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, "writethrough_cache_enable",
331 strncmp(param, "readcache_max_filesize",
333 strncmp(param, "read_cache_enable",
335 strncmp(param, "brw_stats", paramlen) == 0)
344 * Process various configuration parameters.
346 * This function is used by MGS to process specific configurations and
347 * pass them through to the next device in server stack, i.e. the OSD.
349 * \param[in] env execution environment
350 * \param[in] d LU device of OFD
351 * \param[in] cfg parameters to process
353 * \retval 0 if successful
354 * \retval negative value on error
356 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
357 struct lustre_cfg *cfg)
359 struct ofd_device *m = ofd_dev(d);
360 struct dt_device *dt_next = m->ofd_osd;
361 struct lu_device *next = &dt_next->dd_lu_dev;
366 switch (cfg->lcfg_command) {
368 /* For interoperability */
369 struct cfg_interop_param *ptr = NULL;
370 struct lustre_cfg *old_cfg = NULL;
374 param = lustre_cfg_string(cfg, 1);
376 CERROR("param is empty\n");
381 ptr = class_find_old_param(param, ofd_interop_param);
383 if (ptr->new_param == NULL) {
385 CWARN("For interoperability, skip this %s."
386 " It is obsolete.\n", ptr->old_param);
390 CWARN("Found old param %s, changed it to %s.\n",
391 ptr->old_param, ptr->new_param);
394 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
401 if (match_symlink_param(param)) {
402 rc = next->ld_ops->ldo_process_config(env, next, cfg);
406 count = class_modify_config(cfg, PARAM_OST,
407 &d->ld_obd->obd_kset.kobj);
412 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
414 /* we don't understand; pass it on */
415 rc = next->ld_ops->ldo_process_config(env, next, cfg);
418 case LCFG_SPTLRPC_CONF: {
423 /* others are passed further */
424 rc = next->ld_ops->ldo_process_config(env, next, cfg);
431 * Implementation of lu_object_operations::loo_object_init for OFD
433 * Allocate just the next object (OSD) in stack.
435 * \param[in] env execution environment
436 * \param[in] o lu_object of OFD object
437 * \param[in] conf additional configuration parameters, not used here
439 * \retval 0 if successful
440 * \retval negative value on error
442 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
443 const struct lu_object_conf *conf)
445 struct ofd_device *d = ofd_dev(o->lo_dev);
446 struct lu_device *under;
447 struct lu_object *below;
452 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
453 PFID(lu_object_fid(o)));
455 under = &d->ofd_osd->dd_lu_dev;
456 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
458 lu_object_add(o, below);
466 * Implementation of lu_object_operations::loo_object_free.
468 * Finish OFD object lifecycle and free its memory.
470 * \param[in] env execution environment
471 * \param[in] o LU object of OFD object
473 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
475 struct ofd_object *of = ofd_obj(o);
476 struct lu_object_header *h;
481 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
482 PFID(lu_object_fid(o)));
485 lu_object_header_fini(h);
486 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
491 * Implementation of lu_object_operations::loo_object_print.
493 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
494 * LU_OBJECT_DEBUG() for more details about the compound object printing.
496 * \param[in] env execution environment
497 * \param[in] cookie opaque data passed to the printer function
498 * \param[in] p printer function to use
499 * \param[in] o LU object of OFD object
501 * \retval 0 if successful
502 * \retval negative value on error
504 static int ofd_object_print(const struct lu_env *env, void *cookie,
505 lu_printer_t p, const struct lu_object *o)
507 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
510 static struct lu_object_operations ofd_obj_ops = {
511 .loo_object_init = ofd_object_init,
512 .loo_object_free = ofd_object_free,
513 .loo_object_print = ofd_object_print
517 * Implementation of lu_device_operations::lod_object_alloc.
519 * This function allocates OFD part of compound OFD-OSD object and
520 * initializes its header, because OFD is the top device in stack
522 * \param[in] env execution environment
523 * \param[in] hdr object header, NULL for OFD
524 * \param[in] d lu_device
526 * \retval allocated object if successful
527 * \retval NULL value on failed allocation
529 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
530 const struct lu_object_header *hdr,
533 struct ofd_object *of;
537 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
540 struct lu_object_header *h;
542 o = &of->ofo_obj.do_lu;
544 lu_object_header_init(h);
545 lu_object_init(o, h, d);
546 lu_object_add_top(h, o);
547 o->lo_ops = &ofd_obj_ops;
555 * Return the result of LFSCK run to the OFD.
557 * Notify OFD about result of LFSCK run. That may block the new object
558 * creation until problem is fixed by LFSCK.
560 * \param[in] env execution environment
561 * \param[in] data pointer to the OFD device
562 * \param[in] event LFSCK event type
564 * \retval 0 if successful
565 * \retval negative value on unknown event
567 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
568 enum lfsck_events event)
570 struct ofd_device *ofd = data;
571 struct obd_device *obd = ofd_obd(ofd);
574 case LE_LASTID_REBUILDING:
575 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
576 "on the device until the LAST_ID rebuilt successfully.\n",
578 down_write(&ofd->ofd_lastid_rwsem);
579 ofd->ofd_lastid_rebuilding = 1;
580 up_write(&ofd->ofd_lastid_rwsem);
582 case LE_LASTID_REBUILT: {
583 down_write(&ofd->ofd_lastid_rwsem);
584 ofd_seqs_free(env, ofd);
585 ofd->ofd_lastid_rebuilding = 0;
586 ofd->ofd_lastid_gen++;
587 up_write(&ofd->ofd_lastid_rwsem);
588 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
593 CERROR("%s: unknown lfsck event: rc = %d\n",
594 ofd_name(ofd), event);
602 * Implementation of lu_device_operations::ldo_prepare.
604 * This method is called after layer has been initialized and before it starts
605 * serving user requests. In OFD it starts lfsk check routines and initializes
608 * \param[in] env execution environment
609 * \param[in] pdev higher device in stack, NULL for OFD
610 * \param[in] dev lu_device of OFD device
612 * \retval 0 if successful
613 * \retval negative value on error
615 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
616 struct lu_device *dev)
618 struct ofd_thread_info *info;
619 struct ofd_device *ofd = ofd_dev(dev);
620 struct obd_device *obd = ofd_obd(ofd);
621 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
626 info = ofd_info_init(env, NULL);
630 /* initialize lower device */
631 rc = next->ld_ops->ldo_prepare(env, dev, next);
635 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
636 ofd_lfsck_out_notify, ofd, false);
638 CERROR("%s: failed to initialize lfsck: rc = %d\n",
643 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
644 /* The LFSCK instance is registered just now, so it must be there when
645 * register the namespace to such instance. */
646 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
648 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
649 LASSERT(obd->obd_no_conn);
650 spin_lock(&obd->obd_dev_lock);
651 obd->obd_no_conn = 0;
652 spin_unlock(&obd->obd_dev_lock);
654 if (obd->obd_recovering == 0)
655 ofd_postrecov(env, ofd);
661 * Implementation of lu_device_operations::ldo_recovery_complete.
663 * This method notifies all layers about 'recovery complete' event. That means
664 * device is in full state and consistent. An OFD calculates available grant
665 * space upon this event.
667 * \param[in] env execution environment
668 * \param[in] dev lu_device of OFD device
670 * \retval 0 if successful
671 * \retval negative value on error
673 static int ofd_recovery_complete(const struct lu_env *env,
674 struct lu_device *dev)
676 struct ofd_thread_info *oti = ofd_info(env);
677 struct ofd_device *ofd = ofd_dev(dev);
678 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
684 * Grant space for object precreation on the self export.
685 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
686 * is enough to create 10k objects. More space is then acquired for
687 * precreation in tgt_grant_create().
689 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
690 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
691 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
692 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
693 OBD_CONNECT_GRANT_PARAM;
694 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
696 rc = next->ld_ops->ldo_recovery_complete(env, next);
701 * lu_device_operations matrix for OFD device.
703 static struct lu_device_operations ofd_lu_ops = {
704 .ldo_object_alloc = ofd_object_alloc,
705 .ldo_process_config = ofd_process_config,
706 .ldo_recovery_complete = ofd_recovery_complete,
707 .ldo_prepare = ofd_prepare,
710 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
712 * Expose OSD statistics to OFD layer.
714 * The osd interfaces to the backend file system exposes useful data
715 * such as brw_stats and read or write cache states. This same data
716 * needs to be exposed into the obdfilter (ofd) layer to maintain
717 * backwards compatibility. This function creates the symlinks in the
718 * proc layer to enable this.
720 * \param[in] ofd OFD device
722 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
724 struct obd_device *obd = ofd_obd(ofd);
725 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
727 if (obd->obd_proc_entry == NULL)
730 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
731 "../../%s/%s/brw_stats",
732 osd_obd->obd_type->typ_name, obd->obd_name);
734 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
735 "../../%s/%s/read_cache_enable",
736 osd_obd->obd_type->typ_name, obd->obd_name);
738 lprocfs_add_symlink("readcache_max_filesize",
740 "../../%s/%s/readcache_max_filesize",
741 osd_obd->obd_type->typ_name, obd->obd_name);
743 lprocfs_add_symlink("writethrough_cache_enable",
745 "../../%s/%s/writethrough_cache_enable",
746 osd_obd->obd_type->typ_name, obd->obd_name);
751 * Cleanup all procfs entries in OFD.
753 * \param[in] ofd OFD device
755 static void ofd_procfs_fini(struct ofd_device *ofd)
757 struct obd_device *obd = ofd_obd(ofd);
759 lprocfs_free_per_client_stats(obd);
760 lprocfs_obd_cleanup(obd);
761 lprocfs_free_obd_stats(obd);
762 lprocfs_job_stats_fini(obd);
766 * Stop SEQ/FID server on OFD.
768 * \param[in] env execution environment
769 * \param[in] ofd OFD device
771 * \retval 0 if successful
772 * \retval negative value on error
774 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
776 return seq_site_fini(env, &ofd->ofd_seq_site);
780 * Start SEQ/FID server on OFD.
782 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
783 * It also connects to the master server to get own FID sequence (SEQ) range
784 * to this particular OFD. Typically that happens when the OST is first
785 * formatted or in the rare case that it exhausts the local sequence range.
787 * The sequence range is allocated out to the MDTs for OST object allocations,
788 * and not directly to the clients.
790 * \param[in] env execution environment
791 * \param[in] ofd OFD device
793 * \retval 0 if successful
794 * \retval negative value on error
796 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
798 struct seq_server_site *ss = &ofd->ofd_seq_site;
799 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
800 char *obd_name = ofd_name(ofd);
802 int len = strlen(obd_name) + 7;
805 ss = &ofd->ofd_seq_site;
806 lu->ld_site->ld_seq_site = ss;
807 ss->ss_lu = lu->ld_site;
808 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
810 OBD_ALLOC(name, len);
814 OBD_ALLOC_PTR(ss->ss_server_seq);
815 if (ss->ss_server_seq == NULL)
816 GOTO(out_name, rc = -ENOMEM);
818 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
819 LUSTRE_SEQ_SERVER, ss);
821 CERROR("%s: seq server init error: rc = %d\n", obd_name, rc);
822 GOTO(out_server, rc);
824 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
826 OBD_ALLOC_PTR(ss->ss_client_seq);
827 if (ss->ss_client_seq == NULL)
828 GOTO(out_server, rc = -ENOMEM);
830 snprintf(name, len, "%s-super", obd_name);
831 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
834 CERROR("%s: seq client init error: rc = %d\n", obd_name, rc);
835 GOTO(out_client, rc);
838 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
842 seq_client_fini(ss->ss_client_seq);
843 OBD_FREE_PTR(ss->ss_client_seq);
844 ss->ss_client_seq = NULL;
846 seq_server_fini(ss->ss_server_seq, env);
847 OBD_FREE_PTR(ss->ss_server_seq);
848 ss->ss_server_seq = NULL;
857 * OFD request handler for OST_SET_INFO RPC.
859 * This is OFD-specific part of request handling
861 * \param[in] tsi target session environment for this request
863 * \retval 0 if successful
864 * \retval negative value on error
866 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
868 struct ptlrpc_request *req = tgt_ses_req(tsi);
869 struct ost_body *body = NULL, *repbody;
870 void *key, *val = NULL;
871 int keylen, vallen, rc = 0;
872 bool is_grant_shrink;
876 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
878 DEBUG_REQ(D_HA, req, "no set_info key");
879 RETURN(err_serious(-EFAULT));
881 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
884 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
886 DEBUG_REQ(D_HA, req, "no set_info val");
887 RETURN(err_serious(-EFAULT));
889 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
892 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
894 /* In this case the value is actually an RMF_OST_BODY, so we
895 * transmutate the type of this PTLRPC */
896 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
898 rc = req_capsule_server_pack(tsi->tsi_pill);
902 if (is_grant_shrink) {
903 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
905 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
908 /** handle grant shrink, similar to a read request */
909 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
911 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
913 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
916 CERROR("%s: Unsupported key %s\n",
917 tgt_name(tsi->tsi_tgt), (char *)key);
920 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
927 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
929 * This function returns a list of extents which describes how a file's
930 * blocks are laid out on the disk.
932 * \param[in] env execution environment
933 * \param[in] ofd OFD device
934 * \param[in] fid FID of object
935 * \param[in] fiemap fiemap structure to fill with data
937 * \retval 0 if \a fiemap is filled with data successfully
938 * \retval negative value on error
940 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
941 struct lu_fid *fid, struct fiemap *fiemap)
943 struct ofd_object *fo;
946 fo = ofd_object_find(env, ofd, fid);
948 CERROR("%s: error finding object "DFID"\n",
949 ofd_name(ofd), PFID(fid));
953 ofd_read_lock(env, fo);
954 if (ofd_object_exists(fo))
955 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
958 ofd_read_unlock(env, fo);
959 ofd_object_put(env, fo);
964 static int ofd_lock_unlock_region(const struct lu_env *env,
965 struct ldlm_namespace *ns,
966 struct ldlm_res_id *res_id,
967 unsigned long long begin,
968 unsigned long long end)
972 struct lustre_handle lh = { 0 };
974 LASSERT(begin <= end);
976 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
980 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
981 tgt_extent_unlock(&lh, LCK_PR);
987 * Lock the sparse areas of given resource.
989 * The locking of sparse areas will cause dirty data to be flushed back from
990 * clients. This is used when getting the FIEMAP of an object to make sure
991 * there is no unaccounted cached data on clients.
993 * This function goes through \a fiemap list of extents and locks only sparse
994 * areas between extents.
996 * \param[in] ns LDLM namespace
997 * \param[in] res_id resource ID
998 * \param[in] fiemap file extents mapping on disk
999 * \param[in] locked list head of regions list
1001 * \retval 0 if successful
1002 * \retval negative value on error
1004 static int lock_zero_regions(const struct lu_env *env,
1005 struct ldlm_namespace *ns,
1006 struct ldlm_res_id *res_id,
1007 struct fiemap *fiemap)
1009 __u64 begin = fiemap->fm_start;
1012 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1016 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1017 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1018 if (fiemap_start[i].fe_logical > begin) {
1019 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1020 begin, fiemap_start[i].fe_logical);
1021 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1022 fiemap_start[i].fe_logical);
1027 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1030 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1031 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1032 begin, fiemap->fm_start + fiemap->fm_length);
1033 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1034 fiemap->fm_start + fiemap->fm_length);
1042 * OFD request handler for OST_GET_INFO RPC.
1044 * This is OFD-specific part of request handling. The OFD-specific keys are:
1045 * - KEY_LAST_ID (obsolete)
1049 * This function reads needed data from storage and fills reply with it.
1051 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1052 * and is kept for compatibility.
1054 * \param[in] tsi target session environment for this request
1056 * \retval 0 if successful
1057 * \retval negative value on error
1059 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1061 struct obd_export *exp = tsi->tsi_exp;
1062 struct ofd_device *ofd = ofd_exp(exp);
1063 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1066 int replylen, rc = 0;
1070 /* this common part for get_info rpc */
1071 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1073 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1074 RETURN(err_serious(-EPROTO));
1076 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1079 if (KEY_IS(KEY_LAST_ID)) {
1081 struct ofd_seq *oseq;
1083 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1084 rc = req_capsule_server_pack(tsi->tsi_pill);
1086 RETURN(err_serious(rc));
1088 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1090 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1091 (u64)exp->exp_filter_data.fed_group);
1095 *last_id = ofd_seq_last_oid(oseq);
1096 ofd_seq_put(tsi->tsi_env, oseq);
1097 } else if (KEY_IS(KEY_FIEMAP)) {
1098 struct ll_fiemap_info_key *fm_key;
1099 struct fiemap *fiemap;
1102 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1104 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1105 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1107 RETURN(err_serious(rc));
1109 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1111 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1113 replylen = fiemap_count_to_size(
1114 fm_key->lfik_fiemap.fm_extent_count);
1115 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1116 RCL_SERVER, replylen);
1118 rc = req_capsule_server_pack(tsi->tsi_pill);
1120 RETURN(err_serious(rc));
1122 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1126 *fiemap = fm_key->lfik_fiemap;
1127 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1129 /* LU-3219: Lock the sparse areas to make sure dirty
1130 * flushed back from client, then call fiemap again. */
1131 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1132 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1133 ost_fid_build_resid(fid, &fti->fti_resid);
1134 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1135 &fti->fti_resid, fiemap);
1137 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1140 } else if (KEY_IS(KEY_LAST_FID)) {
1141 struct ofd_device *ofd = ofd_exp(exp);
1142 struct ofd_seq *oseq;
1146 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1147 rc = req_capsule_server_pack(tsi->tsi_pill);
1149 RETURN(err_serious(rc));
1151 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1153 RETURN(err_serious(-EPROTO));
1155 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1157 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1161 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1162 ostid_seq(&fti->fti_ostid));
1164 RETURN(PTR_ERR(oseq));
1166 rc = ostid_to_fid(fid, &oseq->os_oi,
1167 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1171 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1174 ofd_seq_put(tsi->tsi_env, oseq);
1176 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1180 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1187 * OFD request handler for OST_GETATTR RPC.
1189 * This is OFD-specific part of request handling. It finds the OFD object
1190 * by its FID, gets attributes from storage and packs result to the reply.
1192 * \param[in] tsi target session environment for this request
1194 * \retval 0 if successful
1195 * \retval negative value on error
1197 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1199 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1200 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1201 struct ost_body *repbody;
1202 struct lustre_handle lh = { 0 };
1203 struct ofd_object *fo;
1205 enum ldlm_mode lock_mode = LCK_PR;
1210 LASSERT(tsi->tsi_ost_body != NULL);
1212 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1213 if (repbody == NULL)
1216 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1217 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1219 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1220 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1223 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1226 rc = tgt_extent_lock(tsi->tsi_env,
1227 tsi->tsi_tgt->lut_obd->obd_namespace,
1228 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1234 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1236 GOTO(out, rc = PTR_ERR(fo));
1238 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1242 obdo_from_la(&repbody->oa, &fti->fti_attr,
1243 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1245 /* Store object version in reply */
1246 curr_version = dt_version_get(tsi->tsi_env,
1247 ofd_object_child(fo));
1248 if ((__s64)curr_version != -EOPNOTSUPP) {
1249 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1250 repbody->oa.o_data_version = curr_version;
1253 if (fo->ofo_ff.ff_layout_version > 0) {
1254 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1255 repbody->oa.o_layout_version =
1256 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1258 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1259 PFID(&tsi->tsi_fid),
1260 repbody->oa.o_layout_version);
1264 ofd_object_put(tsi->tsi_env, fo);
1267 tgt_extent_unlock(&lh, lock_mode);
1269 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1272 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1273 repbody->oa.o_flags = OBD_FL_FLUSH;
1279 * OFD request handler for OST_SETATTR RPC.
1281 * This is OFD-specific part of request handling. It finds the OFD object
1282 * by its FID, sets attributes from request and packs result to the reply.
1284 * \param[in] tsi target session environment for this request
1286 * \retval 0 if successful
1287 * \retval negative value on error
1289 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1291 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1292 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1293 struct ost_body *body = tsi->tsi_ost_body;
1294 struct ost_body *repbody;
1295 struct ldlm_resource *res;
1296 struct ofd_object *fo;
1301 LASSERT(body != NULL);
1303 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1304 if (repbody == NULL)
1307 repbody->oa.o_oi = body->oa.o_oi;
1308 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1310 /* This would be very bad - accidentally truncating a file when
1311 * changing the time or similar - bug 12203. */
1312 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1313 body->oa.o_size != OBD_OBJECT_EOF) {
1314 static char mdsinum[48];
1316 if (body->oa.o_valid & OBD_MD_FLFID)
1317 snprintf(mdsinum, sizeof(mdsinum) - 1,
1318 "of parent "DFID, body->oa.o_parent_seq,
1319 body->oa.o_parent_oid, 0);
1323 CERROR("%s: setattr from %s is trying to truncate object "DFID
1324 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1325 PFID(&tsi->tsi_fid), mdsinum);
1329 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1331 GOTO(out, rc = PTR_ERR(fo));
1333 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1334 fti->fti_attr.la_valid &= ~LA_TYPE;
1336 /* setting objects attributes (including owner/group) */
1337 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1341 obdo_from_la(&repbody->oa, &fti->fti_attr,
1342 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1344 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1348 ofd_object_put(tsi->tsi_env, fo);
1351 /* we do not call this before to avoid lu_object_find() in
1352 * ->lvbo_update() holding another reference on the object.
1353 * otherwise concurrent destroy can make the object unavailable
1354 * for 2nd lu_object_find() waiting for the first reference
1355 * to go... deadlock! */
1356 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1357 &tsi->tsi_resid, LDLM_EXTENT, 0);
1359 ldlm_res_lvbo_update(tsi->tsi_env, res, NULL, 0);
1360 ldlm_resource_putref(res);
1367 * Destroy OST orphans.
1369 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1370 * set then we must destroy possible orphaned objects.
1372 * \param[in] env execution environment
1373 * \param[in] exp OBD export
1374 * \param[in] ofd OFD device
1375 * \param[in] oa obdo structure for reply
1377 * \retval 0 if successful
1378 * \retval negative value on error
1380 static int ofd_orphans_destroy(const struct lu_env *env,
1381 struct obd_export *exp,
1382 struct ofd_device *ofd, struct obdo *oa)
1384 struct ofd_thread_info *info = ofd_info(env);
1385 struct lu_fid *fid = &info->fti_fid;
1386 struct ost_id *oi = &oa->o_oi;
1387 struct ofd_seq *oseq;
1388 u64 seq = ostid_seq(oi);
1389 u64 end_id = ostid_id(oi);
1397 oseq = ofd_seq_get(ofd, seq);
1399 CERROR("%s: Can not find seq for "DOSTID"\n",
1400 ofd_name(ofd), POSTID(oi));
1405 last = ofd_seq_last_oid(oseq);
1408 LASSERT(exp != NULL);
1409 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1411 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1414 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1415 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1417 while (oid > end_id) {
1418 rc = fid_set_id(fid, oid);
1419 if (unlikely(rc != 0))
1422 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1423 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1424 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1425 /* this is pretty fatal... */
1426 CEMERG("%s: error destroying precreated id "
1428 ofd_name(ofd), PFID(fid), rc);
1432 ofd_seq_last_oid_set(oseq, oid);
1433 /* update last_id on disk periodically so that if we
1434 * restart * we don't need to re-scan all of the just
1435 * deleted objects. */
1436 if ((oid & 511) == 0)
1437 ofd_seq_last_oid_write(env, ofd, oseq);
1441 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1442 ofd_name(ofd), seq, oid);
1446 ofd_seq_last_oid_set(oseq, oid);
1447 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1449 /* don't reuse orphan object, return last used objid */
1450 rc = ostid_set_id(oi, last);
1456 ofd_seq_put(env, oseq);
1461 * OFD request handler for OST_CREATE RPC.
1463 * This is OFD-specific part of request handling. Its main purpose is to
1464 * create new data objects on OST, but it also used to destroy orphans.
1466 * \param[in] tsi target session environment for this request
1468 * \retval 0 if successful
1469 * \retval negative value on error
1471 static int ofd_create_hdl(struct tgt_session_info *tsi)
1473 struct ptlrpc_request *req = tgt_ses_req(tsi);
1474 struct ost_body *repbody;
1475 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1476 struct obdo *rep_oa;
1477 struct obd_export *exp = tsi->tsi_exp;
1478 struct ofd_device *ofd = ofd_exp(exp);
1479 u64 seq = ostid_seq(&oa->o_oi);
1480 u64 oid = ostid_id(&oa->o_oi);
1481 struct ofd_seq *oseq;
1489 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1492 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1493 if (repbody == NULL)
1496 down_read(&ofd->ofd_lastid_rwsem);
1497 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1498 * we may do that in the future.
1499 * Return -ENOSPC until the LAST_ID rebuilt. */
1500 if (unlikely(ofd->ofd_lastid_rebuilding))
1501 GOTO(out_sem, rc = -ENOSPC);
1503 rep_oa = &repbody->oa;
1504 rep_oa->o_oi = oa->o_oi;
1506 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1508 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1510 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1512 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1513 ofd_name(ofd), seq, PTR_ERR(oseq));
1514 GOTO(out_sem, rc = -EINVAL);
1517 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1518 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1519 if (!ofd_obd(ofd)->obd_recovering ||
1520 oid > ofd_seq_last_oid(oseq)) {
1521 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1522 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1523 ofd_seq_last_oid(oseq));
1524 GOTO(out_nolock, rc = -EINVAL);
1526 /* Do nothing here, we re-create objects during recovery
1527 * upon write replay, see ofd_preprw_write() */
1528 GOTO(out_nolock, rc = 0);
1530 /* former ofd_handle_precreate */
1531 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1532 (oa->o_flags & OBD_FL_DELORPHAN)) {
1533 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1535 /* destroy orphans */
1536 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1537 exp->exp_conn_cnt) {
1538 CERROR("%s: dropping old orphan cleanup request\n",
1540 GOTO(out_nolock, rc = 0);
1542 /* This causes inflight precreates to abort and drop lock */
1543 oseq->os_destroys_in_progress = 1;
1544 mutex_lock(&oseq->os_create_lock);
1545 if (!oseq->os_destroys_in_progress) {
1547 "%s:[%llu] destroys_in_progress already cleared\n",
1548 ofd_name(ofd), seq);
1549 rc = ostid_set_id(&rep_oa->o_oi,
1550 ofd_seq_last_oid(oseq));
1553 diff = oid - ofd_seq_last_oid(oseq);
1554 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1555 ofd_seq_last_oid(oseq), diff);
1556 if (-diff > OST_MAX_PRECREATE) {
1557 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1558 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1559 "(%llu), trust the OST\n",
1560 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1561 PFID(&oseq->os_oi.oi_fid),
1562 ofd_seq_last_oid(oseq));
1564 /* Let MDS know that we are so far ahead. */
1565 rc = ostid_set_id(&rep_oa->o_oi,
1566 ofd_seq_last_oid(oseq) + 1);
1567 } else if (diff < 0) {
1568 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1570 oseq->os_destroys_in_progress = 0;
1572 /* XXX: Used by MDS for the first time! */
1573 oseq->os_destroys_in_progress = 0;
1576 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1577 ofd->ofd_lastid_gen)) {
1578 /* Keep the export ref so we can send the reply. */
1579 ofd_obd_disconnect(class_export_get(exp));
1580 GOTO(out_nolock, rc = -ENOTCONN);
1583 mutex_lock(&oseq->os_create_lock);
1584 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1585 exp->exp_conn_cnt) {
1586 CERROR("%s: dropping old precreate request\n",
1590 /* only precreate if seq is 0, IDIF or normal and also o_id
1591 * must be specfied */
1592 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1593 !fid_seq_is_idif(seq)) || oid == 0) {
1594 diff = 1; /* shouldn't we create this right now? */
1596 diff = oid - ofd_seq_last_oid(oseq);
1597 /* Do sync create if the seq is about to used up */
1598 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1599 if (unlikely(oid >= IDIF_MAX_OID - 1))
1601 } else if (fid_seq_is_norm(seq)) {
1603 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1606 CERROR("%s : invalid o_seq "DOSTID"\n",
1607 ofd_name(ofd), POSTID(&oa->o_oi));
1608 GOTO(out, rc = -EINVAL);
1613 CERROR("%s: invalid precreate request for "
1614 DOSTID", last_id %llu. "
1615 "Likely MDS last_id corruption\n",
1616 ofd_name(ofd), POSTID(&oa->o_oi),
1617 ofd_seq_last_oid(oseq));
1618 GOTO(out, rc = -EINVAL);
1623 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1629 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1630 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1631 /* don't enforce grant during orphan recovery */
1632 granted = tgt_grant_create(tsi->tsi_env,
1633 ofd_obd(ofd)->obd_self_export,
1638 CDEBUG(D_HA, "%s: failed to acquire grant "
1639 "space for precreate (%lld): rc = %d\n",
1640 ofd_name(ofd), diff, rc);
1645 /* This can happen if a new OST is formatted and installed
1646 * in place of an old one at the same index. Instead of
1647 * precreating potentially millions of deleted old objects
1648 * (possibly filling the OST), only precreate the last batch.
1649 * LFSCK will eventually clean up any orphans. LU-14 */
1650 if (diff > 5 * OST_MAX_PRECREATE) {
1651 diff = OST_MAX_PRECREATE / 2;
1652 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1653 "OST replaced or reformatted: "
1654 "LFSCK will clean up",
1657 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1658 "%u larger than the LAST_ID "DOSTID", only "
1659 "precreating the last %lld objects.\n",
1660 ofd_name(ofd), POSTID(&oa->o_oi),
1661 5 * OST_MAX_PRECREATE,
1662 POSTID(&oseq->os_oi), diff);
1663 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1667 next_id = ofd_seq_last_oid(oseq) + 1;
1668 count = ofd_precreate_batch(ofd, (int)diff);
1670 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1671 " at %llu\n", ofd_name(ofd),
1672 count, seq, next_id);
1674 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1675 && ktime_get_seconds() > enough_time) {
1676 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1677 " created at a rate of %d/s\n",
1678 ofd_name(ofd), created, diff + created,
1679 created / DISK_TIMEOUT);
1683 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1684 oseq, count, sync_trans);
1688 } else if (rc < 0) {
1694 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1695 LCONSOLE_WARN("%s: can't create the same count of"
1696 " objects when replaying the request"
1697 " (diff is %lld). see LU-4621\n",
1698 ofd_name(ofd), diff);
1701 /* some objects got created, we can return
1702 * them, even if last creation failed */
1705 CERROR("%s: unable to precreate: rc = %d\n",
1708 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1709 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1710 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1715 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1719 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1722 mutex_unlock(&oseq->os_create_lock);
1725 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1726 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1727 struct lu_fid *fid = &info->fti_fid;
1729 /* For compatible purpose, it needs to convert back to
1730 * OST ID before put it on wire. */
1731 *fid = rep_oa->o_oi.oi_fid;
1732 fid_to_ostid(fid, &rep_oa->o_oi);
1734 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1736 ofd_seq_put(tsi->tsi_env, oseq);
1739 up_read(&ofd->ofd_lastid_rwsem);
1744 * OFD request handler for OST_DESTROY RPC.
1746 * This is OFD-specific part of request handling. It destroys data objects
1747 * related to destroyed object on MDT.
1749 * \param[in] tsi target session environment for this request
1751 * \retval 0 if successful
1752 * \retval negative value on error
1754 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1756 const struct ost_body *body = tsi->tsi_ost_body;
1757 struct ost_body *repbody;
1758 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1759 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1760 struct lu_fid *fid = &fti->fti_fid;
1767 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1770 /* This is old case for clients before Lustre 2.4 */
1771 /* If there's a DLM request, cancel the locks mentioned in it */
1772 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1774 struct ldlm_request *dlm;
1776 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1779 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1782 *fid = body->oa.o_oi.oi_fid;
1783 oid = ostid_id(&body->oa.o_oi);
1786 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1788 /* check that o_misc makes sense */
1789 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1790 count = body->oa.o_misc;
1792 count = 1; /* default case - single destroy */
1794 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1795 POSTID(&body->oa.o_oi), count);
1800 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1801 if (lrc == -ENOENT) {
1803 "%s: destroying non-existent object "DFID"\n",
1804 ofd_name(ofd), PFID(fid));
1805 /* rewrite rc with -ENOENT only if it is 0 */
1808 } else if (lrc != 0) {
1809 CERROR("%s: error destroying object "DFID": %d\n",
1810 ofd_name(ofd), PFID(fid), lrc);
1816 lrc = fid_set_id(fid, oid);
1817 if (unlikely(lrc != 0 && count > 0))
1818 GOTO(out, rc = lrc);
1821 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1827 fid_to_ostid(fid, &repbody->oa.o_oi);
1832 * OFD request handler for OST_STATFS RPC.
1834 * This function gets statfs data from storage as part of request
1837 * \param[in] tsi target session environment for this request
1839 * \retval 0 if successful
1840 * \retval negative value on error
1842 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1844 struct obd_statfs *osfs;
1849 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1851 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1853 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1854 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1856 CERROR("%s: statfs failed: rc = %d\n",
1857 tgt_name(tsi->tsi_tgt), rc);
1859 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1862 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1869 * OFD request handler for OST_SYNC RPC.
1871 * Sync object data or all filesystem data to the disk and pack the
1874 * \param[in] tsi target session environment for this request
1876 * \retval 0 if successful
1877 * \retval negative value on error
1879 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1881 struct ost_body *body = tsi->tsi_ost_body;
1882 struct ost_body *repbody;
1883 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1884 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1885 struct ofd_object *fo = NULL;
1890 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1892 /* if no objid is specified, it means "sync whole filesystem" */
1893 if (!fid_is_zero(&tsi->tsi_fid)) {
1894 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1896 RETURN(PTR_ERR(fo));
1899 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1900 fo != NULL ? ofd_object_child(fo) : NULL,
1901 repbody->oa.o_size, repbody->oa.o_blocks);
1905 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1910 repbody->oa.o_oi = body->oa.o_oi;
1911 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1913 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1915 obdo_from_la(&repbody->oa, &fti->fti_attr,
1918 /* don't return rc from getattr */
1923 ofd_object_put(tsi->tsi_env, fo);
1928 * OFD request handler for OST_PUNCH RPC.
1930 * This is part of request processing. Validate request fields,
1931 * punch (truncate) the given OFD object and pack reply.
1933 * \param[in] tsi target session environment for this request
1935 * \retval 0 if successful
1936 * \retval negative value on error
1938 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1940 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1941 struct ost_body *repbody;
1942 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1943 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1944 struct ldlm_resource *res;
1945 struct ofd_object *fo;
1947 struct lustre_handle lh = { 0, };
1954 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
1956 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1957 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
1959 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1960 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1961 RETURN(err_serious(-EPROTO));
1963 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1964 if (repbody == NULL)
1965 RETURN(err_serious(-ENOMEM));
1967 /* punch start,end are passed in o_size,o_blocks throught wire */
1971 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1974 /* standard truncate optimization: if file body is completely
1975 * destroyed, don't send data back to the server. */
1977 flags |= LDLM_FL_AST_DISCARD_DATA;
1979 repbody->oa.o_oi = oa->o_oi;
1980 repbody->oa.o_valid = OBD_MD_FLID;
1982 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1983 oa->o_flags & OBD_FL_SRVLOCK;
1986 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
1987 end, &lh, LCK_PW, &flags);
1992 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
1993 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
1994 oa->o_valid, start, end);
1996 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
1999 GOTO(out, rc = PTR_ERR(fo));
2001 la_from_obdo(&info->fti_attr, oa,
2002 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2003 info->fti_attr.la_size = start;
2004 info->fti_attr.la_valid |= LA_SIZE;
2006 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2011 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2015 ofd_object_put(tsi->tsi_env, fo);
2018 tgt_extent_unlock(&lh, LCK_PW);
2020 /* we do not call this before to avoid lu_object_find() in
2021 * ->lvbo_update() holding another reference on the object.
2022 * otherwise concurrent destroy can make the object unavailable
2023 * for 2nd lu_object_find() waiting for the first reference
2024 * to go... deadlock! */
2025 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2028 struct ost_lvb *res_lvb;
2030 ldlm_res_lvbo_update(tsi->tsi_env, res, NULL, 0);
2031 res_lvb = res->lr_lvb_data;
2032 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2033 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2034 ldlm_resource_putref(res);
2040 static int ofd_ladvise_prefetch(const struct lu_env *env,
2041 struct ofd_object *fo,
2042 struct niobuf_local *lnb,
2043 __u64 start, __u64 end, enum dt_bufs_type dbt)
2045 struct ofd_thread_info *info = ofd_info(env);
2046 pgoff_t start_index, end_index, pages;
2047 struct niobuf_remote rnb;
2048 unsigned long nr_local;
2054 ofd_read_lock(env, fo);
2055 if (!ofd_object_exists(fo))
2056 GOTO(out_unlock, rc = -ENOENT);
2058 rc = ofd_attr_get(env, fo, &info->fti_attr);
2060 GOTO(out_unlock, rc);
2062 if (end > info->fti_attr.la_size)
2063 end = info->fti_attr.la_size;
2066 GOTO(out_unlock, rc);
2068 /* We need page aligned offset and length */
2069 start_index = start >> PAGE_SHIFT;
2070 end_index = (end - 1) >> PAGE_SHIFT;
2071 pages = end_index - start_index + 1;
2073 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2074 PTLRPC_MAX_BRW_PAGES;
2075 rnb.rnb_offset = start_index << PAGE_SHIFT;
2076 rnb.rnb_len = nr_local << PAGE_SHIFT;
2077 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, dbt);
2078 if (unlikely(rc < 0))
2081 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2082 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2085 start_index += nr_local;
2090 ofd_read_unlock(env, fo);
2095 * OFD request handler for OST_LADVISE RPC.
2097 * Tune cache or perfetch policies according to advices.
2099 * \param[in] tsi target session environment for this request
2101 * \retval 0 if successful
2102 * \retval negative errno on error
2104 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2106 struct ptlrpc_request *req = tgt_ses_req(tsi);
2107 struct obd_export *exp = tsi->tsi_exp;
2108 struct ofd_device *ofd = ofd_exp(exp);
2109 struct ost_body *body, *repbody;
2110 struct ofd_thread_info *info;
2111 struct ofd_object *fo;
2112 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2113 const struct lu_env *env = svc_thread->t_env;
2114 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2115 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2116 struct lu_ladvise *ladvise;
2118 struct ladvise_hdr *ladvise_hdr;
2119 struct obd_ioobj ioo;
2120 struct lustre_handle lockh = { 0 };
2123 struct dt_object *dob;
2129 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2130 body = tsi->tsi_ost_body;
2132 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2133 RETURN(err_serious(-EPROTO));
2135 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2136 &RMF_OST_LADVISE_HDR);
2137 if (ladvise_hdr == NULL)
2138 RETURN(err_serious(-EPROTO));
2140 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2141 ladvise_hdr->lah_count < 1)
2142 RETURN(err_serious(-EPROTO));
2144 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2145 RETURN(err_serious(-EPROTO));
2147 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2148 if (ladvise == NULL)
2149 RETURN(err_serious(-EPROTO));
2151 num_advise = req_capsule_get_size(&req->rq_pill,
2152 &RMF_OST_LADVISE, RCL_CLIENT) /
2154 if (num_advise < ladvise_hdr->lah_count)
2155 RETURN(err_serious(-EPROTO));
2157 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2158 repbody->oa = body->oa;
2160 info = ofd_info_init(env, exp);
2162 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2163 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2167 fo = ofd_object_find(env, ofd, &info->fti_fid);
2172 LASSERT(fo != NULL);
2173 dob = ofd_object_child(fo);
2175 if (ptlrpc_connection_is_local(exp->exp_connection))
2176 dbt |= DT_BUFS_TYPE_LOCAL;
2178 for (i = 0; i < num_advise; i++, ladvise++) {
2179 start = ladvise->lla_start;
2180 end = ladvise->lla_end;
2182 rc = err_serious(-EPROTO);
2186 /* Handle different advice types */
2187 switch (ladvise->lla_advice) {
2191 case LU_LADVISE_WILLREAD:
2195 ioo.ioo_oid = body->oa.o_oi;
2197 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2198 &tsi->tsi_resid, start, end - 1,
2199 &lockh, LCK_PR, &flags);
2203 req->rq_status = ofd_ladvise_prefetch(env, fo,
2206 tgt_extent_unlock(&lockh, LCK_PR);
2208 case LU_LADVISE_DONTNEED:
2209 rc = dt_ladvise(env, dob, ladvise->lla_start,
2210 ladvise->lla_end, LU_LADVISE_DONTNEED);
2217 ofd_object_put(env, fo);
2218 req->rq_status = rc;
2223 * OFD request handler for OST_QUOTACTL RPC.
2225 * This is part of request processing to validate incoming request fields,
2226 * get the requested data from OSD and pack reply.
2228 * \param[in] tsi target session environment for this request
2230 * \retval 0 if successful
2231 * \retval negative value on error
2233 static int ofd_quotactl(struct tgt_session_info *tsi)
2235 struct obd_quotactl *oqctl, *repoqc;
2236 struct lu_nodemap *nodemap;
2242 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2244 RETURN(err_serious(-EPROTO));
2246 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2248 RETURN(err_serious(-ENOMEM));
2252 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2253 if (IS_ERR(nodemap))
2254 RETURN(PTR_ERR(nodemap));
2257 if (oqctl->qc_type == USRQUOTA)
2258 id = nodemap_map_id(nodemap, NODEMAP_UID,
2259 NODEMAP_CLIENT_TO_FS,
2261 else if (oqctl->qc_type == GRPQUOTA)
2262 id = nodemap_map_id(nodemap, NODEMAP_GID,
2263 NODEMAP_CLIENT_TO_FS,
2266 nodemap_putref(nodemap);
2268 if (repoqc->qc_id != id)
2269 swap(repoqc->qc_id, id);
2271 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2273 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2276 if (repoqc->qc_id != id)
2277 swap(repoqc->qc_id, id);
2283 * Calculate the amount of time for lock prolongation.
2285 * This is helper for ofd_prolong_extent_locks() function to get
2286 * the timeout extra time.
2288 * \param[in] req current request
2290 * \retval amount of time to extend the timeout with
2292 static inline time64_t prolong_timeout(struct ptlrpc_request *req)
2294 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2295 time64_t req_timeout;
2298 return obd_timeout / 2;
2300 req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec;
2301 return max_t(time64_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
2306 * Prolong lock timeout for the given extent.
2308 * This function finds all locks related with incoming request and
2309 * prolongs their timeout.
2311 * If a client is holding a lock for a long time while it sends
2312 * read or write RPCs to the OST for the object under this lock,
2313 * then we don't want the OST to evict the client. Otherwise,
2314 * if the network or disk is very busy then the client may not
2315 * be able to make any progress to clear out dirty pages under
2316 * the lock and the application will fail.
2318 * Every time a Bulk Read/Write (BRW) request arrives for the object
2319 * covered by the lock, extend the timeout on that lock. The RPC should
2320 * contain a lock handle for the lock it is using, but this
2321 * isn't handled correctly by all client versions, and the
2322 * request may cover multiple locks.
2324 * \param[in] tsi target session environment for this request
2325 * \param[in] data struct of data to prolong locks
2328 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2329 struct ldlm_prolong_args *data)
2331 struct obdo *oa = &tsi->tsi_ost_body->oa;
2332 struct ldlm_lock *lock;
2336 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2337 data->lpa_export = tsi->tsi_exp;
2338 data->lpa_resid = tsi->tsi_resid;
2340 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2341 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2342 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2343 data->lpa_extent.end);
2345 if (oa->o_valid & OBD_MD_FLHANDLE) {
2346 /* mostly a request should be covered by only one lock, try
2348 lock = ldlm_handle2lock(&oa->o_handle);
2350 /* Fast path to check if the lock covers the whole IO
2351 * region exclusively. */
2352 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2353 &data->lpa_extent)) {
2355 LASSERT(lock->l_export == data->lpa_export);
2356 ldlm_lock_prolong_one(lock, data);
2357 LDLM_LOCK_PUT(lock);
2358 if (data->lpa_locks_cnt > 0)
2360 /* The lock was destroyed probably lets try
2363 lock->l_last_used = ktime_get();
2364 LDLM_LOCK_PUT(lock);
2369 ldlm_resource_prolong(data);
2374 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2376 * Determine if \a lock and the lock from request \a req are equivalent
2377 * by comparing their resource names, modes, and extents.
2379 * It is used to give priority to read and write RPCs being done
2380 * under this lock so that the client can drop the contended
2381 * lock more quickly and let other clients use it. This improves
2382 * overall performance in the case where the first client gets a
2383 * very large lock extent that prevents other clients from
2384 * submitting their writes.
2386 * \param[in] req ptlrpc_request being processed
2387 * \param[in] lock contended lock to match
2389 * \retval 1 if lock is matched
2390 * \retval 0 otherwise
2392 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2393 struct ldlm_lock *lock)
2395 struct niobuf_remote *rnb;
2396 struct obd_ioobj *ioo;
2397 enum ldlm_mode mode;
2398 struct ldlm_extent ext;
2399 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2403 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2404 LASSERT(ioo != NULL);
2406 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2407 LASSERT(rnb != NULL);
2409 ext.start = rnb->rnb_offset;
2410 rnb += ioo->ioo_bufcnt - 1;
2411 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2413 LASSERT(lock->l_resource != NULL);
2414 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2417 /* a bulk write can only hold a reference on a PW extent lock
2420 mode = LCK_PW | LCK_GROUP;
2421 if (opc == OST_READ)
2422 /* whereas a bulk read can be protected by either a PR or PW
2426 if (!(lock->l_granted_mode & mode))
2429 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2433 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2435 * Check for whether the given PTLRPC request (\a req) is blocking
2436 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2439 * \param[in] req the incoming request
2441 * \retval 1 if \a req is blocking an LDLM lock cancel
2442 * \retval 0 if it is not
2443 * \retval -ESTALE if lock is not found
2445 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2447 struct tgt_session_info *tsi;
2448 struct obd_ioobj *ioo;
2449 struct niobuf_remote *rnb;
2451 struct ldlm_prolong_args pa = { 0 };
2455 /* Don't use tgt_ses_info() to get session info, because lock_match()
2456 * can be called while request has no processing thread yet. */
2457 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2460 * Use LASSERT below because malformed RPCs should have
2461 * been filtered out in tgt_hpreq_handler().
2463 opc = lustre_msg_get_opc(req->rq_reqmsg);
2464 LASSERT(opc == OST_READ || opc == OST_WRITE);
2466 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2467 LASSERT(ioo != NULL);
2469 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2470 LASSERT(rnb != NULL);
2471 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2473 pa.lpa_mode = LCK_PW | LCK_GROUP;
2474 if (opc == OST_READ)
2475 pa.lpa_mode |= LCK_PR;
2477 pa.lpa_extent.start = rnb->rnb_offset;
2478 rnb += ioo->ioo_bufcnt - 1;
2479 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2481 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
2482 " (%llu->%llu)\n", tgt_name(tsi->tsi_tgt),
2483 current->comm, PFID(&tsi->tsi_fid), pa.lpa_extent.start,
2486 ofd_prolong_extent_locks(tsi, &pa);
2488 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2489 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2491 if (pa.lpa_blocks_cnt > 0)
2494 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2498 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2500 * Called after the request has been handled. It refreshes lock timeout again
2501 * so that client has more time to send lock cancel RPC.
2503 * \param[in] req request which is being processed.
2505 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2507 ofd_rw_hpreq_check(req);
2511 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2513 * This function checks if the given lock is the same by its resname, mode
2514 * and extent as one taken from the request.
2515 * It is used to give priority to punch/truncate RPCs that might lead to
2516 * the fastest release of that lock when a lock is contended.
2518 * \param[in] req ptlrpc_request being processed
2519 * \param[in] lock contended lock to match
2521 * \retval 1 if lock is matched
2522 * \retval 0 otherwise
2524 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2525 struct ldlm_lock *lock)
2527 struct tgt_session_info *tsi;
2529 struct ldlm_extent ext;
2533 /* Don't use tgt_ses_info() to get session info, because lock_match()
2534 * can be called while request has no processing thread yet. */
2535 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2538 * Use LASSERT below because malformed RPCs should have
2539 * been filtered out in tgt_hpreq_handler().
2541 LASSERT(tsi->tsi_ost_body != NULL);
2542 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2543 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2546 oa = &tsi->tsi_ost_body->oa;
2547 ext.start = oa->o_size;
2548 ext.end = oa->o_blocks;
2550 LASSERT(lock->l_resource != NULL);
2551 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2554 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2557 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2561 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2563 * High-priority queue request check for whether the given punch request
2564 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2565 * covered by an LDLM lock.
2569 * \param[in] req the incoming request
2571 * \retval 1 if \a req is blocking an LDLM lock cancel
2572 * \retval 0 if it is not
2573 * \retval -ESTALE if lock is not found
2575 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2577 struct tgt_session_info *tsi;
2579 struct ldlm_prolong_args pa = { 0 };
2583 /* Don't use tgt_ses_info() to get session info, because lock_match()
2584 * can be called while request has no processing thread yet. */
2585 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2586 LASSERT(tsi != NULL);
2587 oa = &tsi->tsi_ost_body->oa;
2589 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2590 oa->o_flags & OBD_FL_SRVLOCK));
2592 pa.lpa_mode = LCK_PW | LCK_GROUP;
2593 pa.lpa_extent.start = oa->o_size;
2594 pa.lpa_extent.end = oa->o_blocks;
2597 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2598 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2599 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2601 ofd_prolong_extent_locks(tsi, &pa);
2603 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2604 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2606 if (pa.lpa_blocks_cnt > 0)
2609 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2613 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2615 * Called after the request has been handled. It refreshes lock timeout again
2616 * so that client has more time to send lock cancel RPC.
2618 * \param[in] req request which is being processed.
2620 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2622 ofd_punch_hpreq_check(req);
2625 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2626 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2627 .hpreq_check = ofd_rw_hpreq_check,
2628 .hpreq_fini = ofd_rw_hpreq_fini
2631 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2632 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2633 .hpreq_check = ofd_punch_hpreq_check,
2634 .hpreq_fini = ofd_punch_hpreq_fini
2638 * Assign high priority operations to an IO request.
2640 * Check if the incoming request is a candidate for
2641 * high-priority processing. If it is, assign it a high
2642 * priority operations table.
2644 * \param[in] tsi target session environment for this request
2646 static void ofd_hp_brw(struct tgt_session_info *tsi)
2648 struct niobuf_remote *rnb;
2649 struct obd_ioobj *ioo;
2653 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2654 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2655 if (ioo->ioo_bufcnt > 0) {
2656 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2657 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2659 /* no high priority if server lock is needed */
2660 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2661 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2665 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2669 * Assign high priority operations to an punch request.
2671 * Check if the incoming request is a candidate for
2672 * high-priority processing. If it is, assign it a high
2673 * priority operations table.
2675 * \param[in] tsi target session environment for this request
2677 static void ofd_hp_punch(struct tgt_session_info *tsi)
2679 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2680 /* no high-priority if server lock is needed */
2681 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2682 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2683 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2684 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2686 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2689 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2690 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2691 #define OST_BRW_READ OST_READ
2692 #define OST_BRW_WRITE OST_WRITE
2695 * Table of OFD-specific request handlers
2697 * This table contains all opcodes accepted by OFD and
2698 * specifies handlers for them. The tgt_request_handler()
2699 * uses such table from each target to process incoming
2702 static struct tgt_handler ofd_tgt_handlers[] = {
2703 TGT_RPC_HANDLER(OST_FIRST_OPC,
2704 0, OST_CONNECT, tgt_connect,
2705 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2706 TGT_RPC_HANDLER(OST_FIRST_OPC,
2707 0, OST_DISCONNECT, tgt_disconnect,
2708 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2709 TGT_RPC_HANDLER(OST_FIRST_OPC,
2710 0, OST_SET_INFO, ofd_set_info_hdl,
2711 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2712 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2713 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
2714 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2715 OST_SETATTR, ofd_setattr_hdl),
2716 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2717 OST_CREATE, ofd_create_hdl),
2718 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2719 OST_DESTROY, ofd_destroy_hdl),
2720 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
2721 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2722 OST_BRW_READ, tgt_brw_read,
2724 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2725 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
2727 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2728 OST_PUNCH, ofd_punch_hdl,
2730 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
2731 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
2732 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_LADVISE, ofd_ladvise_hdl),
2735 static struct tgt_opc_slice ofd_common_slice[] = {
2737 .tos_opc_start = OST_FIRST_OPC,
2738 .tos_opc_end = OST_LAST_OPC,
2739 .tos_hs = ofd_tgt_handlers
2742 .tos_opc_start = OBD_FIRST_OPC,
2743 .tos_opc_end = OBD_LAST_OPC,
2744 .tos_hs = tgt_obd_handlers
2747 .tos_opc_start = LDLM_FIRST_OPC,
2748 .tos_opc_end = LDLM_LAST_OPC,
2749 .tos_hs = tgt_dlm_handlers
2752 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2753 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2754 .tos_hs = tgt_out_handlers
2757 .tos_opc_start = SEQ_FIRST_OPC,
2758 .tos_opc_end = SEQ_LAST_OPC,
2759 .tos_hs = seq_handlers
2762 .tos_opc_start = LFSCK_FIRST_OPC,
2763 .tos_opc_end = LFSCK_LAST_OPC,
2764 .tos_hs = tgt_lfsck_handlers
2767 .tos_opc_start = SEC_FIRST_OPC,
2768 .tos_opc_end = SEC_LAST_OPC,
2769 .tos_hs = tgt_sec_ctx_handlers
2776 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2777 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2780 * Implementation of lu_context_key::lct_key_exit.
2782 * Optional method called on lu_context_exit() for all allocated
2784 * It is used in OFD to sanitize context values which may be re-used
2785 * during another request processing by the same thread.
2787 * \param[in] ctx execution context
2788 * \param[in] key context key
2789 * \param[in] data ofd_thread_info
2791 static void ofd_key_exit(const struct lu_context *ctx,
2792 struct lu_context_key *key, void *data)
2794 struct ofd_thread_info *info = data;
2796 info->fti_env = NULL;
2797 info->fti_exp = NULL;
2800 info->fti_pre_version = 0;
2802 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2805 struct lu_context_key ofd_thread_key = {
2806 .lct_tags = LCT_DT_THREAD,
2807 .lct_init = ofd_key_init,
2808 .lct_fini = ofd_key_fini,
2809 .lct_exit = ofd_key_exit
2813 * Initialize OFD device according to parameters in the config log \a cfg.
2815 * This is the main starting point of OFD initialization. It fills all OFD
2816 * parameters with their initial values and calls other initializing functions
2817 * to set up all OFD subsystems.
2819 * \param[in] env execution environment
2820 * \param[in] m OFD device
2821 * \param[in] ldt LU device type of OFD
2822 * \param[in] cfg configuration log
2824 * \retval 0 if successful
2825 * \retval negative value on error
2827 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2828 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2830 const char *dev = lustre_cfg_string(cfg, 0);
2831 struct ofd_thread_info *info = NULL;
2832 struct obd_device *obd;
2833 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2835 struct nm_config_file *nodemap_config;
2836 struct obd_device_target *obt;
2841 obd = class_name2obd(dev);
2843 CERROR("Cannot find obd with name %s\n", dev);
2847 rc = lu_env_refill((struct lu_env *)env);
2852 obt->obt_magic = OBT_MAGIC;
2854 spin_lock_init(&m->ofd_flags_lock);
2855 m->ofd_raid_degraded = 0;
2856 m->ofd_checksum_t10pi_enforce = 0;
2857 m->ofd_sync_journal = 0;
2859 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2861 m->ofd_seq_count = 0;
2862 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2863 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2864 spin_lock_init(&m->ofd_inconsistency_lock);
2866 spin_lock_init(&m->ofd_batch_lock);
2867 init_rwsem(&m->ofd_lastid_rwsem);
2869 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2870 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2871 /* set this lu_device to obd, because error handling need it */
2872 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2874 rc = ofd_tunables_init(m);
2876 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2880 /* No connection accepted until configurations will finish */
2881 spin_lock(&obd->obd_dev_lock);
2882 obd->obd_no_conn = 1;
2883 spin_unlock(&obd->obd_dev_lock);
2884 obd->obd_replayable = 1;
2885 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2886 char *str = lustre_cfg_string(cfg, 4);
2888 if (strchr(str, 'n')) {
2889 CWARN("%s: recovery disabled\n", obd->obd_name);
2890 obd->obd_replayable = 0;
2894 info = ofd_info_init(env, NULL);
2896 GOTO(err_fini_proc, rc = -EFAULT);
2898 rc = ofd_stack_init(env, m, cfg);
2900 CERROR("Can't init device stack, rc %d\n", rc);
2901 GOTO(err_fini_proc, rc);
2904 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
2905 ofd_procfs_add_brw_stats_symlink(m);
2908 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2909 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2910 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2911 LDLM_NAMESPACE_SERVER,
2912 LDLM_NAMESPACE_GREEDY,
2914 if (m->ofd_namespace == NULL)
2915 GOTO(err_fini_stack, rc = -ENOMEM);
2916 /* set obd_namespace for compatibility with old code */
2917 obd->obd_namespace = m->ofd_namespace;
2918 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2919 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2920 m->ofd_namespace->ns_lvbp = m;
2922 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2923 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2925 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2926 OBD_FAIL_OST_ALL_REQUEST_NET,
2927 OBD_FAIL_OST_ALL_REPLY_NET);
2929 GOTO(err_free_ns, rc);
2931 tgd->tgd_reserved_pcnt = 0;
2933 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
2934 m->ofd_cksum_types_supported =
2935 obd_cksum_types_supported_server(obd->obd_name);
2936 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2937 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
2938 OFD_PRECREATE_SMALL_FS)
2939 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2941 rc = ofd_fs_setup(env, m, obd);
2943 GOTO(err_fini_lut, rc);
2945 fid.f_seq = FID_SEQ_LOCAL_NAME;
2948 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
2951 GOTO(err_fini_fs, rc);
2953 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
2955 if (IS_ERR(nodemap_config)) {
2956 rc = PTR_ERR(nodemap_config);
2958 GOTO(err_fini_los, rc);
2960 obt->obt_nodemap_config_file = nodemap_config;
2963 rc = ofd_start_inconsistency_verification_thread(m);
2965 GOTO(err_fini_nm, rc);
2967 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
2972 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
2973 obt->obt_nodemap_config_file = NULL;
2975 local_oid_storage_fini(env, m->ofd_los);
2978 ofd_fs_cleanup(env, m);
2980 tgt_fini(env, &m->ofd_lut);
2982 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
2983 obd->obd_namespace = m->ofd_namespace = NULL;
2985 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
2992 * Stop the OFD device
2994 * This function stops the OFD device and all its subsystems.
2995 * This is the end of OFD lifecycle.
2997 * \param[in] env execution environment
2998 * \param[in] m OFD device
3000 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3002 struct obd_device *obd = ofd_obd(m);
3003 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3004 struct lfsck_stop stop;
3006 stop.ls_status = LS_PAUSED;
3008 lfsck_stop(env, m->ofd_osd, &stop);
3009 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3010 target_recovery_fini(obd);
3011 if (m->ofd_namespace != NULL)
3012 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3013 d->ld_obd->obd_force);
3015 obd_exports_barrier(obd);
3016 obd_zombie_barrier();
3018 tgt_fini(env, &m->ofd_lut);
3019 ofd_stop_inconsistency_verification_thread(m);
3020 lfsck_degister(env, m->ofd_osd);
3021 ofd_fs_cleanup(env, m);
3022 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3023 obd->u.obt.obt_nodemap_config_file = NULL;
3025 if (m->ofd_namespace != NULL) {
3026 ldlm_namespace_free_post(m->ofd_namespace);
3027 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3030 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3032 LASSERT(atomic_read(&d->ld_ref) == 0);
3033 server_put_mount(obd->obd_name, true);
3038 * Implementation of lu_device_type_operations::ldto_device_fini.
3040 * Finalize device. Dual to ofd_device_init(). It is called from
3041 * obd_precleanup() and stops the current device.
3043 * \param[in] env execution environment
3044 * \param[in] d LU device of OFD
3048 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3049 struct lu_device *d)
3052 ofd_fini(env, ofd_dev(d));
3057 * Implementation of lu_device_type_operations::ldto_device_free.
3059 * Free OFD device. Dual to ofd_device_alloc().
3061 * \param[in] env execution environment
3062 * \param[in] d LU device of OFD
3066 static struct lu_device *ofd_device_free(const struct lu_env *env,
3067 struct lu_device *d)
3069 struct ofd_device *m = ofd_dev(d);
3071 dt_device_fini(&m->ofd_dt_dev);
3077 * Implementation of lu_device_type_operations::ldto_device_alloc.
3079 * This function allocates the new OFD device. It is called from
3080 * obd_setup() if OBD device had lu_device_type defined.
3082 * \param[in] env execution environment
3083 * \param[in] t lu_device_type of OFD device
3084 * \param[in] cfg configuration log
3086 * \retval pointer to the lu_device of just allocated OFD
3087 * \retval ERR_PTR of return value on error
3089 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3090 struct lu_device_type *t,
3091 struct lustre_cfg *cfg)
3093 struct ofd_device *m;
3094 struct lu_device *l;
3099 return ERR_PTR(-ENOMEM);
3101 l = &m->ofd_dt_dev.dd_lu_dev;
3102 dt_device_init(&m->ofd_dt_dev, t);
3103 rc = ofd_init0(env, m, t, cfg);
3105 ofd_device_free(env, l);
3112 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3113 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3115 static struct lu_device_type_operations ofd_device_type_ops = {
3116 .ldto_init = ofd_type_init,
3117 .ldto_fini = ofd_type_fini,
3119 .ldto_start = ofd_type_start,
3120 .ldto_stop = ofd_type_stop,
3122 .ldto_device_alloc = ofd_device_alloc,
3123 .ldto_device_free = ofd_device_free,
3124 .ldto_device_fini = ofd_device_fini
3127 static struct lu_device_type ofd_device_type = {
3128 .ldt_tags = LU_DEVICE_DT,
3129 .ldt_name = LUSTRE_OST_NAME,
3130 .ldt_ops = &ofd_device_type_ops,
3131 .ldt_ctx_tags = LCT_DT_THREAD
3135 * Initialize OFD module.
3137 * This function is called upon module loading. It registers OFD device type
3138 * and prepares all in-memory structures used by all OFD devices.
3140 * \retval 0 if successful
3141 * \retval negative value on error
3143 static int __init ofd_init(void)
3147 rc = lu_kmem_init(ofd_caches);
3150 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3151 LUSTRE_OST_NAME, &ofd_device_type);
3158 * This function is called upon OFD module unloading.
3159 * It frees all related structures and unregisters OFD device type.
3161 static void __exit ofd_exit(void)
3163 lu_kmem_fini(ofd_caches);
3164 class_unregister_type(LUSTRE_OST_NAME);
3167 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3168 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3169 MODULE_VERSION(LUSTRE_VERSION_STRING);
3170 MODULE_LICENSE("GPL");
3172 module_init(ofd_init);
3173 module_exit(ofd_exit);