd->ofd_write_cache_enable =
740 osd_type->default_attrs[i];
744 if (obd->obd_proc_entry == NULL)
747 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
748 "../../%s/%s/brw_stats",
749 osd_obd->obd_type->typ_name, obd->obd_name);
754 * Cleanup all procfs entries in OFD.
756 * \param[in] ofd OFD device
758 static void ofd_procfs_fini(struct ofd_device *ofd)
760 struct obd_device *obd = ofd_obd(ofd);
762 lprocfs_free_per_client_stats(obd);
763 lprocfs_obd_cleanup(obd);
764 lprocfs_free_obd_stats(obd);
765 lprocfs_job_stats_fini(obd);
769 * Stop SEQ/FID server on OFD.
771 * \param[in] env execution environment
772 * \param[in] ofd OFD device
774 * \retval 0 if successful
775 * \retval negative value on error
777 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
779 return seq_site_fini(env, &ofd->ofd_seq_site);
783 * Start SEQ/FID server on OFD.
785 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
786 * It also connects to the master server to get own FID sequence (SEQ) range
787 * to this particular OFD. Typically that happens when the OST is first
788 * formatted or in the rare case that it exhausts the local sequence range.
790 * The sequence range is allocated out to the MDTs for OST object allocations,
791 * and not directly to the clients.
793 * \param[in] env execution environment
794 * \param[in] ofd OFD device
796 * \retval 0 if successful
797 * \retval negative value on error
799 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
801 struct seq_server_site *ss = &ofd->ofd_seq_site;
802 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
803 char *obd_name = ofd_name(ofd);
805 int len = strlen(obd_name) + 7;
808 ss = &ofd->ofd_seq_site;
809 lu->ld_site->ld_seq_site = ss;
810 ss->ss_lu = lu->ld_site;
811 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
813 OBD_ALLOC(name, len);
817 OBD_ALLOC_PTR(ss->ss_server_seq);
818 if (ss->ss_server_seq == NULL)
819 GOTO(out_name, rc = -ENOMEM);
821 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
822 LUSTRE_SEQ_SERVER, ss);
824 CERROR("%s: seq server init error: rc = %d\n", obd_name, rc);
825 GOTO(out_server, rc);
827 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
829 OBD_ALLOC_PTR(ss->ss_client_seq);
830 if (ss->ss_client_seq == NULL)
831 GOTO(out_server, rc = -ENOMEM);
833 snprintf(name, len, "%s-super", obd_name);
834 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
837 CERROR("%s: seq client init error: rc = %d\n", obd_name, rc);
838 GOTO(out_client, rc);
841 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
845 seq_client_fini(ss->ss_client_seq);
846 OBD_FREE_PTR(ss->ss_client_seq);
847 ss->ss_client_seq = NULL;
849 seq_server_fini(ss->ss_server_seq, env);
850 OBD_FREE_PTR(ss->ss_server_seq);
851 ss->ss_server_seq = NULL;
860 * OFD request handler for OST_SET_INFO RPC.
862 * This is OFD-specific part of request handling
864 * \param[in] tsi target session environment for this request
866 * \retval 0 if successful
867 * \retval negative value on error
869 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
871 struct ptlrpc_request *req = tgt_ses_req(tsi);
872 struct ost_body *body = NULL, *repbody;
873 void *key, *val = NULL;
874 int keylen, vallen, rc = 0;
875 bool is_grant_shrink;
879 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
881 DEBUG_REQ(D_HA, req, "no set_info key");
882 RETURN(err_serious(-EFAULT));
884 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
887 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
889 DEBUG_REQ(D_HA, req, "no set_info val");
890 RETURN(err_serious(-EFAULT));
892 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
895 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
897 /* In this case the value is actually an RMF_OST_BODY, so we
898 * transmutate the type of this PTLRPC */
899 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
901 rc = req_capsule_server_pack(tsi->tsi_pill);
905 if (is_grant_shrink) {
906 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
908 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
911 /** handle grant shrink, similar to a read request */
912 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
914 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
916 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
919 CERROR("%s: Unsupported key %s\n",
920 tgt_name(tsi->tsi_tgt), (char *)key);
923 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
930 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
932 * This function returns a list of extents which describes how a file's
933 * blocks are laid out on the disk.
935 * \param[in] env execution environment
936 * \param[in] ofd OFD device
937 * \param[in] fid FID of object
938 * \param[in] fiemap fiemap structure to fill with data
940 * \retval 0 if \a fiemap is filled with data successfully
941 * \retval negative value on error
943 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
944 struct lu_fid *fid, struct fiemap *fiemap)
946 struct ofd_object *fo;
949 fo = ofd_object_find(env, ofd, fid);
951 CERROR("%s: error finding object "DFID"\n",
952 ofd_name(ofd), PFID(fid));
956 ofd_read_lock(env, fo);
957 if (ofd_object_exists(fo))
958 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
961 ofd_read_unlock(env, fo);
962 ofd_object_put(env, fo);
967 static int ofd_lock_unlock_region(const struct lu_env *env,
968 struct ldlm_namespace *ns,
969 struct ldlm_res_id *res_id,
970 unsigned long long begin,
971 unsigned long long end)
975 struct lustre_handle lh = { 0 };
977 LASSERT(begin <= end);
979 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
983 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
984 tgt_extent_unlock(&lh, LCK_PR);
990 * Lock the sparse areas of given resource.
992 * The locking of sparse areas will cause dirty data to be flushed back from
993 * clients. This is used when getting the FIEMAP of an object to make sure
994 * there is no unaccounted cached data on clients.
996 * This function goes through \a fiemap list of extents and locks only sparse
997 * areas between extents.
999 * \param[in] ns LDLM namespace
1000 * \param[in] res_id resource ID
1001 * \param[in] fiemap file extents mapping on disk
1002 * \param[in] locked list head of regions list
1004 * \retval 0 if successful
1005 * \retval negative value on error
1007 static int lock_zero_regions(const struct lu_env *env,
1008 struct ldlm_namespace *ns,
1009 struct ldlm_res_id *res_id,
1010 struct fiemap *fiemap)
1012 __u64 begin = fiemap->fm_start;
1015 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1019 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1020 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1021 if (fiemap_start[i].fe_logical > begin) {
1022 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1023 begin, fiemap_start[i].fe_logical);
1024 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1025 fiemap_start[i].fe_logical);
1030 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1033 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1034 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1035 begin, fiemap->fm_start + fiemap->fm_length);
1036 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1037 fiemap->fm_start + fiemap->fm_length);
1045 * OFD request handler for OST_GET_INFO RPC.
1047 * This is OFD-specific part of request handling. The OFD-specific keys are:
1048 * - KEY_LAST_ID (obsolete)
1052 * This function reads needed data from storage and fills reply with it.
1054 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1055 * and is kept for compatibility.
1057 * \param[in] tsi target session environment for this request
1059 * \retval 0 if successful
1060 * \retval negative value on error
1062 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1064 struct obd_export *exp = tsi->tsi_exp;
1065 struct ofd_device *ofd = ofd_exp(exp);
1066 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1069 int replylen, rc = 0;
1073 /* this common part for get_info rpc */
1074 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1076 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1077 RETURN(err_serious(-EPROTO));
1079 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1082 if (KEY_IS(KEY_LAST_ID)) {
1084 struct ofd_seq *oseq;
1086 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1087 rc = req_capsule_server_pack(tsi->tsi_pill);
1089 RETURN(err_serious(rc));
1091 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1093 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1094 (u64)exp->exp_filter_data.fed_group);
1098 *last_id = ofd_seq_last_oid(oseq);
1099 ofd_seq_put(tsi->tsi_env, oseq);
1100 } else if (KEY_IS(KEY_FIEMAP)) {
1101 struct ll_fiemap_info_key *fm_key;
1102 struct fiemap *fiemap;
1105 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1107 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1108 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1110 RETURN(err_serious(rc));
1112 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1114 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1116 replylen = fiemap_count_to_size(
1117 fm_key->lfik_fiemap.fm_extent_count);
1118 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1119 RCL_SERVER, replylen);
1121 rc = req_capsule_server_pack(tsi->tsi_pill);
1123 RETURN(err_serious(rc));
1125 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1129 *fiemap = fm_key->lfik_fiemap;
1130 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1132 /* LU-3219: Lock the sparse areas to make sure dirty
1133 * flushed back from client, then call fiemap again. */
1134 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1135 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1136 ost_fid_build_resid(fid, &fti->fti_resid);
1137 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1138 &fti->fti_resid, fiemap);
1140 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1143 } else if (KEY_IS(KEY_LAST_FID)) {
1144 struct ofd_device *ofd = ofd_exp(exp);
1145 struct ofd_seq *oseq;
1149 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1150 rc = req_capsule_server_pack(tsi->tsi_pill);
1152 RETURN(err_serious(rc));
1154 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1156 RETURN(err_serious(-EPROTO));
1158 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1160 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1164 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1165 ostid_seq(&fti->fti_ostid));
1167 RETURN(PTR_ERR(oseq));
1169 rc = ostid_to_fid(fid, &oseq->os_oi,
1170 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1174 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1177 ofd_seq_put(tsi->tsi_env, oseq);
1179 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1183 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1190 * OFD request handler for OST_GETATTR RPC.
1192 * This is OFD-specific part of request handling. It finds the OFD object
1193 * by its FID, gets attributes from storage and packs result to the reply.
1195 * \param[in] tsi target session environment for this request
1197 * \retval 0 if successful
1198 * \retval negative value on error
1200 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1202 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1203 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1204 struct ost_body *repbody;
1205 struct lustre_handle lh = { 0 };
1206 struct ofd_object *fo;
1208 enum ldlm_mode lock_mode = LCK_PR;
1213 LASSERT(tsi->tsi_ost_body != NULL);
1215 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1216 if (repbody == NULL)
1219 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1220 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1222 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1223 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1226 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1229 rc = tgt_extent_lock(tsi->tsi_env,
1230 tsi->tsi_tgt->lut_obd->obd_namespace,
1231 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1237 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1239 GOTO(out, rc = PTR_ERR(fo));
1241 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1245 obdo_from_la(&repbody->oa, &fti->fti_attr,
1246 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1248 /* Store object version in reply */
1249 curr_version = dt_version_get(tsi->tsi_env,
1250 ofd_object_child(fo));
1251 if ((__s64)curr_version != -EOPNOTSUPP) {
1252 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1253 repbody->oa.o_data_version = curr_version;
1256 if (fo->ofo_ff.ff_layout_version > 0) {
1257 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1258 repbody->oa.o_layout_version =
1259 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1261 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1262 PFID(&tsi->tsi_fid),
1263 repbody->oa.o_layout_version);
1267 ofd_object_put(tsi->tsi_env, fo);
1270 tgt_extent_unlock(&lh, lock_mode);
1272 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1275 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1276 repbody->oa.o_flags = OBD_FL_FLUSH;
1282 * OFD request handler for OST_SETATTR RPC.
1284 * This is OFD-specific part of request handling. It finds the OFD object
1285 * by its FID, sets attributes from request and packs result to the reply.
1287 * \param[in] tsi target session environment for this request
1289 * \retval 0 if successful
1290 * \retval negative value on error
1292 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1294 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1295 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1296 struct ost_body *body = tsi->tsi_ost_body;
1297 struct ost_body *repbody;
1298 struct ldlm_resource *res;
1299 struct ofd_object *fo;
1304 LASSERT(body != NULL);
1306 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1307 if (repbody == NULL)
1310 repbody->oa.o_oi = body->oa.o_oi;
1311 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1313 /* This would be very bad - accidentally truncating a file when
1314 * changing the time or similar - bug 12203. */
1315 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1316 body->oa.o_size != OBD_OBJECT_EOF) {
1317 static char mdsinum[48];
1319 if (body->oa.o_valid & OBD_MD_FLFID)
1320 snprintf(mdsinum, sizeof(mdsinum) - 1,
1321 "of parent "DFID, body->oa.o_parent_seq,
1322 body->oa.o_parent_oid, 0);
1326 CERROR("%s: setattr from %s is trying to truncate object "DFID
1327 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1328 PFID(&tsi->tsi_fid), mdsinum);
1332 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1334 GOTO(out, rc = PTR_ERR(fo));
1336 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1337 fti->fti_attr.la_valid &= ~LA_TYPE;
1339 /* setting objects attributes (including owner/group) */
1340 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1344 obdo_from_la(&repbody->oa, &fti->fti_attr,
1345 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1347 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1351 ofd_object_put(tsi->tsi_env, fo);
1354 /* we do not call this before to avoid lu_object_find() in
1355 * ->lvbo_update() holding another reference on the object.
1356 * otherwise concurrent destroy can make the object unavailable
1357 * for 2nd lu_object_find() waiting for the first reference
1358 * to go... deadlock! */
1359 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1360 &tsi->tsi_resid, LDLM_EXTENT, 0);
1362 ldlm_res_lvbo_update(tsi->tsi_env, res, NULL, 0);
1363 ldlm_resource_putref(res);
1370 * Destroy OST orphans.
1372 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1373 * set then we must destroy possible orphaned objects.
1375 * \param[in] env execution environment
1376 * \param[in] exp OBD export
1377 * \param[in] ofd OFD device
1378 * \param[in] oa obdo structure for reply
1380 * \retval 0 if successful
1381 * \retval negative value on error
1383 static int ofd_orphans_destroy(const struct lu_env *env,
1384 struct obd_export *exp,
1385 struct ofd_device *ofd, struct obdo *oa)
1387 struct ofd_thread_info *info = ofd_info(env);
1388 struct lu_fid *fid = &info->fti_fid;
1389 struct ost_id *oi = &oa->o_oi;
1390 struct ofd_seq *oseq;
1391 u64 seq = ostid_seq(oi);
1392 u64 end_id = ostid_id(oi);
1400 oseq = ofd_seq_get(ofd, seq);
1402 CERROR("%s: Can not find seq for "DOSTID"\n",
1403 ofd_name(ofd), POSTID(oi));
1408 last = ofd_seq_last_oid(oseq);
1411 LASSERT(exp != NULL);
1412 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1414 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1417 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1418 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1420 while (oid > end_id) {
1421 rc = fid_set_id(fid, oid);
1422 if (unlikely(rc != 0))
1425 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1426 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1427 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1428 /* this is pretty fatal... */
1429 CEMERG("%s: error destroying precreated id "
1431 ofd_name(ofd), PFID(fid), rc);
1435 ofd_seq_last_oid_set(oseq, oid);
1436 /* update last_id on disk periodically so that if we
1437 * restart * we don't need to re-scan all of the just
1438 * deleted objects. */
1439 if ((oid & 511) == 0)
1440 ofd_seq_last_oid_write(env, ofd, oseq);
1444 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1445 ofd_name(ofd), seq, oid);
1449 ofd_seq_last_oid_set(oseq, oid);
1450 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1452 /* don't reuse orphan object, return last used objid */
1453 rc = ostid_set_id(oi, last);
1459 ofd_seq_put(env, oseq);
1464 * OFD request handler for OST_CREATE RPC.
1466 * This is OFD-specific part of request handling. Its main purpose is to
1467 * create new data objects on OST, but it also used to destroy orphans.
1469 * \param[in] tsi target session environment for this request
1471 * \retval 0 if successful
1472 * \retval negative value on error
1474 static int ofd_create_hdl(struct tgt_session_info *tsi)
1476 struct ptlrpc_request *req = tgt_ses_req(tsi);
1477 struct ost_body *repbody;
1478 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1479 struct obdo *rep_oa;
1480 struct obd_export *exp = tsi->tsi_exp;
1481 struct ofd_device *ofd = ofd_exp(exp);
1482 u64 seq = ostid_seq(&oa->o_oi);
1483 u64 oid = ostid_id(&oa->o_oi);
1484 struct ofd_seq *oseq;
1492 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1495 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1496 if (repbody == NULL)
1499 down_read(&ofd->ofd_lastid_rwsem);
1500 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1501 * we may do that in the future.
1502 * Return -ENOSPC until the LAST_ID rebuilt. */
1503 if (unlikely(ofd->ofd_lastid_rebuilding))
1504 GOTO(out_sem, rc = -ENOSPC);
1506 rep_oa = &repbody->oa;
1507 rep_oa->o_oi = oa->o_oi;
1509 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1511 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1513 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1515 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1516 ofd_name(ofd), seq, PTR_ERR(oseq));
1517 GOTO(out_sem, rc = -EINVAL);
1520 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1521 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1522 if (!ofd_obd(ofd)->obd_recovering ||
1523 oid > ofd_seq_last_oid(oseq)) {
1524 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1525 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1526 ofd_seq_last_oid(oseq));
1527 GOTO(out_nolock, rc = -EINVAL);
1529 /* Do nothing here, we re-create objects during recovery
1530 * upon write replay, see ofd_preprw_write() */
1531 GOTO(out_nolock, rc = 0);
1533 /* former ofd_handle_precreate */
1534 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1535 (oa->o_flags & OBD_FL_DELORPHAN)) {
1536 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1538 /* destroy orphans */
1539 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1540 exp->exp_conn_cnt) {
1541 CERROR("%s: dropping old orphan cleanup request\n",
1543 GOTO(out_nolock, rc = 0);
1545 /* This causes inflight precreates to abort and drop lock */
1546 oseq->os_destroys_in_progress = 1;
1547 mutex_lock(&oseq->os_create_lock);
1548 if (!oseq->os_destroys_in_progress) {
1550 "%s:[%llu] destroys_in_progress already cleared\n",
1551 ofd_name(ofd), seq);
1552 rc = ostid_set_id(&rep_oa->o_oi,
1553 ofd_seq_last_oid(oseq));
1556 diff = oid - ofd_seq_last_oid(oseq);
1557 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1558 ofd_seq_last_oid(oseq), diff);
1559 if (-diff > OST_MAX_PRECREATE) {
1560 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1561 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1562 "(%llu), trust the OST\n",
1563 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1564 PFID(&oseq->os_oi.oi_fid),
1565 ofd_seq_last_oid(oseq));
1567 /* Let MDS know that we are so far ahead. */
1568 rc = ostid_set_id(&rep_oa->o_oi,
1569 ofd_seq_last_oid(oseq) + 1);
1570 } else if (diff < 0) {
1571 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1573 oseq->os_destroys_in_progress = 0;
1575 /* XXX: Used by MDS for the first time! */
1576 oseq->os_destroys_in_progress = 0;
1579 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1580 ofd->ofd_lastid_gen)) {
1581 /* Keep the export ref so we can send the reply. */
1582 ofd_obd_disconnect(class_export_get(exp));
1583 GOTO(out_nolock, rc = -ENOTCONN);
1586 mutex_lock(&oseq->os_create_lock);
1587 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1588 exp->exp_conn_cnt) {
1589 CERROR("%s: dropping old precreate request\n",
1593 /* only precreate if seq is 0, IDIF or normal and also o_id
1594 * must be specfied */
1595 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1596 !fid_seq_is_idif(seq)) || oid == 0) {
1597 diff = 1; /* shouldn't we create this right now? */
1599 diff = oid - ofd_seq_last_oid(oseq);
1600 /* Do sync create if the seq is about to used up */
1601 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1602 if (unlikely(oid >= IDIF_MAX_OID - 1))
1604 } else if (fid_seq_is_norm(seq)) {
1606 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1609 CERROR("%s : invalid o_seq "DOSTID"\n",
1610 ofd_name(ofd), POSTID(&oa->o_oi));
1611 GOTO(out, rc = -EINVAL);
1616 CERROR("%s: invalid precreate request for "
1617 DOSTID", last_id %llu. "
1618 "Likely MDS last_id corruption\n",
1619 ofd_name(ofd), POSTID(&oa->o_oi),
1620 ofd_seq_last_oid(oseq));
1621 GOTO(out, rc = -EINVAL);
1626 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1632 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1633 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1634 /* don't enforce grant during orphan recovery */
1635 granted = tgt_grant_create(tsi->tsi_env,
1636 ofd_obd(ofd)->obd_self_export,
1641 CDEBUG(D_HA, "%s: failed to acquire grant "
1642 "space for precreate (%lld): rc = %d\n",
1643 ofd_name(ofd), diff, rc);
1648 /* This can happen if a new OST is formatted and installed
1649 * in place of an old one at the same index. Instead of
1650 * precreating potentially millions of deleted old objects
1651 * (possibly filling the OST), only precreate the last batch.
1652 * LFSCK will eventually clean up any orphans. LU-14 */
1653 if (diff > 5 * OST_MAX_PRECREATE) {
1654 diff = OST_MAX_PRECREATE / 2;
1655 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1656 "OST replaced or reformatted: "
1657 "LFSCK will clean up",
1660 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1661 "%u larger than the LAST_ID "DOSTID", only "
1662 "precreating the last %lld objects.\n",
1663 ofd_name(ofd), POSTID(&oa->o_oi),
1664 5 * OST_MAX_PRECREATE,
1665 POSTID(&oseq->os_oi), diff);
1666 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1670 next_id = ofd_seq_last_oid(oseq) + 1;
1671 count = ofd_precreate_batch(ofd, (int)diff);
1673 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1674 " at %llu\n", ofd_name(ofd),
1675 count, seq, next_id);
1677 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1678 && ktime_get_seconds() > enough_time) {
1679 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1680 " created at a rate of %d/s\n",
1681 ofd_name(ofd), created, diff + created,
1682 created / DISK_TIMEOUT);
1686 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1687 oseq, count, sync_trans);
1691 } else if (rc < 0) {
1697 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1698 LCONSOLE_WARN("%s: can't create the same count of"
1699 " objects when replaying the request"
1700 " (diff is %lld). see LU-4621\n",
1701 ofd_name(ofd), diff);
1704 /* some objects got created, we can return
1705 * them, even if last creation failed */
1708 CERROR("%s: unable to precreate: rc = %d\n",
1711 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1712 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1713 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1718 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1722 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1724 if (unlikely(!oseq->os_last_id_synced))
1725 oseq->os_last_id_synced = 1;
1727 mutex_unlock(&oseq->os_create_lock);
1730 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1731 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1732 struct lu_fid *fid = &info->fti_fid;
1734 /* For compatible purpose, it needs to convert back to
1735 * OST ID before put it on wire. */
1736 *fid = rep_oa->o_oi.oi_fid;
1737 fid_to_ostid(fid, &rep_oa->o_oi);
1739 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1741 ofd_seq_put(tsi->tsi_env, oseq);
1744 up_read(&ofd->ofd_lastid_rwsem);
1749 * OFD request handler for OST_DESTROY RPC.
1751 * This is OFD-specific part of request handling. It destroys data objects
1752 * related to destroyed object on MDT.
1754 * \param[in] tsi target session environment for this request
1756 * \retval 0 if successful
1757 * \retval negative value on error
1759 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1761 const struct ost_body *body = tsi->tsi_ost_body;
1762 struct ost_body *repbody;
1763 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1764 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1765 struct lu_fid *fid = &fti->fti_fid;
1772 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1775 /* This is old case for clients before Lustre 2.4 */
1776 /* If there's a DLM request, cancel the locks mentioned in it */
1777 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1779 struct ldlm_request *dlm;
1781 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1784 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1787 *fid = body->oa.o_oi.oi_fid;
1788 oid = ostid_id(&body->oa.o_oi);
1791 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1793 /* check that o_misc makes sense */
1794 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1795 count = body->oa.o_misc;
1797 count = 1; /* default case - single destroy */
1799 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1800 POSTID(&body->oa.o_oi), count);
1805 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1806 if (lrc == -ENOENT) {
1808 "%s: destroying non-existent object "DFID"\n",
1809 ofd_name(ofd), PFID(fid));
1810 /* rewrite rc with -ENOENT only if it is 0 */
1813 } else if (lrc != 0) {
1814 CERROR("%s: error destroying object "DFID": %d\n",
1815 ofd_name(ofd), PFID(fid), lrc);
1821 lrc = fid_set_id(fid, oid);
1822 if (unlikely(lrc != 0 && count > 0))
1823 GOTO(out, rc = lrc);
1826 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1832 fid_to_ostid(fid, &repbody->oa.o_oi);
1837 * OFD request handler for OST_STATFS RPC.
1839 * This function gets statfs data from storage as part of request
1842 * \param[in] tsi target session environment for this request
1844 * \retval 0 if successful
1845 * \retval negative value on error
1847 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1849 struct obd_statfs *osfs;
1854 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1856 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1858 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1859 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1861 CERROR("%s: statfs failed: rc = %d\n",
1862 tgt_name(tsi->tsi_tgt), rc);
1864 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1867 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1874 * OFD request handler for OST_SYNC RPC.
1876 * Sync object data or all filesystem data to the disk and pack the
1879 * \param[in] tsi target session environment for this request
1881 * \retval 0 if successful
1882 * \retval negative value on error
1884 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1886 struct ost_body *body = tsi->tsi_ost_body;
1887 struct ost_body *repbody;
1888 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1889 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1890 struct ofd_object *fo = NULL;
1895 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1897 /* if no objid is specified, it means "sync whole filesystem" */
1898 if (!fid_is_zero(&tsi->tsi_fid)) {
1899 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1901 RETURN(PTR_ERR(fo));
1904 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1905 fo != NULL ? ofd_object_child(fo) : NULL,
1906 repbody->oa.o_size, repbody->oa.o_blocks);
1910 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1915 repbody->oa.o_oi = body->oa.o_oi;
1916 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1918 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1920 obdo_from_la(&repbody->oa, &fti->fti_attr,
1923 /* don't return rc from getattr */
1928 ofd_object_put(tsi->tsi_env, fo);
1933 * OFD request handler for OST_PUNCH RPC.
1935 * This is part of request processing. Validate request fields,
1936 * punch (truncate) the given OFD object and pack reply.
1938 * \param[in] tsi target session environment for this request
1940 * \retval 0 if successful
1941 * \retval negative value on error
1943 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1945 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1946 struct ost_body *repbody;
1947 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1948 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1949 struct ldlm_resource *res;
1950 struct ofd_object *fo;
1952 struct lustre_handle lh = { 0, };
1959 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
1961 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1962 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
1964 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1965 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1966 RETURN(err_serious(-EPROTO));
1968 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1969 if (repbody == NULL)
1970 RETURN(err_serious(-ENOMEM));
1972 /* punch start,end are passed in o_size,o_blocks throught wire */
1976 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1979 /* standard truncate optimization: if file body is completely
1980 * destroyed, don't send data back to the server. */
1982 flags |= LDLM_FL_AST_DISCARD_DATA;
1984 repbody->oa.o_oi = oa->o_oi;
1985 repbody->oa.o_valid = OBD_MD_FLID;
1987 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1988 oa->o_flags & OBD_FL_SRVLOCK;
1991 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
1992 end, &lh, LCK_PW, &flags);
1997 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
1998 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
1999 oa->o_valid, start, end);
2001 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2004 GOTO(out, rc = PTR_ERR(fo));
2006 la_from_obdo(&info->fti_attr, oa,
2007 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2008 info->fti_attr.la_size = start;
2009 info->fti_attr.la_valid |= LA_SIZE;
2011 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2016 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2020 ofd_object_put(tsi->tsi_env, fo);
2023 tgt_extent_unlock(&lh, LCK_PW);
2025 /* we do not call this before to avoid lu_object_find() in
2026 * ->lvbo_update() holding another reference on the object.
2027 * otherwise concurrent destroy can make the object unavailable
2028 * for 2nd lu_object_find() waiting for the first reference
2029 * to go... deadlock! */
2030 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2033 struct ost_lvb *res_lvb;
2035 ldlm_res_lvbo_update(tsi->tsi_env, res, NULL, 0);
2036 res_lvb = res->lr_lvb_data;
2037 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2038 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2039 ldlm_resource_putref(res);
2045 static int ofd_ladvise_prefetch(const struct lu_env *env,
2046 struct ofd_object *fo,
2047 struct niobuf_local *lnb,
2048 __u64 start, __u64 end, enum dt_bufs_type dbt)
2050 struct ofd_thread_info *info = ofd_info(env);
2051 pgoff_t start_index, end_index, pages;
2052 struct niobuf_remote rnb;
2053 unsigned long nr_local;
2059 ofd_read_lock(env, fo);
2060 if (!ofd_object_exists(fo))
2061 GOTO(out_unlock, rc = -ENOENT);
2063 rc = ofd_attr_get(env, fo, &info->fti_attr);
2065 GOTO(out_unlock, rc);
2067 if (end > info->fti_attr.la_size)
2068 end = info->fti_attr.la_size;
2071 GOTO(out_unlock, rc);
2073 /* We need page aligned offset and length */
2074 start_index = start >> PAGE_SHIFT;
2075 end_index = (end - 1) >> PAGE_SHIFT;
2076 pages = end_index - start_index + 1;
2078 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2079 PTLRPC_MAX_BRW_PAGES;
2080 rnb.rnb_offset = start_index << PAGE_SHIFT;
2081 rnb.rnb_len = nr_local << PAGE_SHIFT;
2082 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, dbt);
2083 if (unlikely(rc < 0))
2086 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2087 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2090 start_index += nr_local;
2095 ofd_read_unlock(env, fo);
2100 * OFD request handler for OST_LADVISE RPC.
2102 * Tune cache or perfetch policies according to advices.
2104 * \param[in] tsi target session environment for this request
2106 * \retval 0 if successful
2107 * \retval negative errno on error
2109 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2111 struct ptlrpc_request *req = tgt_ses_req(tsi);
2112 struct obd_export *exp = tsi->tsi_exp;
2113 struct ofd_device *ofd = ofd_exp(exp);
2114 struct ost_body *body, *repbody;
2115 struct ofd_thread_info *info;
2116 struct ofd_object *fo;
2117 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2118 const struct lu_env *env = svc_thread->t_env;
2119 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2120 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2121 struct lu_ladvise *ladvise;
2123 struct ladvise_hdr *ladvise_hdr;
2124 struct obd_ioobj ioo;
2125 struct lustre_handle lockh = { 0 };
2128 struct dt_object *dob;
2134 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2135 body = tsi->tsi_ost_body;
2137 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2138 RETURN(err_serious(-EPROTO));
2140 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2141 &RMF_OST_LADVISE_HDR);
2142 if (ladvise_hdr == NULL)
2143 RETURN(err_serious(-EPROTO));
2145 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2146 ladvise_hdr->lah_count < 1)
2147 RETURN(err_serious(-EPROTO));
2149 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2150 RETURN(err_serious(-EPROTO));
2152 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2153 if (ladvise == NULL)
2154 RETURN(err_serious(-EPROTO));
2156 num_advise = req_capsule_get_size(&req->rq_pill,
2157 &RMF_OST_LADVISE, RCL_CLIENT) /
2159 if (num_advise < ladvise_hdr->lah_count)
2160 RETURN(err_serious(-EPROTO));
2162 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2163 repbody->oa = body->oa;
2165 info = ofd_info_init(env, exp);
2167 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2168 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2172 fo = ofd_object_find(env, ofd, &info->fti_fid);
2177 LASSERT(fo != NULL);
2178 dob = ofd_object_child(fo);
2180 if (ptlrpc_connection_is_local(exp->exp_connection))
2181 dbt |= DT_BUFS_TYPE_LOCAL;
2183 for (i = 0; i < num_advise; i++, ladvise++) {
2184 start = ladvise->lla_start;
2185 end = ladvise->lla_end;
2187 rc = err_serious(-EPROTO);
2191 /* Handle different advice types */
2192 switch (ladvise->lla_advice) {
2196 case LU_LADVISE_WILLREAD:
2200 ioo.ioo_oid = body->oa.o_oi;
2202 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2203 &tsi->tsi_resid, start, end - 1,
2204 &lockh, LCK_PR, &flags);
2208 req->rq_status = ofd_ladvise_prefetch(env, fo,
2211 tgt_extent_unlock(&lockh, LCK_PR);
2213 case LU_LADVISE_DONTNEED:
2214 rc = dt_ladvise(env, dob, ladvise->lla_start,
2215 ladvise->lla_end, LU_LADVISE_DONTNEED);
2222 ofd_object_put(env, fo);
2223 req->rq_status = rc;
2228 * OFD request handler for OST_QUOTACTL RPC.
2230 * This is part of request processing to validate incoming request fields,
2231 * get the requested data from OSD and pack reply.
2233 * \param[in] tsi target session environment for this request
2235 * \retval 0 if successful
2236 * \retval negative value on error
2238 static int ofd_quotactl(struct tgt_session_info *tsi)
2240 struct obd_quotactl *oqctl, *repoqc;
2241 struct lu_nodemap *nodemap;
2247 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2249 RETURN(err_serious(-EPROTO));
2251 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2253 RETURN(err_serious(-ENOMEM));
2257 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2258 if (IS_ERR(nodemap))
2259 RETURN(PTR_ERR(nodemap));
2262 if (oqctl->qc_type == USRQUOTA)
2263 id = nodemap_map_id(nodemap, NODEMAP_UID,
2264 NODEMAP_CLIENT_TO_FS,
2266 else if (oqctl->qc_type == GRPQUOTA)
2267 id = nodemap_map_id(nodemap, NODEMAP_GID,
2268 NODEMAP_CLIENT_TO_FS,
2271 nodemap_putref(nodemap);
2273 if (repoqc->qc_id != id)
2274 swap(repoqc->qc_id, id);
2276 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2278 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2281 if (repoqc->qc_id != id)
2282 swap(repoqc->qc_id, id);
2288 * Calculate the amount of time for lock prolongation.
2290 * This is helper for ofd_prolong_extent_locks() function to get
2291 * the timeout extra time.
2293 * \param[in] req current request
2295 * \retval amount of time to extend the timeout with
2297 static inline time64_t prolong_timeout(struct ptlrpc_request *req)
2299 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2300 time64_t req_timeout;
2303 return obd_timeout / 2;
2305 req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec;
2306 return max_t(time64_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
2311 * Prolong lock timeout for the given extent.
2313 * This function finds all locks related with incoming request and
2314 * prolongs their timeout.
2316 * If a client is holding a lock for a long time while it sends
2317 * read or write RPCs to the OST for the object under this lock,
2318 * then we don't want the OST to evict the client. Otherwise,
2319 * if the network or disk is very busy then the client may not
2320 * be able to make any progress to clear out dirty pages under
2321 * the lock and the application will fail.
2323 * Every time a Bulk Read/Write (BRW) request arrives for the object
2324 * covered by the lock, extend the timeout on that lock. The RPC should
2325 * contain a lock handle for the lock it is using, but this
2326 * isn't handled correctly by all client versions, and the
2327 * request may cover multiple locks.
2329 * \param[in] tsi target session environment for this request
2330 * \param[in] data struct of data to prolong locks
2333 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2334 struct ldlm_prolong_args *data)
2336 struct obdo *oa = &tsi->tsi_ost_body->oa;
2337 struct ldlm_lock *lock;
2341 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2342 data->lpa_export = tsi->tsi_exp;
2343 data->lpa_resid = tsi->tsi_resid;
2345 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2346 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2347 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2348 data->lpa_extent.end);
2350 if (oa->o_valid & OBD_MD_FLHANDLE) {
2351 /* mostly a request should be covered by only one lock, try
2353 lock = ldlm_handle2lock(&oa->o_handle);
2355 /* Fast path to check if the lock covers the whole IO
2356 * region exclusively. */
2357 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2358 &data->lpa_extent)) {
2360 LASSERT(lock->l_export == data->lpa_export);
2361 ldlm_lock_prolong_one(lock, data);
2362 LDLM_LOCK_PUT(lock);
2363 if (data->lpa_locks_cnt > 0)
2365 /* The lock was destroyed probably lets try
2368 lock->l_last_used = ktime_get();
2369 LDLM_LOCK_PUT(lock);
2374 ldlm_resource_prolong(data);
2379 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2381 * Determine if \a lock and the lock from request \a req are equivalent
2382 * by comparing their resource names, modes, and extents.
2384 * It is used to give priority to read and write RPCs being done
2385 * under this lock so that the client can drop the contended
2386 * lock more quickly and let other clients use it. This improves
2387 * overall performance in the case where the first client gets a
2388 * very large lock extent that prevents other clients from
2389 * submitting their writes.
2391 * \param[in] req ptlrpc_request being processed
2392 * \param[in] lock contended lock to match
2394 * \retval 1 if lock is matched
2395 * \retval 0 otherwise
2397 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2398 struct ldlm_lock *lock)
2400 struct niobuf_remote *rnb;
2401 struct obd_ioobj *ioo;
2402 enum ldlm_mode mode;
2403 struct ldlm_extent ext;
2404 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2408 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2409 LASSERT(ioo != NULL);
2411 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2412 LASSERT(rnb != NULL);
2414 ext.start = rnb->rnb_offset;
2415 rnb += ioo->ioo_bufcnt - 1;
2416 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2418 LASSERT(lock->l_resource != NULL);
2419 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2422 /* a bulk write can only hold a reference on a PW extent lock
2425 mode = LCK_PW | LCK_GROUP;
2426 if (opc == OST_READ)
2427 /* whereas a bulk read can be protected by either a PR or PW
2431 if (!(lock->l_granted_mode & mode))
2434 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2438 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2440 * Check for whether the given PTLRPC request (\a req) is blocking
2441 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2444 * \param[in] req the incoming request
2446 * \retval 1 if \a req is blocking an LDLM lock cancel
2447 * \retval 0 if it is not
2448 * \retval -ESTALE if lock is not found
2450 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2452 struct tgt_session_info *tsi;
2453 struct obd_ioobj *ioo;
2454 struct niobuf_remote *rnb;
2456 struct ldlm_prolong_args pa = { 0 };
2460 /* Don't use tgt_ses_info() to get session info, because lock_match()
2461 * can be called while request has no processing thread yet. */
2462 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2465 * Use LASSERT below because malformed RPCs should have
2466 * been filtered out in tgt_hpreq_handler().
2468 opc = lustre_msg_get_opc(req->rq_reqmsg);
2469 LASSERT(opc == OST_READ || opc == OST_WRITE);
2471 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2472 LASSERT(ioo != NULL);
2474 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2475 LASSERT(rnb != NULL);
2476 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2478 pa.lpa_mode = LCK_PW | LCK_GROUP;
2479 if (opc == OST_READ)
2480 pa.lpa_mode |= LCK_PR;
2482 pa.lpa_extent.start = rnb->rnb_offset;
2483 rnb += ioo->ioo_bufcnt - 1;
2484 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2486 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
2487 " (%llu->%llu)\n", tgt_name(tsi->tsi_tgt),
2488 current->comm, PFID(&tsi->tsi_fid), pa.lpa_extent.start,
2491 ofd_prolong_extent_locks(tsi, &pa);
2493 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2494 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2496 if (pa.lpa_blocks_cnt > 0)
2499 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2503 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2505 * Called after the request has been handled. It refreshes lock timeout again
2506 * so that client has more time to send lock cancel RPC.
2508 * \param[in] req request which is being processed.
2510 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2512 ofd_rw_hpreq_check(req);
2516 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2518 * This function checks if the given lock is the same by its resname, mode
2519 * and extent as one taken from the request.
2520 * It is used to give priority to punch/truncate RPCs that might lead to
2521 * the fastest release of that lock when a lock is contended.
2523 * \param[in] req ptlrpc_request being processed
2524 * \param[in] lock contended lock to match
2526 * \retval 1 if lock is matched
2527 * \retval 0 otherwise
2529 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2530 struct ldlm_lock *lock)
2532 struct tgt_session_info *tsi;
2534 struct ldlm_extent ext;
2538 /* Don't use tgt_ses_info() to get session info, because lock_match()
2539 * can be called while request has no processing thread yet. */
2540 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2543 * Use LASSERT below because malformed RPCs should have
2544 * been filtered out in tgt_hpreq_handler().
2546 LASSERT(tsi->tsi_ost_body != NULL);
2547 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2548 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2551 oa = &tsi->tsi_ost_body->oa;
2552 ext.start = oa->o_size;
2553 ext.end = oa->o_blocks;
2555 LASSERT(lock->l_resource != NULL);
2556 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2559 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2562 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2566 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2568 * High-priority queue request check for whether the given punch request
2569 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2570 * covered by an LDLM lock.
2574 * \param[in] req the incoming request
2576 * \retval 1 if \a req is blocking an LDLM lock cancel
2577 * \retval 0 if it is not
2578 * \retval -ESTALE if lock is not found
2580 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2582 struct tgt_session_info *tsi;
2584 struct ldlm_prolong_args pa = { 0 };
2588 /* Don't use tgt_ses_info() to get session info, because lock_match()
2589 * can be called while request has no processing thread yet. */
2590 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2591 LASSERT(tsi != NULL);
2592 oa = &tsi->tsi_ost_body->oa;
2594 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2595 oa->o_flags & OBD_FL_SRVLOCK));
2597 pa.lpa_mode = LCK_PW | LCK_GROUP;
2598 pa.lpa_extent.start = oa->o_size;
2599 pa.lpa_extent.end = oa->o_blocks;
2602 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2603 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2604 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2606 ofd_prolong_extent_locks(tsi, &pa);
2608 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2609 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2611 if (pa.lpa_blocks_cnt > 0)
2614 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2618 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2620 * Called after the request has been handled. It refreshes lock timeout again
2621 * so that client has more time to send lock cancel RPC.
2623 * \param[in] req request which is being processed.
2625 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2627 ofd_punch_hpreq_check(req);
2630 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2631 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2632 .hpreq_check = ofd_rw_hpreq_check,
2633 .hpreq_fini = ofd_rw_hpreq_fini
2636 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2637 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2638 .hpreq_check = ofd_punch_hpreq_check,
2639 .hpreq_fini = ofd_punch_hpreq_fini
2643 * Assign high priority operations to an IO request.
2645 * Check if the incoming request is a candidate for
2646 * high-priority processing. If it is, assign it a high
2647 * priority operations table.
2649 * \param[in] tsi target session environment for this request
2651 static void ofd_hp_brw(struct tgt_session_info *tsi)
2653 struct niobuf_remote *rnb;
2654 struct obd_ioobj *ioo;
2658 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2659 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2660 if (ioo->ioo_bufcnt > 0) {
2661 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2662 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2664 /* no high priority if server lock is needed */
2665 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2666 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2670 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2674 * Assign high priority operations to an punch request.
2676 * Check if the incoming request is a candidate for
2677 * high-priority processing. If it is, assign it a high
2678 * priority operations table.
2680 * \param[in] tsi target session environment for this request
2682 static void ofd_hp_punch(struct tgt_session_info *tsi)
2684 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2685 /* no high-priority if server lock is needed */
2686 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2687 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2688 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2689 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2691 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2694 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2695 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2696 #define OST_BRW_READ OST_READ
2697 #define OST_BRW_WRITE OST_WRITE
2700 * Table of OFD-specific request handlers
2702 * This table contains all opcodes accepted by OFD and
2703 * specifies handlers for them. The tgt_request_handler()
2704 * uses such table from each target to process incoming
2707 static struct tgt_handler ofd_tgt_handlers[] = {
2708 TGT_RPC_HANDLER(OST_FIRST_OPC,
2709 0, OST_CONNECT, tgt_connect,
2710 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2711 TGT_RPC_HANDLER(OST_FIRST_OPC,
2712 0, OST_DISCONNECT, tgt_disconnect,
2713 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2714 TGT_RPC_HANDLER(OST_FIRST_OPC,
2715 0, OST_SET_INFO, ofd_set_info_hdl,
2716 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2717 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2718 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl),
2719 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2720 OST_SETATTR, ofd_setattr_hdl),
2721 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2722 OST_CREATE, ofd_create_hdl),
2723 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2724 OST_DESTROY, ofd_destroy_hdl),
2725 TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl),
2726 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
2728 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2729 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write,
2731 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2732 OST_PUNCH, ofd_punch_hdl,
2734 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl),
2735 TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl),
2736 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl),
2739 static struct tgt_opc_slice ofd_common_slice[] = {
2741 .tos_opc_start = OST_FIRST_OPC,
2742 .tos_opc_end = OST_LAST_OPC,
2743 .tos_hs = ofd_tgt_handlers
2746 .tos_opc_start = OBD_FIRST_OPC,
2747 .tos_opc_end = OBD_LAST_OPC,
2748 .tos_hs = tgt_obd_handlers
2751 .tos_opc_start = LDLM_FIRST_OPC,
2752 .tos_opc_end = LDLM_LAST_OPC,
2753 .tos_hs = tgt_dlm_handlers
2756 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2757 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2758 .tos_hs = tgt_out_handlers
2761 .tos_opc_start = SEQ_FIRST_OPC,
2762 .tos_opc_end = SEQ_LAST_OPC,
2763 .tos_hs = seq_handlers
2766 .tos_opc_start = LFSCK_FIRST_OPC,
2767 .tos_opc_end = LFSCK_LAST_OPC,
2768 .tos_hs = tgt_lfsck_handlers
2771 .tos_opc_start = SEC_FIRST_OPC,
2772 .tos_opc_end = SEC_LAST_OPC,
2773 .tos_hs = tgt_sec_ctx_handlers
2780 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2781 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2784 * Implementation of lu_context_key::lct_key_exit.
2786 * Optional method called on lu_context_exit() for all allocated
2788 * It is used in OFD to sanitize context values which may be re-used
2789 * during another request processing by the same thread.
2791 * \param[in] ctx execution context
2792 * \param[in] key context key
2793 * \param[in] data ofd_thread_info
2795 static void ofd_key_exit(const struct lu_context *ctx,
2796 struct lu_context_key *key, void *data)
2798 struct ofd_thread_info *info = data;
2800 info->fti_env = NULL;
2801 info->fti_exp = NULL;
2804 info->fti_pre_version = 0;
2806 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2809 struct lu_context_key ofd_thread_key = {
2810 .lct_tags = LCT_DT_THREAD,
2811 .lct_init = ofd_key_init,
2812 .lct_fini = ofd_key_fini,
2813 .lct_exit = ofd_key_exit
2817 * Initialize OFD device according to parameters in the config log \a cfg.
2819 * This is the main starting point of OFD initialization. It fills all OFD
2820 * parameters with their initial values and calls other initializing functions
2821 * to set up all OFD subsystems.
2823 * \param[in] env execution environment
2824 * \param[in] m OFD device
2825 * \param[in] ldt LU device type of OFD
2826 * \param[in] cfg configuration log
2828 * \retval 0 if successful
2829 * \retval negative value on error
2831 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2832 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2834 const char *dev = lustre_cfg_string(cfg, 0);
2835 struct ofd_thread_info *info = NULL;
2836 struct obd_device *obd;
2837 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2839 struct nm_config_file *nodemap_config;
2840 struct obd_device_target *obt;
2845 obd = class_name2obd(dev);
2847 CERROR("Cannot find obd with name %s\n", dev);
2851 rc = lu_env_refill((struct lu_env *)env);
2856 obt->obt_magic = OBT_MAGIC;
2858 spin_lock_init(&m->ofd_flags_lock);
2859 m->ofd_raid_degraded = 0;
2860 m->ofd_checksum_t10pi_enforce = 0;
2861 m->ofd_sync_journal = 0;
2863 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2865 m->ofd_seq_count = 0;
2866 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2867 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2868 spin_lock_init(&m->ofd_inconsistency_lock);
2870 spin_lock_init(&m->ofd_batch_lock);
2871 init_rwsem(&m->ofd_lastid_rwsem);
2873 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2874 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2875 /* set this lu_device to obd, because error handling need it */
2876 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2878 rc = ofd_tunables_init(m);
2880 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2884 /* No connection accepted until configurations will finish */
2885 spin_lock(&obd->obd_dev_lock);
2886 obd->obd_no_conn = 1;
2887 spin_unlock(&obd->obd_dev_lock);
2888 obd->obd_replayable = 1;
2889 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2890 char *str = lustre_cfg_string(cfg, 4);
2892 if (strchr(str, 'n')) {
2893 CWARN("%s: recovery disabled\n", obd->obd_name);
2894 obd->obd_replayable = 0;
2898 info = ofd_info_init(env, NULL);
2900 GOTO(err_fini_proc, rc = -EFAULT);
2902 rc = ofd_stack_init(env, m, cfg);
2904 CERROR("Can't init device stack, rc %d\n", rc);
2905 GOTO(err_fini_proc, rc);
2908 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
2909 ofd_procfs_add_brw_stats_symlink(m);
2912 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2913 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2914 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2915 LDLM_NAMESPACE_SERVER,
2916 LDLM_NAMESPACE_GREEDY,
2918 if (m->ofd_namespace == NULL)
2919 GOTO(err_fini_stack, rc = -ENOMEM);
2920 /* set obd_namespace for compatibility with old code */
2921 obd->obd_namespace = m->ofd_namespace;
2922 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2923 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2924 m->ofd_namespace->ns_lvbp = m;
2926 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2927 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2929 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2930 OBD_FAIL_OST_ALL_REQUEST_NET,
2931 OBD_FAIL_OST_ALL_REPLY_NET);
2933 GOTO(err_free_ns, rc);
2935 tgd->tgd_reserved_pcnt = 0;
2937 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
2938 m->ofd_cksum_types_supported =
2939 obd_cksum_types_supported_server(obd->obd_name);
2940 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2941 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
2942 OFD_PRECREATE_SMALL_FS)
2943 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2945 rc = ofd_fs_setup(env, m, obd);
2947 GOTO(err_fini_lut, rc);
2949 fid.f_seq = FID_SEQ_LOCAL_NAME;
2952 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
2955 GOTO(err_fini_fs, rc);
2957 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
2959 if (IS_ERR(nodemap_config)) {
2960 rc = PTR_ERR(nodemap_config);
2962 GOTO(err_fini_los, rc);
2964 obt->obt_nodemap_config_file = nodemap_config;
2967 rc = ofd_start_inconsistency_verification_thread(m);
2969 GOTO(err_fini_nm, rc);
2971 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
2976 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
2977 obt->obt_nodemap_config_file = NULL;
2979 local_oid_storage_fini(env, m->ofd_los);
2982 ofd_fs_cleanup(env, m);
2984 tgt_fini(env, &m->ofd_lut);
2986 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
2987 obd->obd_namespace = m->ofd_namespace = NULL;
2989 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
2996 * Stop the OFD device
2998 * This function stops the OFD device and all its subsystems.
2999 * This is the end of OFD lifecycle.
3001 * \param[in] env execution environment
3002 * \param[in] m OFD device
3004 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3006 struct obd_device *obd = ofd_obd(m);
3007 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3008 struct lfsck_stop stop;
3010 stop.ls_status = LS_PAUSED;
3012 lfsck_stop(env, m->ofd_osd, &stop);
3013 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3014 target_recovery_fini(obd);
3015 if (m->ofd_namespace != NULL)
3016 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3017 d->ld_obd->obd_force);
3019 obd_exports_barrier(obd);
3020 obd_zombie_barrier();
3022 tgt_fini(env, &m->ofd_lut);
3023 ofd_stop_inconsistency_verification_thread(m);
3024 lfsck_degister(env, m->ofd_osd);
3025 ofd_fs_cleanup(env, m);
3026 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3027 obd->u.obt.obt_nodemap_config_file = NULL;
3029 if (m->ofd_namespace != NULL) {
3030 ldlm_namespace_free_post(m->ofd_namespace);
3031 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3034 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3036 LASSERT(atomic_read(&d->ld_ref) == 0);
3037 server_put_mount(obd->obd_name, true);
3042 * Implementation of lu_device_type_operations::ldto_device_fini.
3044 * Finalize device. Dual to ofd_device_init(). It is called from
3045 * obd_precleanup() and stops the current device.
3047 * \param[in] env execution environment
3048 * \param[in] d LU device of OFD
3052 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3053 struct lu_device *d)
3056 ofd_fini(env, ofd_dev(d));
3061 * Implementation of lu_device_type_operations::ldto_device_free.
3063 * Free OFD device. Dual to ofd_device_alloc().
3065 * \param[in] env execution environment
3066 * \param[in] d LU device of OFD
3070 static struct lu_device *ofd_device_free(const struct lu_env *env,
3071 struct lu_device *d)
3073 struct ofd_device *m = ofd_dev(d);
3075 dt_device_fini(&m->ofd_dt_dev);
3081 * Implementation of lu_device_type_operations::ldto_device_alloc.
3083 * This function allocates the new OFD device. It is called from
3084 * obd_setup() if OBD device had lu_device_type defined.
3086 * \param[in] env execution environment
3087 * \param[in] t lu_device_type of OFD device
3088 * \param[in] cfg configuration log
3090 * \retval pointer to the lu_device of just allocated OFD
3091 * \retval ERR_PTR of return value on error
3093 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3094 struct lu_device_type *t,
3095 struct lustre_cfg *cfg)
3097 struct ofd_device *m;
3098 struct lu_device *l;
3103 return ERR_PTR(-ENOMEM);
3105 l = &m->ofd_dt_dev.dd_lu_dev;
3106 dt_device_init(&m->ofd_dt_dev, t);
3107 rc = ofd_init0(env, m, t, cfg);
3109 ofd_device_free(env, l);
3116 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3117 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3119 static struct lu_device_type_operations ofd_device_type_ops = {
3120 .ldto_init = ofd_type_init,
3121 .ldto_fini = ofd_type_fini,
3123 .ldto_start = ofd_type_start,
3124 .ldto_stop = ofd_type_stop,
3126 .ldto_device_alloc = ofd_device_alloc,
3127 .ldto_device_free = ofd_device_free,
3128 .ldto_device_fini = ofd_device_fini
3131 static struct lu_device_type ofd_device_type = {
3132 .ldt_tags = LU_DEVICE_DT,
3133 .ldt_name = LUSTRE_OST_NAME,
3134 .ldt_ops = &ofd_device_type_ops,
3135 .ldt_ctx_tags = LCT_DT_THREAD
3139 * Initialize OFD module.
3141 * This function is called upon module loading. It registers OFD device type
3142 * and prepares all in-memory structures used by all OFD devices.
3144 * \retval 0 if successful
3145 * \retval negative value on error
3147 static int __init ofd_init(void)
3151 rc = lu_kmem_init(ofd_caches);
3154 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3155 LUSTRE_OST_NAME, &ofd_device_type);
3162 * This function is called upon OFD module unloading.
3163 * It frees all related structures and unregisters OFD device type.
3165 static void __exit ofd_exit(void)
3167 lu_kmem_fini(ofd_caches);
3168 class_unregister_type(LUSTRE_OST_NAME);
3171 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3172 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3173 MODULE_VERSION(LUSTRE_VERSION_STRING);
3174 MODULE_LICENSE("GPL");
3176 module_init(ofd_init);
3177 module_exit(ofd_exit);