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>
80 #include <linux/falloc.h>
82 #include "ofd_internal.h"
84 /* Slab for OFD object allocation */
85 static struct kmem_cache *ofd_object_kmem;
86 static struct lu_kmem_descr ofd_caches[] = {
88 .ckd_cache = &ofd_object_kmem,
89 .ckd_name = "ofd_obj",
90 .ckd_size = sizeof(struct ofd_object)
98 * Connect OFD to the next device in the stack.
100 * This function is used for device stack configuration and links OFD
101 * device with bottom OSD device.
103 * \param[in] env execution environment
104 * \param[in] m OFD device
105 * \param[in] next name of next device in the stack
106 * \param[out] exp export to return
108 * \retval 0 and export in \a exp if successful
109 * \retval negative value on error
111 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
112 const char *next, struct obd_export **exp)
114 struct obd_connect_data *data = NULL;
115 struct obd_device *obd;
121 GOTO(out, rc = -ENOMEM);
123 obd = class_name2obd(next);
125 CERROR("%s: can't locate next device: %s\n",
127 GOTO(out, rc = -ENOTCONN);
130 data->ocd_connect_flags = OBD_CONNECT_VERSION;
131 data->ocd_version = LUSTRE_VERSION_CODE;
133 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
135 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
136 ofd_name(m), next, rc);
140 m->ofd_dt_dev.dd_lu_dev.ld_site =
141 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
142 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
143 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
144 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
153 * Initialize stack of devices.
155 * This function initializes OFD-OSD device stack to serve OST requests
157 * \param[in] env execution environment
158 * \param[in] m OFD device
159 * \param[in] cfg Lustre config for this server
161 * \retval 0 if successful
162 * \retval negative value on error
164 static int ofd_stack_init(const struct lu_env *env,
165 struct ofd_device *m, struct lustre_cfg *cfg,
168 const char *dev = lustre_cfg_string(cfg, 0);
170 struct ofd_thread_info *info = ofd_info(env);
171 struct lustre_mount_info *lmi;
172 struct lustre_mount_data *lmd;
178 lmi = server_get_mount(dev);
180 CERROR("Cannot get mount info for %s!\n", dev);
184 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
186 if (lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
187 m->ofd_skip_lfsck = 1;
188 if (lmd->lmd_flags & LMD_FLG_NO_PRECREATE)
189 m->ofd_no_precreate = 1;
190 *lmd_flags = lmd->lmd_flags;
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 lu_site_print(env, top->ld_site, &top->ld_site->ls_obj_hash.nelems,
261 D_OTHER, lu_cdebug_printer);
262 LASSERT(m->ofd_osd_exp);
263 obd_disconnect(m->ofd_osd_exp);
268 static void ofd_stack_pre_fini(const struct lu_env *env, struct ofd_device *m,
269 struct lu_device *top)
271 struct lustre_cfg_bufs bufs;
272 struct lustre_cfg *lcfg;
277 lustre_cfg_bufs_reset(&bufs, ofd_name(m));
278 lustre_cfg_bufs_set_string(&bufs, 1, NULL);
279 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
281 CERROR("%s: failed to trigger LCFG_PRE_CLEANUP\n", ofd_name(m));
283 lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, &bufs);
284 top->ld_ops->ldo_process_config(env, top, lcfg);
285 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
286 lcfg->lcfg_buflens));
292 /* For interoperability, see mdt_interop_param[]. */
293 static struct cfg_interop_param ofd_interop_param[] = {
294 { "ost.quota_type", NULL },
299 * Check if parameters are symlinks to the OSD.
301 * Some parameters were moved from ofd to osd and only their
302 * symlinks were kept in ofd by LU-3106. They are:
303 * -writehthrough_cache_enable
304 * -readcache_max_filesize
308 * Since they are not included by the static lprocfs var list, a pre-check
309 * is added for them to avoid "unknown param" errors. If they are matched
310 * in this check, they will be passed to the OSD directly.
312 * \param[in] param parameters to check
314 * \retval true if param is symlink to OSD param
317 static bool match_symlink_param(char *param)
322 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
323 sval = strchr(param, '=');
325 paramlen = sval - param;
326 if (strncmp(param, "brw_stats", paramlen) == 0)
335 * Process various configuration parameters.
337 * This function is used by MGS to process specific configurations and
338 * pass them through to the next device in server stack, i.e. the OSD.
340 * \param[in] env execution environment
341 * \param[in] d LU device of OFD
342 * \param[in] cfg parameters to process
344 * \retval 0 if successful
345 * \retval negative value on error
347 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
348 struct lustre_cfg *cfg)
350 struct ofd_device *m = ofd_dev(d);
351 struct dt_device *dt_next = m->ofd_osd;
352 struct lu_device *next = &dt_next->dd_lu_dev;
357 switch (cfg->lcfg_command) {
359 /* For interoperability */
360 struct cfg_interop_param *ptr = NULL;
361 struct lustre_cfg *old_cfg = NULL;
365 param = lustre_cfg_string(cfg, 1);
367 CERROR("param is empty\n");
372 ptr = class_find_old_param(param, ofd_interop_param);
374 if (ptr->new_param == NULL) {
376 CWARN("For interoperability, skip this %s."
377 " It is obsolete.\n", ptr->old_param);
381 CWARN("Found old param %s, changed it to %s.\n",
382 ptr->old_param, ptr->new_param);
385 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
392 if (match_symlink_param(param)) {
393 rc = next->ld_ops->ldo_process_config(env, next, cfg);
397 count = class_modify_config(cfg, PARAM_OST,
398 &d->ld_obd->obd_kset.kobj);
403 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
405 /* we don't understand; pass it on */
406 rc = next->ld_ops->ldo_process_config(env, next, cfg);
409 case LCFG_SPTLRPC_CONF: {
414 /* others are passed further */
415 rc = next->ld_ops->ldo_process_config(env, next, cfg);
422 * Implementation of lu_object_operations::loo_object_init for OFD
424 * Allocate just the next object (OSD) in stack.
426 * \param[in] env execution environment
427 * \param[in] o lu_object of OFD object
428 * \param[in] conf additional configuration parameters, not used here
430 * \retval 0 if successful
431 * \retval negative value on error
433 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
434 const struct lu_object_conf *conf)
436 struct ofd_device *d = ofd_dev(o->lo_dev);
437 struct lu_device *under;
438 struct lu_object *below;
443 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
444 PFID(lu_object_fid(o)));
446 under = &d->ofd_osd->dd_lu_dev;
447 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
449 lu_object_add(o, below);
456 static void ofd_object_free_rcu(struct rcu_head *head)
458 struct ofd_object *of = container_of(head, struct ofd_object,
461 kmem_cache_free(ofd_object_kmem, of);
465 * Implementation of lu_object_operations::loo_object_free.
467 * Finish OFD object lifecycle and free its memory.
469 * \param[in] env execution environment
470 * \param[in] o LU object of OFD object
472 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
474 struct ofd_object *of = ofd_obj(o);
475 struct lu_object_header *h;
480 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
481 PFID(lu_object_fid(o)));
484 lu_object_header_fini(h);
485 OBD_FREE_PRE(of, sizeof(*of), "slab-freed");
486 call_rcu(&of->ofo_header.loh_rcu, ofd_object_free_rcu);
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 const 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;
548 range_lock_tree_init(&of->ofo_write_tree);
556 * Return the result of LFSCK run to the OFD.
558 * Notify OFD about result of LFSCK run. That may block the new object
559 * creation until problem is fixed by LFSCK.
561 * \param[in] env execution environment
562 * \param[in] data pointer to the OFD device
563 * \param[in] event LFSCK event type
565 * \retval 0 if successful
566 * \retval negative value on unknown event
568 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
569 enum lfsck_events event)
571 struct ofd_device *ofd = data;
572 struct obd_device *obd = ofd_obd(ofd);
575 case LE_LASTID_REBUILDING:
576 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
577 "on the device until the LAST_ID rebuilt successfully.\n",
579 down_write(&ofd->ofd_lastid_rwsem);
580 ofd->ofd_lastid_rebuilding = 1;
581 up_write(&ofd->ofd_lastid_rwsem);
583 case LE_LASTID_REBUILT: {
584 down_write(&ofd->ofd_lastid_rwsem);
585 ofd_seqs_free(env, ofd);
586 ofd->ofd_lastid_rebuilding = 0;
587 ofd->ofd_lastid_gen++;
588 up_write(&ofd->ofd_lastid_rwsem);
589 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
594 CERROR("%s: unknown lfsck event: rc = %d\n",
595 ofd_name(ofd), event);
603 * Implementation of lu_device_operations::ldo_prepare.
605 * This method is called after layer has been initialized and before it starts
606 * serving user requests. In OFD it starts lfsk check routines and initializes
609 * \param[in] env execution environment
610 * \param[in] pdev higher device in stack, NULL for OFD
611 * \param[in] dev lu_device of OFD device
613 * \retval 0 if successful
614 * \retval negative value on error
616 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
617 struct lu_device *dev)
619 struct ofd_thread_info *info;
620 struct ofd_device *ofd = ofd_dev(dev);
621 struct obd_device *obd = ofd_obd(ofd);
622 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
627 info = ofd_info_init(env, NULL);
631 /* initialize lower device */
632 rc = next->ld_ops->ldo_prepare(env, dev, next);
636 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
637 ofd_lfsck_out_notify, ofd, false);
639 CERROR("%s: failed to initialize lfsck: rc = %d\n",
644 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
645 /* The LFSCK instance is registered just now, so it must be there when
646 * register the namespace to such instance. */
647 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
649 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
650 OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_OST_PREPARE_DELAY, OBD_FAIL_ONCE,
651 (OBD_TIMEOUT_DEFAULT + 1) / 4);
652 LASSERT(obd->obd_no_conn);
653 spin_lock(&obd->obd_dev_lock);
654 obd->obd_no_conn = 0;
655 spin_unlock(&obd->obd_dev_lock);
657 if (obd->obd_recovering == 0)
658 ofd_postrecov(env, ofd);
664 * Implementation of lu_device_operations::ldo_recovery_complete.
666 * This method notifies all layers about 'recovery complete' event. That means
667 * device is in full state and consistent. An OFD calculates available grant
668 * space upon this event.
670 * \param[in] env execution environment
671 * \param[in] dev lu_device of OFD device
673 * \retval 0 if successful
674 * \retval negative value on error
676 static int ofd_recovery_complete(const struct lu_env *env,
677 struct lu_device *dev)
679 struct ofd_thread_info *oti = ofd_info(env);
680 struct ofd_device *ofd = ofd_dev(dev);
681 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
687 * Grant space for object precreation on the self export.
688 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
689 * is enough to create 10k objects. More space is then acquired for
690 * precreation in tgt_grant_create().
692 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
693 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
694 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
695 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
696 OBD_CONNECT_GRANT_PARAM;
697 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
699 rc = next->ld_ops->ldo_recovery_complete(env, next);
704 * lu_device_operations matrix for OFD device.
706 static const struct lu_device_operations ofd_lu_ops = {
707 .ldo_object_alloc = ofd_object_alloc,
708 .ldo_process_config = ofd_process_config,
709 .ldo_recovery_complete = ofd_recovery_complete,
710 .ldo_prepare = ofd_prepare,
713 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
715 * Expose OSD statistics to OFD layer.
717 * The osd interfaces to the backend file system exposes useful data
718 * such as brw_stats and read or write cache states. This same data
719 * needs to be exposed into the obdfilter (ofd) layer to maintain
720 * backwards compatibility. This function creates the symlinks in the
721 * proc layer to enable this.
723 * \param[in] ofd OFD device
725 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
727 struct obd_device *obd = ofd_obd(ofd);
728 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
729 struct kobj_type *osd_type;
732 osd_type = get_ktype(&ofd->ofd_osd->dd_kobj);
733 for (i = 0; osd_type->default_attrs[i]; i++) {
734 if (strcmp(osd_type->default_attrs[i]->name,
735 "read_cache_enable") == 0) {
736 ofd->ofd_read_cache_enable =
737 osd_type->default_attrs[i];
740 if (strcmp(osd_type->default_attrs[i]->name,
741 "readcache_max_filesize") == 0) {
742 ofd->ofd_read_cache_max_filesize =
743 osd_type->default_attrs[i];
746 if (strcmp(osd_type->default_attrs[i]->name,
747 "writethrough_cache_enable") == 0) {
748 ofd->ofd_write_cache_enable =
749 osd_type->default_attrs[i];
753 if (obd->obd_proc_entry == NULL)
756 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
757 "../../%s/%s/brw_stats",
758 osd_obd->obd_type->typ_name, obd->obd_name);
763 * Cleanup all procfs entries in OFD.
765 * \param[in] ofd OFD device
767 static void ofd_procfs_fini(struct ofd_device *ofd)
769 struct obd_device *obd = ofd_obd(ofd);
771 tgt_tunables_fini(&ofd->ofd_lut);
772 lprocfs_free_per_client_stats(obd);
773 lprocfs_obd_cleanup(obd);
774 lprocfs_free_obd_stats(obd);
775 lprocfs_job_stats_fini(obd);
779 * Stop SEQ/FID server on OFD.
781 * \param[in] env execution environment
782 * \param[in] ofd OFD device
784 * \retval 0 if successful
785 * \retval negative value on error
787 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
789 return seq_site_fini(env, &ofd->ofd_seq_site);
793 * Start SEQ/FID server on OFD.
795 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
796 * It also connects to the master server to get own FID sequence (SEQ) range
797 * to this particular OFD. Typically that happens when the OST is first
798 * formatted or in the rare case that it exhausts the local sequence range.
800 * The sequence range is allocated out to the MDTs for OST object allocations,
801 * and not directly to the clients.
803 * \param[in] env execution environment
804 * \param[in] ofd OFD device
806 * \retval 0 if successful
807 * \retval negative value on error
809 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
811 struct seq_server_site *ss = &ofd->ofd_seq_site;
812 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
813 char *obd_name = ofd_name(ofd);
815 int len = strlen(obd_name) + 7;
818 ss = &ofd->ofd_seq_site;
819 lu->ld_site->ld_seq_site = ss;
820 ss->ss_lu = lu->ld_site;
821 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
823 OBD_ALLOC(name, len);
827 OBD_ALLOC_PTR(ss->ss_server_seq);
828 if (ss->ss_server_seq == NULL)
829 GOTO(out_name, rc = -ENOMEM);
831 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
832 LUSTRE_SEQ_SERVER, ss);
834 CERROR("%s: seq server init error: rc = %d\n", obd_name, rc);
835 GOTO(out_server, rc);
837 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
839 OBD_ALLOC_PTR(ss->ss_client_seq);
840 if (ss->ss_client_seq == NULL)
841 GOTO(out_server, rc = -ENOMEM);
843 snprintf(name, len, "%s-super", obd_name);
844 seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
847 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
850 seq_client_fini(ss->ss_client_seq);
851 OBD_FREE_PTR(ss->ss_client_seq);
852 ss->ss_client_seq = NULL;
854 seq_server_fini(ss->ss_server_seq, env);
855 OBD_FREE_PTR(ss->ss_server_seq);
856 ss->ss_server_seq = NULL;
865 * OFD request handler for OST_SET_INFO RPC.
867 * This is OFD-specific part of request handling
869 * \param[in] tsi target session environment for this request
871 * \retval 0 if successful
872 * \retval negative value on error
874 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
876 struct ptlrpc_request *req = tgt_ses_req(tsi);
877 struct ost_body *body = NULL, *repbody;
878 void *key, *val = NULL;
879 int keylen, vallen, rc = 0;
880 bool is_grant_shrink;
881 ktime_t kstart = ktime_get();
885 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
887 DEBUG_REQ(D_HA, req, "no set_info key");
888 RETURN(err_serious(-EFAULT));
890 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
893 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
895 DEBUG_REQ(D_HA, req, "no set_info val");
896 RETURN(err_serious(-EFAULT));
898 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
901 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
903 /* In this case the value is actually an RMF_OST_BODY, so we
904 * transmutate the type of this PTLRPC */
905 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
907 rc = req_capsule_server_pack(tsi->tsi_pill);
911 if (is_grant_shrink) {
912 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
914 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
917 /** handle grant shrink, similar to a read request */
918 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
920 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
922 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
925 CERROR("%s: Unsupported key %s\n",
926 tgt_name(tsi->tsi_tgt), (char *)key);
929 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
930 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
936 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
938 * This function returns a list of extents which describes how a file's
939 * blocks are laid out on the disk.
941 * \param[in] env execution environment
942 * \param[in] ofd OFD device
943 * \param[in] fid FID of object
944 * \param[in] fiemap fiemap structure to fill with data
946 * \retval 0 if \a fiemap is filled with data successfully
947 * \retval negative value on error
949 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
950 struct lu_fid *fid, struct fiemap *fiemap)
952 struct ofd_object *fo;
955 fo = ofd_object_find(env, ofd, fid);
957 CERROR("%s: error finding object "DFID"\n",
958 ofd_name(ofd), PFID(fid));
962 ofd_read_lock(env, fo);
963 if (ofd_object_exists(fo))
964 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
967 ofd_read_unlock(env, fo);
968 ofd_object_put(env, fo);
973 static int ofd_lock_unlock_region(const struct lu_env *env,
974 struct ldlm_namespace *ns,
975 struct ldlm_res_id *res_id,
976 unsigned long long begin,
977 unsigned long long end)
981 struct lustre_handle lh = { 0 };
983 LASSERT(begin <= end);
985 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
989 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
990 tgt_data_unlock(&lh, LCK_PR);
996 * Lock the sparse areas of given resource.
998 * The locking of sparse areas will cause dirty data to be flushed back from
999 * clients. This is used when getting the FIEMAP of an object to make sure
1000 * there is no unaccounted cached data on clients.
1002 * This function goes through \a fiemap list of extents and locks only sparse
1003 * areas between extents.
1005 * \param[in] ns LDLM namespace
1006 * \param[in] res_id resource ID
1007 * \param[in] fiemap file extents mapping on disk
1008 * \param[in] locked list head of regions list
1010 * \retval 0 if successful
1011 * \retval negative value on error
1013 static int lock_zero_regions(const struct lu_env *env,
1014 struct ldlm_namespace *ns,
1015 struct ldlm_res_id *res_id,
1016 struct fiemap *fiemap)
1018 __u64 begin = fiemap->fm_start;
1021 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1025 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1026 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1027 if (fiemap_start[i].fe_logical > begin) {
1028 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1029 begin, fiemap_start[i].fe_logical);
1030 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1031 fiemap_start[i].fe_logical);
1036 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1039 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1040 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1041 begin, fiemap->fm_start + fiemap->fm_length);
1042 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1043 fiemap->fm_start + fiemap->fm_length);
1051 * OFD request handler for OST_GET_INFO RPC.
1053 * This is OFD-specific part of request handling. The OFD-specific keys are:
1054 * - KEY_LAST_ID (obsolete)
1058 * This function reads needed data from storage and fills reply with it.
1060 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1061 * and is kept for compatibility.
1063 * \param[in] tsi target session environment for this request
1065 * \retval 0 if successful
1066 * \retval negative value on error
1068 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1070 struct obd_export *exp = tsi->tsi_exp;
1071 struct ofd_device *ofd = ofd_exp(exp);
1072 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1075 int replylen, rc = 0;
1076 ktime_t kstart = ktime_get();
1080 /* this common part for get_info rpc */
1081 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1083 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1084 RETURN(err_serious(-EPROTO));
1086 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1089 if (KEY_IS(KEY_LAST_ID)) {
1091 struct ofd_seq *oseq;
1093 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1094 rc = req_capsule_server_pack(tsi->tsi_pill);
1096 RETURN(err_serious(rc));
1098 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1100 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1101 (u64)exp->exp_filter_data.fed_group);
1105 *last_id = ofd_seq_last_oid(oseq);
1106 ofd_seq_put(tsi->tsi_env, oseq);
1107 } else if (KEY_IS(KEY_FIEMAP)) {
1108 struct ll_fiemap_info_key *fm_key;
1109 struct fiemap *fiemap;
1112 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1114 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1115 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1117 RETURN(err_serious(rc));
1119 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1121 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1123 replylen = fiemap_count_to_size(
1124 fm_key->lfik_fiemap.fm_extent_count);
1125 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1126 RCL_SERVER, replylen);
1128 rc = req_capsule_server_pack(tsi->tsi_pill);
1130 RETURN(err_serious(rc));
1132 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1136 *fiemap = fm_key->lfik_fiemap;
1137 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1139 /* LU-3219: Lock the sparse areas to make sure dirty
1140 * flushed back from client, then call fiemap again. */
1141 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1142 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1143 ost_fid_build_resid(fid, &fti->fti_resid);
1144 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1145 &fti->fti_resid, fiemap);
1147 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1150 } else if (KEY_IS(KEY_LAST_FID)) {
1151 struct ofd_device *ofd = ofd_exp(exp);
1152 struct ofd_seq *oseq;
1156 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1157 rc = req_capsule_server_pack(tsi->tsi_pill);
1159 RETURN(err_serious(rc));
1161 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1163 RETURN(err_serious(-EPROTO));
1165 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1167 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1171 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1172 ostid_seq(&fti->fti_ostid));
1174 RETURN(PTR_ERR(oseq));
1176 rc = ostid_to_fid(fid, &oseq->os_oi,
1177 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1181 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1184 ofd_seq_put(tsi->tsi_env, oseq);
1186 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1190 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1191 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1197 * OFD request handler for OST_GETATTR RPC.
1199 * This is OFD-specific part of request handling. It finds the OFD object
1200 * by its FID, gets attributes from storage and packs result to the reply.
1202 * \param[in] tsi target session environment for this request
1204 * \retval 0 if successful
1205 * \retval negative value on error
1207 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1209 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1210 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1211 struct ost_body *repbody;
1212 struct lustre_handle lh = { 0 };
1213 struct ofd_object *fo;
1215 enum ldlm_mode lock_mode = LCK_PR;
1216 ktime_t kstart = ktime_get();
1221 LASSERT(tsi->tsi_ost_body != NULL);
1223 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1224 if (repbody == NULL)
1227 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1228 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1230 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1231 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1234 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1237 rc = tgt_extent_lock(tsi->tsi_env,
1238 tsi->tsi_tgt->lut_obd->obd_namespace,
1239 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1245 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1247 GOTO(out, rc = PTR_ERR(fo));
1249 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1253 obdo_from_la(&repbody->oa, &fti->fti_attr,
1254 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1256 /* Store object version in reply */
1257 curr_version = dt_version_get(tsi->tsi_env,
1258 ofd_object_child(fo));
1259 if ((__s64)curr_version != -EOPNOTSUPP) {
1260 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1261 repbody->oa.o_data_version = curr_version;
1264 if (fo->ofo_ff.ff_layout_version > 0) {
1265 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1266 repbody->oa.o_layout_version =
1267 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1269 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1270 PFID(&tsi->tsi_fid),
1271 repbody->oa.o_layout_version);
1275 ofd_object_put(tsi->tsi_env, fo);
1278 tgt_data_unlock(&lh, lock_mode);
1280 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1281 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1283 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1284 repbody->oa.o_flags = OBD_FL_FLUSH;
1290 * OFD request handler for OST_SETATTR RPC.
1292 * This is OFD-specific part of request handling. It finds the OFD object
1293 * by its FID, sets attributes from request and packs result to the reply.
1295 * \param[in] tsi target session environment for this request
1297 * \retval 0 if successful
1298 * \retval negative value on error
1300 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1302 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1303 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1304 struct ost_body *body = tsi->tsi_ost_body;
1305 struct ost_body *repbody;
1306 struct ldlm_resource *res;
1307 struct ofd_object *fo;
1308 ktime_t kstart = ktime_get();
1313 LASSERT(body != NULL);
1315 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1316 if (repbody == NULL)
1319 repbody->oa.o_oi = body->oa.o_oi;
1320 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1322 /* This would be very bad - accidentally truncating a file when
1323 * changing the time or similar - bug 12203. */
1324 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1325 body->oa.o_size != OBD_OBJECT_EOF) {
1326 static char mdsinum[48];
1328 if (body->oa.o_valid & OBD_MD_FLFID)
1329 snprintf(mdsinum, sizeof(mdsinum) - 1,
1330 "of parent "DFID, body->oa.o_parent_seq,
1331 body->oa.o_parent_oid, 0);
1335 CERROR("%s: setattr from %s is trying to truncate object "DFID
1336 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1337 PFID(&tsi->tsi_fid), mdsinum);
1341 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1343 GOTO(out, rc = PTR_ERR(fo));
1345 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1346 fti->fti_attr.la_valid &= ~LA_TYPE;
1348 /* setting objects attributes (including owner/group) */
1349 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1353 obdo_from_la(&repbody->oa, &fti->fti_attr,
1354 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1356 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1357 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1360 ofd_object_put(tsi->tsi_env, fo);
1363 /* we do not call this before to avoid lu_object_find() in
1364 * ->lvbo_update() holding another reference on the object.
1365 * otherwise concurrent destroy can make the object unavailable
1366 * for 2nd lu_object_find() waiting for the first reference
1367 * to go... deadlock! */
1368 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1369 &tsi->tsi_resid, LDLM_EXTENT, 0);
1371 ldlm_res_lvbo_update(res, NULL, 0);
1372 ldlm_resource_putref(res);
1379 * Destroy OST orphans.
1381 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1382 * set then we must destroy possible orphaned objects.
1384 * \param[in] env execution environment
1385 * \param[in] exp OBD export
1386 * \param[in] ofd OFD device
1387 * \param[in] oa obdo structure for reply
1389 * \retval 0 if successful
1390 * \retval negative value on error
1392 static int ofd_orphans_destroy(const struct lu_env *env,
1393 struct obd_export *exp,
1394 struct ofd_device *ofd, struct obdo *oa)
1396 struct ofd_thread_info *info = ofd_info(env);
1397 struct lu_fid *fid = &info->fti_fid;
1398 struct ost_id *oi = &oa->o_oi;
1399 struct ofd_seq *oseq;
1400 u64 seq = ostid_seq(oi);
1401 u64 end_id = ostid_id(oi);
1409 oseq = ofd_seq_get(ofd, seq);
1411 CERROR("%s: Can not find seq for "DOSTID"\n",
1412 ofd_name(ofd), POSTID(oi));
1417 last = ofd_seq_last_oid(oseq);
1420 LASSERT(exp != NULL);
1421 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1423 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1426 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1427 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1429 while (oid > end_id) {
1430 rc = fid_set_id(fid, oid);
1431 if (unlikely(rc != 0))
1434 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1435 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1436 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1437 /* this is pretty fatal... */
1438 CEMERG("%s: error destroying precreated id "
1440 ofd_name(ofd), PFID(fid), rc);
1444 ofd_seq_last_oid_set(oseq, oid);
1445 /* update last_id on disk periodically so that if we
1446 * restart * we don't need to re-scan all of the just
1447 * deleted objects. */
1448 if ((oid & 511) == 0)
1449 ofd_seq_last_oid_write(env, ofd, oseq);
1453 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1454 ofd_name(ofd), seq, oid);
1458 ofd_seq_last_oid_set(oseq, oid);
1459 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1461 /* don't reuse orphan object, return last used objid */
1462 rc = ostid_set_id(oi, last);
1468 ofd_seq_put(env, oseq);
1473 * OFD request handler for OST_CREATE RPC.
1475 * This is OFD-specific part of request handling. Its main purpose is to
1476 * create new data objects on OST, but it also used to destroy orphans.
1478 * \param[in] tsi target session environment for this request
1480 * \retval 0 if successful
1481 * \retval negative value on error
1483 static int ofd_create_hdl(struct tgt_session_info *tsi)
1485 struct ptlrpc_request *req = tgt_ses_req(tsi);
1486 struct ost_body *repbody;
1487 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1488 struct obdo *rep_oa;
1489 struct obd_export *exp = tsi->tsi_exp;
1490 struct ofd_device *ofd = ofd_exp(exp);
1491 u64 seq = ostid_seq(&oa->o_oi);
1492 u64 oid = ostid_id(&oa->o_oi);
1493 struct ofd_seq *oseq;
1496 ktime_t kstart = ktime_get();
1502 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1505 if (ofd->ofd_no_precreate)
1508 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1509 if (repbody == NULL)
1512 down_read(&ofd->ofd_lastid_rwsem);
1513 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1514 * we may do that in the future.
1515 * Return -ENOSPC until the LAST_ID rebuilt. */
1516 if (unlikely(ofd->ofd_lastid_rebuilding))
1517 GOTO(out_sem, rc = -ENOSPC);
1519 rep_oa = &repbody->oa;
1520 rep_oa->o_oi = oa->o_oi;
1522 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1524 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1526 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1528 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1529 ofd_name(ofd), seq, PTR_ERR(oseq));
1530 GOTO(out_sem, rc = -EINVAL);
1533 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1534 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1535 if (!ofd_obd(ofd)->obd_recovering ||
1536 oid > ofd_seq_last_oid(oseq)) {
1537 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1538 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1539 ofd_seq_last_oid(oseq));
1540 GOTO(out_nolock, rc = -EINVAL);
1542 /* Do nothing here, we re-create objects during recovery
1543 * upon write replay, see ofd_preprw_write() */
1544 GOTO(out_nolock, rc = 0);
1546 /* former ofd_handle_precreate */
1547 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1548 (oa->o_flags & OBD_FL_DELORPHAN)) {
1549 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1551 /* destroy orphans */
1552 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1553 exp->exp_conn_cnt) {
1554 CERROR("%s: dropping old orphan cleanup request\n",
1556 GOTO(out_nolock, rc = 0);
1558 /* This causes inflight precreates to abort and drop lock */
1559 oseq->os_destroys_in_progress = 1;
1560 mutex_lock(&oseq->os_create_lock);
1561 if (!oseq->os_destroys_in_progress) {
1563 "%s:[%llu] destroys_in_progress already cleared\n",
1564 ofd_name(ofd), seq);
1565 rc = ostid_set_id(&rep_oa->o_oi,
1566 ofd_seq_last_oid(oseq));
1569 diff = oid - ofd_seq_last_oid(oseq);
1570 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1571 ofd_seq_last_oid(oseq), diff);
1572 if (-diff > OST_MAX_PRECREATE) {
1573 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1574 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1575 "(%llu), trust the OST\n",
1576 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1577 PFID(&oseq->os_oi.oi_fid),
1578 ofd_seq_last_oid(oseq));
1580 /* Let MDS know that we are so far ahead. */
1581 rc = ostid_set_id(&rep_oa->o_oi,
1582 ofd_seq_last_oid(oseq) + 1);
1583 } else if (diff < 0) {
1584 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1586 oseq->os_destroys_in_progress = 0;
1588 /* XXX: Used by MDS for the first time! */
1589 oseq->os_destroys_in_progress = 0;
1592 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1593 ofd->ofd_lastid_gen)) {
1594 /* Keep the export ref so we can send the reply. */
1595 ofd_obd_disconnect(class_export_get(exp));
1596 GOTO(out_nolock, rc = -ENOTCONN);
1599 mutex_lock(&oseq->os_create_lock);
1600 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1601 exp->exp_conn_cnt) {
1602 CERROR("%s: dropping old precreate request\n",
1606 /* only precreate if seq is 0, IDIF or normal and also o_id
1607 * must be specfied */
1608 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1609 !fid_seq_is_idif(seq)) || oid == 0) {
1610 diff = 1; /* shouldn't we create this right now? */
1612 diff = oid - ofd_seq_last_oid(oseq);
1613 /* Do sync create if the seq is about to used up */
1614 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1615 if (unlikely(oid >= IDIF_MAX_OID - 1))
1617 } else if (fid_seq_is_norm(seq)) {
1619 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1622 CERROR("%s : invalid o_seq "DOSTID"\n",
1623 ofd_name(ofd), POSTID(&oa->o_oi));
1624 GOTO(out, rc = -EINVAL);
1629 CERROR("%s: invalid precreate request for "
1630 DOSTID", last_id %llu. "
1631 "Likely MDS last_id corruption\n",
1632 ofd_name(ofd), POSTID(&oa->o_oi),
1633 ofd_seq_last_oid(oseq));
1634 GOTO(out, rc = -EINVAL);
1639 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1645 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1646 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1647 /* don't enforce grant during orphan recovery */
1648 granted = tgt_grant_create(tsi->tsi_env,
1649 ofd_obd(ofd)->obd_self_export,
1654 CDEBUG(D_HA, "%s: failed to acquire grant "
1655 "space for precreate (%lld): rc = %d\n",
1656 ofd_name(ofd), diff, rc);
1661 /* This can happen if a new OST is formatted and installed
1662 * in place of an old one at the same index. Instead of
1663 * precreating potentially millions of deleted old objects
1664 * (possibly filling the OST), only precreate the last batch.
1665 * LFSCK will eventually clean up any orphans. LU-14 */
1666 if (diff > 5 * OST_MAX_PRECREATE) {
1667 diff = OST_MAX_PRECREATE / 2;
1668 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1669 "OST replaced or reformatted: "
1670 "LFSCK will clean up",
1673 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1674 "%u larger than the LAST_ID "DOSTID", only "
1675 "precreating the last %lld objects.\n",
1676 ofd_name(ofd), POSTID(&oa->o_oi),
1677 5 * OST_MAX_PRECREATE,
1678 POSTID(&oseq->os_oi), diff);
1679 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1683 next_id = ofd_seq_last_oid(oseq) + 1;
1684 count = ofd_precreate_batch(ofd, (int)diff);
1686 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1687 " at %llu\n", ofd_name(ofd),
1688 count, seq, next_id);
1690 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1691 && ktime_get_seconds() > enough_time) {
1692 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1693 " created at a rate of %d/s\n",
1694 ofd_name(ofd), created, diff + created,
1695 created / DISK_TIMEOUT);
1699 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1700 oseq, count, sync_trans);
1704 } else if (rc < 0) {
1710 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1711 LCONSOLE_WARN("%s: can't create the same count of"
1712 " objects when replaying the request"
1713 " (diff is %lld). see LU-4621\n",
1714 ofd_name(ofd), diff);
1717 /* some objects got created, we can return
1718 * them, even if last creation failed */
1721 CERROR("%s: unable to precreate: rc = %d\n",
1724 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1725 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1726 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1731 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1735 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1736 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1737 if (unlikely(!oseq->os_last_id_synced))
1738 oseq->os_last_id_synced = 1;
1740 mutex_unlock(&oseq->os_create_lock);
1743 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1744 ofd_seq_put(tsi->tsi_env, oseq);
1747 up_read(&ofd->ofd_lastid_rwsem);
1752 * OFD request handler for OST_DESTROY RPC.
1754 * This is OFD-specific part of request handling. It destroys data objects
1755 * related to destroyed object on MDT.
1757 * \param[in] tsi target session environment for this request
1759 * \retval 0 if successful
1760 * \retval negative value on error
1762 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1764 const struct ost_body *body = tsi->tsi_ost_body;
1765 struct ost_body *repbody;
1766 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1767 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1768 struct lu_fid *fid = &fti->fti_fid;
1769 ktime_t kstart = ktime_get();
1776 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1779 /* This is old case for clients before Lustre 2.4 */
1780 /* If there's a DLM request, cancel the locks mentioned in it */
1781 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1783 struct ldlm_request *dlm;
1785 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1788 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1791 *fid = body->oa.o_oi.oi_fid;
1792 oid = ostid_id(&body->oa.o_oi);
1795 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1797 /* check that o_misc makes sense */
1798 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1799 count = body->oa.o_misc;
1801 count = 1; /* default case - single destroy */
1803 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1804 POSTID(&body->oa.o_oi), count);
1809 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1810 if (lrc == -ENOENT) {
1812 "%s: destroying non-existent object "DFID"\n",
1813 ofd_name(ofd), PFID(fid));
1814 /* rewrite rc with -ENOENT only if it is 0 */
1817 } else if (lrc != 0) {
1818 CERROR("%s: error destroying object "DFID": %d\n",
1819 ofd_name(ofd), PFID(fid), lrc);
1825 lrc = fid_set_id(fid, oid);
1826 if (unlikely(lrc != 0 && count > 0))
1827 GOTO(out, rc = lrc);
1830 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1831 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1836 fid_to_ostid(fid, &repbody->oa.o_oi);
1841 * OFD request handler for OST_STATFS RPC.
1843 * This function gets statfs data from storage as part of request
1846 * \param[in] tsi target session environment for this request
1848 * \retval 0 if successful
1849 * \retval negative value on error
1851 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1853 ktime_t kstart = ktime_get();
1854 struct obd_statfs *osfs;
1859 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1861 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1863 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1864 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1866 CERROR("%s: statfs failed: rc = %d\n",
1867 tgt_name(tsi->tsi_tgt), rc);
1869 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1872 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1873 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1879 * OFD request handler for OST_SYNC RPC.
1881 * Sync object data or all filesystem data to the disk and pack the
1884 * \param[in] tsi target session environment for this request
1886 * \retval 0 if successful
1887 * \retval negative value on error
1889 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1891 struct ost_body *body = tsi->tsi_ost_body;
1892 struct ost_body *repbody;
1893 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1894 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1895 struct ofd_object *fo = NULL;
1896 ktime_t kstart = ktime_get();
1901 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1903 /* if no objid is specified, it means "sync whole filesystem" */
1904 if (!fid_is_zero(&tsi->tsi_fid)) {
1905 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1907 RETURN(PTR_ERR(fo));
1910 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1911 fo != NULL ? ofd_object_child(fo) : NULL,
1912 repbody->oa.o_size, repbody->oa.o_blocks);
1916 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1917 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1921 repbody->oa.o_oi = body->oa.o_oi;
1922 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1924 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1926 obdo_from_la(&repbody->oa, &fti->fti_attr,
1929 /* don't return rc from getattr */
1934 ofd_object_put(tsi->tsi_env, fo);
1939 * OFD request handler for OST_FALLOCATE RPC.
1941 * This is part of request processing. Validate request fields,
1942 * preallocate the given OFD object and pack reply.
1944 * \param[in] tsi target session environment for this request
1946 * \retval 0 if successful
1947 * \retval negative value on error
1949 static int ofd_fallocate_hdl(struct tgt_session_info *tsi)
1951 struct obdo *oa = &tsi->tsi_ost_body->oa;
1952 struct ost_body *repbody;
1953 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1954 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1955 struct ldlm_resource *res;
1956 struct ofd_object *fo;
1958 struct lustre_handle lh = { 0, };
1962 ktime_t kstart = ktime_get();
1964 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1965 if (repbody == NULL)
1966 RETURN(err_serious(-ENOMEM));
1969 * fallocate start and end are passed in o_size, o_blocks
1972 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1973 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1974 RETURN(err_serious(-EPROTO));
1978 mode = oa->o_falloc_mode;
1980 * mode == 0 (which is standard prealloc) and PUNCH is supported
1981 * Rest of mode options are not supported yet.
1983 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1984 RETURN(-EOPNOTSUPP);
1986 /* PUNCH_HOLE mode should always be accompanied with KEEP_SIZE flag
1987 * Check that and add the missing flag for such invalid call with
1990 if (mode & FALLOC_FL_PUNCH_HOLE && !(mode & FALLOC_FL_KEEP_SIZE)) {
1991 CWARN("%s: PUNCH mode misses KEEP_SIZE flag, setting it\n",
1992 tsi->tsi_tgt->lut_obd->obd_name);
1993 mode |= FALLOC_FL_KEEP_SIZE;
1996 repbody->oa.o_oi = oa->o_oi;
1997 repbody->oa.o_valid = OBD_MD_FLID;
1999 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2000 oa->o_flags & OBD_FL_SRVLOCK;
2003 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid,
2004 start, end, &lh, LCK_PW, &flags);
2009 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2012 GOTO(out, rc = PTR_ERR(fo));
2014 la_from_obdo(&info->fti_attr, oa,
2015 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2017 rc = ofd_object_fallocate(tsi->tsi_env, fo, start, end, mode,
2018 &info->fti_attr, oa);
2022 rc = ofd_attr_get(tsi->tsi_env, fo, &info->fti_attr);
2024 obdo_from_la(&repbody->oa, &info->fti_attr,
2029 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PREALLOC,
2030 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2034 ofd_object_put(tsi->tsi_env, fo);
2037 tgt_data_unlock(&lh, LCK_PW);
2039 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2042 struct ost_lvb *res_lvb;
2044 ldlm_res_lvbo_update(res, NULL, 0);
2045 res_lvb = res->lr_lvb_data;
2047 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2048 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2050 repbody->oa.o_valid |= OBD_MD_FLSIZE;
2051 repbody->oa.o_size = res_lvb->lvb_size;
2053 ldlm_resource_putref(res);
2061 * OFD request handler for OST_PUNCH RPC.
2063 * This is part of request processing. Validate request fields,
2064 * punch (truncate) the given OFD object and pack reply.
2066 * \param[in] tsi target session environment for this request
2068 * \retval 0 if successful
2069 * \retval negative value on error
2071 static int ofd_punch_hdl(struct tgt_session_info *tsi)
2073 const struct obdo *oa = &tsi->tsi_ost_body->oa;
2074 struct ost_body *repbody;
2075 struct ofd_thread_info *info = tsi2ofd_info(tsi);
2076 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
2077 struct ldlm_resource *res;
2078 struct ofd_object *fo;
2080 struct lustre_handle lh = { 0, };
2083 ktime_t kstart = ktime_get();
2088 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
2090 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
2091 BUILD_BUG_ON(!(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK));
2093 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2094 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2095 RETURN(err_serious(-EPROTO));
2097 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2098 if (repbody == NULL)
2099 RETURN(err_serious(-ENOMEM));
2101 /* punch start,end are passed in o_size,o_blocks throught wire */
2105 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2108 /* standard truncate optimization: if file body is completely
2109 * destroyed, don't send data back to the server. */
2111 flags |= LDLM_FL_AST_DISCARD_DATA;
2113 repbody->oa.o_oi = oa->o_oi;
2114 repbody->oa.o_valid = OBD_MD_FLID;
2116 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2117 oa->o_flags & OBD_FL_SRVLOCK;
2120 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
2121 end, &lh, LCK_PW, &flags);
2126 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2127 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2128 oa->o_valid, start, end);
2130 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2133 GOTO(out, rc = PTR_ERR(fo));
2135 la_from_obdo(&info->fti_attr, oa,
2136 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2137 info->fti_attr.la_size = start;
2138 info->fti_attr.la_valid |= LA_SIZE;
2140 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2145 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2146 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2149 ofd_object_put(tsi->tsi_env, fo);
2152 tgt_data_unlock(&lh, LCK_PW);
2154 /* we do not call this before to avoid lu_object_find() in
2155 * ->lvbo_update() holding another reference on the object.
2156 * otherwise concurrent destroy can make the object unavailable
2157 * for 2nd lu_object_find() waiting for the first reference
2158 * to go... deadlock! */
2159 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2162 struct ost_lvb *res_lvb;
2164 ldlm_res_lvbo_update(res, NULL, 0);
2165 res_lvb = res->lr_lvb_data;
2166 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2167 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2168 ldlm_resource_putref(res);
2174 static int ofd_ladvise_prefetch(const struct lu_env *env,
2175 struct ofd_object *fo,
2176 struct niobuf_local *lnb,
2177 __u64 start, __u64 end, enum dt_bufs_type dbt)
2179 struct ofd_thread_info *info = ofd_info(env);
2180 pgoff_t start_index, end_index, pages;
2181 struct niobuf_remote rnb;
2182 unsigned long nr_local;
2188 ofd_read_lock(env, fo);
2189 if (!ofd_object_exists(fo))
2190 GOTO(out_unlock, rc = -ENOENT);
2192 rc = ofd_attr_get(env, fo, &info->fti_attr);
2194 GOTO(out_unlock, rc);
2196 if (end > info->fti_attr.la_size)
2197 end = info->fti_attr.la_size;
2200 GOTO(out_unlock, rc);
2202 /* We need page aligned offset and length */
2203 start_index = start >> PAGE_SHIFT;
2204 end_index = (end - 1) >> PAGE_SHIFT;
2205 pages = end_index - start_index + 1;
2207 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2208 PTLRPC_MAX_BRW_PAGES;
2209 rnb.rnb_offset = start_index << PAGE_SHIFT;
2210 rnb.rnb_len = nr_local << PAGE_SHIFT;
2211 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb,
2212 PTLRPC_MAX_BRW_PAGES, dbt);
2213 if (unlikely(rc < 0))
2216 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2217 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2220 start_index += nr_local;
2225 ofd_read_unlock(env, fo);
2230 * OFD request handler for OST_LADVISE RPC.
2232 * Tune cache or perfetch policies according to advices.
2234 * \param[in] tsi target session environment for this request
2236 * \retval 0 if successful
2237 * \retval negative errno on error
2239 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2241 struct ptlrpc_request *req = tgt_ses_req(tsi);
2242 struct obd_export *exp = tsi->tsi_exp;
2243 struct ofd_device *ofd = ofd_exp(exp);
2244 struct ost_body *body, *repbody;
2245 struct ofd_thread_info *info;
2246 struct ofd_object *fo;
2247 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2248 const struct lu_env *env = svc_thread->t_env;
2249 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2250 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2251 struct lu_ladvise *ladvise;
2253 struct ladvise_hdr *ladvise_hdr;
2254 struct obd_ioobj ioo;
2255 struct lustre_handle lockh = { 0 };
2258 struct dt_object *dob;
2264 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2265 body = tsi->tsi_ost_body;
2267 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2268 RETURN(err_serious(-EPROTO));
2270 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2271 &RMF_OST_LADVISE_HDR);
2272 if (ladvise_hdr == NULL)
2273 RETURN(err_serious(-EPROTO));
2275 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2276 ladvise_hdr->lah_count < 1)
2277 RETURN(err_serious(-EPROTO));
2279 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2280 RETURN(err_serious(-EPROTO));
2282 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2283 if (ladvise == NULL)
2284 RETURN(err_serious(-EPROTO));
2286 num_advise = req_capsule_get_size(&req->rq_pill,
2287 &RMF_OST_LADVISE, RCL_CLIENT) /
2289 if (num_advise < ladvise_hdr->lah_count)
2290 RETURN(err_serious(-EPROTO));
2292 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2293 repbody->oa = body->oa;
2295 info = ofd_info_init(env, exp);
2297 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2298 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2302 fo = ofd_object_find(env, ofd, &info->fti_fid);
2307 LASSERT(fo != NULL);
2308 dob = ofd_object_child(fo);
2310 if (ptlrpc_connection_is_local(exp->exp_connection))
2311 dbt |= DT_BUFS_TYPE_LOCAL;
2313 for (i = 0; i < num_advise; i++, ladvise++) {
2314 start = ladvise->lla_start;
2315 end = ladvise->lla_end;
2317 rc = err_serious(-EPROTO);
2321 /* Handle different advice types */
2322 switch (ladvise->lla_advice) {
2326 case LU_LADVISE_WILLREAD:
2330 ioo.ioo_oid = body->oa.o_oi;
2332 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2333 &tsi->tsi_resid, start, end - 1,
2334 &lockh, LCK_PR, &flags);
2338 req->rq_status = ofd_ladvise_prefetch(env, fo,
2341 tgt_data_unlock(&lockh, LCK_PR);
2343 case LU_LADVISE_DONTNEED:
2344 rc = dt_ladvise(env, dob, ladvise->lla_start,
2345 ladvise->lla_end, LU_LADVISE_DONTNEED);
2352 ofd_object_put(env, fo);
2353 req->rq_status = rc;
2358 * OFD request handler for OST_QUOTACTL RPC.
2360 * This is part of request processing to validate incoming request fields,
2361 * get the requested data from OSD and pack reply.
2363 * \param[in] tsi target session environment for this request
2365 * \retval 0 if successful
2366 * \retval negative value on error
2368 static int ofd_quotactl(struct tgt_session_info *tsi)
2370 struct obd_quotactl *oqctl, *repoqc;
2371 struct lu_nodemap *nodemap;
2372 ktime_t kstart = ktime_get();
2378 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2380 RETURN(err_serious(-EPROTO));
2382 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2384 RETURN(err_serious(-ENOMEM));
2388 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2389 if (IS_ERR(nodemap))
2390 RETURN(PTR_ERR(nodemap));
2393 if (oqctl->qc_type == USRQUOTA)
2394 id = nodemap_map_id(nodemap, NODEMAP_UID,
2395 NODEMAP_CLIENT_TO_FS,
2397 else if (oqctl->qc_type == GRPQUOTA)
2398 id = nodemap_map_id(nodemap, NODEMAP_GID,
2399 NODEMAP_CLIENT_TO_FS,
2402 nodemap_putref(nodemap);
2404 if (repoqc->qc_id != id)
2405 swap(repoqc->qc_id, id);
2407 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2409 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2410 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2412 if (repoqc->qc_id != id)
2413 swap(repoqc->qc_id, id);
2419 * Prolong lock timeout for the given extent.
2421 * This function finds all locks related with incoming request and
2422 * prolongs their timeout.
2424 * If a client is holding a lock for a long time while it sends
2425 * read or write RPCs to the OST for the object under this lock,
2426 * then we don't want the OST to evict the client. Otherwise,
2427 * if the network or disk is very busy then the client may not
2428 * be able to make any progress to clear out dirty pages under
2429 * the lock and the application will fail.
2431 * Every time a Bulk Read/Write (BRW) request arrives for the object
2432 * covered by the lock, extend the timeout on that lock. The RPC should
2433 * contain a lock handle for the lock it is using, but this
2434 * isn't handled correctly by all client versions, and the
2435 * request may cover multiple locks.
2437 * \param[in] tsi target session environment for this request
2438 * \param[in] data struct of data to prolong locks
2441 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2442 struct ldlm_prolong_args *data)
2444 struct obdo *oa = &tsi->tsi_ost_body->oa;
2445 struct ldlm_lock *lock;
2449 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2450 data->lpa_export = tsi->tsi_exp;
2451 data->lpa_resid = tsi->tsi_resid;
2453 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2454 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2455 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2456 data->lpa_extent.end);
2458 if (oa->o_valid & OBD_MD_FLHANDLE) {
2459 /* mostly a request should be covered by only one lock, try
2461 lock = ldlm_handle2lock(&oa->o_handle);
2463 /* Fast path to check if the lock covers the whole IO
2464 * region exclusively. */
2465 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2466 &data->lpa_extent)) {
2468 LASSERT(lock->l_export == data->lpa_export);
2469 ldlm_lock_prolong_one(lock, data);
2470 LDLM_LOCK_PUT(lock);
2471 if (data->lpa_locks_cnt > 0)
2473 /* The lock was destroyed probably lets try
2476 lock->l_last_used = ktime_get();
2477 LDLM_LOCK_PUT(lock);
2482 ldlm_resource_prolong(data);
2487 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2489 * Determine if \a lock and the lock from request \a req are equivalent
2490 * by comparing their resource names, modes, and extents.
2492 * It is used to give priority to read and write RPCs being done
2493 * under this lock so that the client can drop the contended
2494 * lock more quickly and let other clients use it. This improves
2495 * overall performance in the case where the first client gets a
2496 * very large lock extent that prevents other clients from
2497 * submitting their writes.
2499 * \param[in] req ptlrpc_request being processed
2500 * \param[in] lock contended lock to match
2502 * \retval 1 if lock is matched
2503 * \retval 0 otherwise
2505 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2506 struct ldlm_lock *lock)
2508 struct niobuf_remote *rnb;
2509 struct obd_ioobj *ioo;
2510 enum ldlm_mode mode;
2511 struct ldlm_extent ext;
2512 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2516 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2517 LASSERT(ioo != NULL);
2519 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2520 LASSERT(rnb != NULL);
2522 ext.start = rnb->rnb_offset;
2523 rnb += ioo->ioo_bufcnt - 1;
2524 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2526 LASSERT(lock->l_resource != NULL);
2527 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2530 /* a bulk write can only hold a reference on a PW extent lock
2533 mode = LCK_PW | LCK_GROUP;
2534 if (opc == OST_READ)
2535 /* whereas a bulk read can be protected by either a PR or PW
2539 if (!(lock->l_granted_mode & mode))
2542 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2546 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2548 * Check for whether the given PTLRPC request (\a req) is blocking
2549 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2552 * \param[in] req the incoming request
2554 * \retval 1 if \a req is blocking an LDLM lock cancel
2555 * \retval 0 if it is not
2556 * \retval -ESTALE if lock is not found
2558 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2560 struct tgt_session_info *tsi;
2561 struct obd_ioobj *ioo;
2562 struct niobuf_remote *rnb;
2564 struct ldlm_prolong_args pa = { 0 };
2568 /* Don't use tgt_ses_info() to get session info, because lock_match()
2569 * can be called while request has no processing thread yet. */
2570 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2573 * Use LASSERT below because malformed RPCs should have
2574 * been filtered out in tgt_hpreq_handler().
2576 opc = lustre_msg_get_opc(req->rq_reqmsg);
2577 LASSERT(opc == OST_READ || opc == OST_WRITE);
2579 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2580 LASSERT(ioo != NULL);
2582 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2583 LASSERT(rnb != NULL);
2584 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2586 pa.lpa_mode = LCK_PW | LCK_GROUP;
2587 if (opc == OST_READ)
2588 pa.lpa_mode |= LCK_PR;
2590 pa.lpa_extent.start = rnb->rnb_offset;
2591 rnb += ioo->ioo_bufcnt - 1;
2592 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2594 DEBUG_REQ(D_RPCTRACE, req,
2595 "%s %s: refresh rw locks for "DFID" (%llu->%llu)",
2596 tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid),
2597 pa.lpa_extent.start, pa.lpa_extent.end);
2599 ofd_prolong_extent_locks(tsi, &pa);
2601 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n",
2602 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2604 if (pa.lpa_blocks_cnt > 0)
2607 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2611 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2613 * Called after the request has been handled. It refreshes lock timeout again
2614 * so that client has more time to send lock cancel RPC.
2616 * \param[in] req request which is being processed.
2618 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2620 ofd_rw_hpreq_check(req);
2624 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2626 * This function checks if the given lock is the same by its resname, mode
2627 * and extent as one taken from the request.
2628 * It is used to give priority to punch/truncate RPCs that might lead to
2629 * the fastest release of that lock when a lock is contended.
2631 * \param[in] req ptlrpc_request being processed
2632 * \param[in] lock contended lock to match
2634 * \retval 1 if lock is matched
2635 * \retval 0 otherwise
2637 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2638 struct ldlm_lock *lock)
2640 struct tgt_session_info *tsi;
2642 struct ldlm_extent ext;
2646 /* Don't use tgt_ses_info() to get session info, because lock_match()
2647 * can be called while request has no processing thread yet. */
2648 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2651 * Use LASSERT below because malformed RPCs should have
2652 * been filtered out in tgt_hpreq_handler().
2654 LASSERT(tsi->tsi_ost_body != NULL);
2655 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2656 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2659 oa = &tsi->tsi_ost_body->oa;
2660 ext.start = oa->o_size;
2661 ext.end = oa->o_blocks;
2663 LASSERT(lock->l_resource != NULL);
2664 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2667 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2670 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2674 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2676 * High-priority queue request check for whether the given punch request
2677 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2678 * covered by an LDLM lock.
2682 * \param[in] req the incoming request
2684 * \retval 1 if \a req is blocking an LDLM lock cancel
2685 * \retval 0 if it is not
2686 * \retval -ESTALE if lock is not found
2688 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2690 struct tgt_session_info *tsi;
2692 struct ldlm_prolong_args pa = { 0 };
2696 /* Don't use tgt_ses_info() to get session info, because lock_match()
2697 * can be called while request has no processing thread yet. */
2698 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2699 LASSERT(tsi != NULL);
2700 oa = &tsi->tsi_ost_body->oa;
2702 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2703 oa->o_flags & OBD_FL_SRVLOCK));
2705 pa.lpa_mode = LCK_PW | LCK_GROUP;
2706 pa.lpa_extent.start = oa->o_size;
2707 pa.lpa_extent.end = oa->o_blocks;
2710 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2711 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2712 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2714 ofd_prolong_extent_locks(tsi, &pa);
2716 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2717 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2719 if (pa.lpa_blocks_cnt > 0)
2722 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2726 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2728 * Called after the request has been handled. It refreshes lock timeout again
2729 * so that client has more time to send lock cancel RPC.
2731 * \param[in] req request which is being processed.
2733 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2735 ofd_punch_hpreq_check(req);
2738 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2739 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2740 .hpreq_check = ofd_rw_hpreq_check,
2741 .hpreq_fini = ofd_rw_hpreq_fini
2744 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2745 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2746 .hpreq_check = ofd_punch_hpreq_check,
2747 .hpreq_fini = ofd_punch_hpreq_fini
2751 * Assign high priority operations to an IO request.
2753 * Check if the incoming request is a candidate for
2754 * high-priority processing. If it is, assign it a high
2755 * priority operations table.
2757 * \param[in] tsi target session environment for this request
2759 static void ofd_hp_brw(struct tgt_session_info *tsi)
2761 struct niobuf_remote *rnb;
2762 struct obd_ioobj *ioo;
2766 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2767 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2768 if (ioo->ioo_bufcnt > 0) {
2769 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2770 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2772 /* no high priority if server lock is needed */
2773 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2774 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2778 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2782 * Assign high priority operations to an punch request.
2784 * Check if the incoming request is a candidate for
2785 * high-priority processing. If it is, assign it a high
2786 * priority operations table.
2788 * \param[in] tsi target session environment for this request
2790 static void ofd_hp_punch(struct tgt_session_info *tsi)
2792 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2793 /* no high-priority if server lock is needed */
2794 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2795 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2796 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2797 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2799 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2802 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2803 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2804 #define OST_BRW_READ OST_READ
2805 #define OST_BRW_WRITE OST_WRITE
2808 * Table of OFD-specific request handlers
2810 * This table contains all opcodes accepted by OFD and
2811 * specifies handlers for them. The tgt_request_handler()
2812 * uses such table from each target to process incoming
2815 static struct tgt_handler ofd_tgt_handlers[] = {
2816 TGT_RPC_HANDLER(OST_FIRST_OPC,
2817 0, OST_CONNECT, tgt_connect,
2818 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2819 TGT_RPC_HANDLER(OST_FIRST_OPC,
2820 0, OST_DISCONNECT, tgt_disconnect,
2821 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2822 TGT_RPC_HANDLER(OST_FIRST_OPC,
2823 0, OST_SET_INFO, ofd_set_info_hdl,
2824 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2825 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2826 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl),
2827 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2828 OST_SETATTR, ofd_setattr_hdl),
2829 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2830 OST_CREATE, ofd_create_hdl),
2831 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2832 OST_DESTROY, ofd_destroy_hdl),
2833 TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl),
2834 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
2836 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2837 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write,
2839 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2840 OST_PUNCH, ofd_punch_hdl,
2842 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl),
2843 TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl),
2844 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl),
2845 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE, ofd_fallocate_hdl),
2846 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
2849 static struct tgt_opc_slice ofd_common_slice[] = {
2851 .tos_opc_start = OST_FIRST_OPC,
2852 .tos_opc_end = OST_LAST_OPC,
2853 .tos_hs = ofd_tgt_handlers
2856 .tos_opc_start = OBD_FIRST_OPC,
2857 .tos_opc_end = OBD_LAST_OPC,
2858 .tos_hs = tgt_obd_handlers
2861 .tos_opc_start = LDLM_FIRST_OPC,
2862 .tos_opc_end = LDLM_LAST_OPC,
2863 .tos_hs = tgt_dlm_handlers
2866 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2867 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2868 .tos_hs = tgt_out_handlers
2871 .tos_opc_start = SEQ_FIRST_OPC,
2872 .tos_opc_end = SEQ_LAST_OPC,
2873 .tos_hs = seq_handlers
2876 .tos_opc_start = LFSCK_FIRST_OPC,
2877 .tos_opc_end = LFSCK_LAST_OPC,
2878 .tos_hs = tgt_lfsck_handlers
2881 .tos_opc_start = SEC_FIRST_OPC,
2882 .tos_opc_end = SEC_LAST_OPC,
2883 .tos_hs = tgt_sec_ctx_handlers
2890 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2891 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2894 * Implementation of lu_context_key::lct_key_exit.
2896 * Optional method called on lu_context_exit() for all allocated
2898 * It is used in OFD to sanitize context values which may be re-used
2899 * during another request processing by the same thread.
2901 * \param[in] ctx execution context
2902 * \param[in] key context key
2903 * \param[in] data ofd_thread_info
2905 static void ofd_key_exit(const struct lu_context *ctx,
2906 struct lu_context_key *key, void *data)
2908 struct ofd_thread_info *info = data;
2910 info->fti_env = NULL;
2911 info->fti_exp = NULL;
2914 info->fti_pre_version = 0;
2916 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2919 struct lu_context_key ofd_thread_key = {
2920 .lct_tags = LCT_DT_THREAD,
2921 .lct_init = ofd_key_init,
2922 .lct_fini = ofd_key_fini,
2923 .lct_exit = ofd_key_exit
2927 * Initialize OFD device according to parameters in the config log \a cfg.
2929 * This is the main starting point of OFD initialization. It fills all OFD
2930 * parameters with their initial values and calls other initializing functions
2931 * to set up all OFD subsystems.
2933 * \param[in] env execution environment
2934 * \param[in] m OFD device
2935 * \param[in] ldt LU device type of OFD
2936 * \param[in] cfg configuration log
2938 * \retval 0 if successful
2939 * \retval negative value on error
2941 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2942 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2944 const char *dev = lustre_cfg_string(cfg, 0);
2945 struct ofd_thread_info *info = NULL;
2946 struct obd_device *obd;
2947 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2949 struct nm_config_file *nodemap_config;
2950 struct obd_device_target *obt;
2956 obd = class_name2obd(dev);
2958 CERROR("Cannot find obd with name %s\n", dev);
2962 rc = lu_env_refill((struct lu_env *)env);
2967 obt->obt_magic = OBT_MAGIC;
2969 spin_lock_init(&m->ofd_flags_lock);
2970 m->ofd_raid_degraded = 0;
2971 m->ofd_sync_journal = 0;
2973 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2975 m->ofd_seq_count = 0;
2976 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2977 spin_lock_init(&m->ofd_inconsistency_lock);
2979 m->ofd_access_log_mask = -1; /* Log all accesses if enabled. */
2981 spin_lock_init(&m->ofd_batch_lock);
2982 init_rwsem(&m->ofd_lastid_rwsem);
2984 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2985 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2986 /* set this lu_device to obd, because error handling need it */
2987 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2989 /* No connection accepted until configurations will finish */
2990 spin_lock(&obd->obd_dev_lock);
2991 obd->obd_no_conn = 1;
2992 spin_unlock(&obd->obd_dev_lock);
2993 obd->obd_replayable = 1;
2994 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2995 char *str = lustre_cfg_string(cfg, 4);
2997 if (strchr(str, 'n')) {
2998 CWARN("%s: recovery disabled\n", obd->obd_name);
2999 obd->obd_replayable = 0;
3003 info = ofd_info_init(env, NULL);
3007 rc = ofd_stack_init(env, m, cfg, &lmd_flags);
3009 CERROR("%s: can't init device stack, rc %d\n",
3014 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
3015 ofd_procfs_add_brw_stats_symlink(m);
3018 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
3019 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
3020 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
3021 LDLM_NAMESPACE_SERVER,
3022 LDLM_NAMESPACE_GREEDY,
3024 if (IS_ERR(m->ofd_namespace)) {
3025 rc = PTR_ERR(m->ofd_namespace);
3026 CERROR("%s: unable to create server namespace: rc = %d\n",
3028 m->ofd_namespace = NULL;
3029 GOTO(err_fini_stack, rc);
3031 /* set obd_namespace for compatibility with old code */
3032 obd->obd_namespace = m->ofd_namespace;
3033 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
3034 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
3035 m->ofd_namespace->ns_lvbp = m;
3037 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3038 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
3040 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
3041 OBD_FAIL_OST_ALL_REQUEST_NET,
3042 OBD_FAIL_OST_ALL_REPLY_NET);
3044 GOTO(err_free_ns, rc);
3046 if (lmd_flags & LMD_FLG_SKIP_LFSCK)
3047 m->ofd_skip_lfsck = 1;
3048 if (lmd_flags & LMD_FLG_LOCAL_RECOV)
3049 m->ofd_lut.lut_local_recovery = 1;
3051 rc = ofd_tunables_init(m);
3053 GOTO(err_fini_lut, rc);
3055 tgd->tgd_reserved_pcnt = 0;
3057 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
3058 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
3059 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
3060 OFD_PRECREATE_SMALL_FS)
3061 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
3062 m->ofd_atime_diff = OFD_DEF_ATIME_DIFF;
3064 rc = ofd_fs_setup(env, m, obd);
3066 GOTO(err_fini_proc, rc);
3068 fid.f_seq = FID_SEQ_LOCAL_NAME;
3071 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
3074 GOTO(err_fini_fs, rc);
3076 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
3078 if (IS_ERR(nodemap_config)) {
3079 rc = PTR_ERR(nodemap_config);
3081 GOTO(err_fini_los, rc);
3083 obt->obt_nodemap_config_file = nodemap_config;
3086 rc = ofd_start_inconsistency_verification_thread(m);
3088 GOTO(err_fini_nm, rc);
3090 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
3095 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
3096 obt->obt_nodemap_config_file = NULL;
3098 local_oid_storage_fini(env, m->ofd_los);
3101 ofd_fs_cleanup(env, m);
3105 tgt_fini(env, &m->ofd_lut);
3107 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3108 obd->obd_namespace = m->ofd_namespace = NULL;
3110 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3115 * Stop the OFD device
3117 * This function stops the OFD device and all its subsystems.
3118 * This is the end of OFD lifecycle.
3120 * \param[in] env execution environment
3121 * \param[in] m OFD device
3123 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3125 struct obd_device *obd = ofd_obd(m);
3126 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3127 struct lfsck_stop stop;
3129 stop.ls_status = LS_PAUSED;
3131 lfsck_stop(env, m->ofd_osd, &stop);
3132 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3133 target_recovery_fini(obd);
3134 if (m->ofd_namespace != NULL)
3135 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3136 d->ld_obd->obd_force);
3138 obd_exports_barrier(obd);
3139 obd_zombie_barrier();
3142 tgt_fini(env, &m->ofd_lut);
3143 ofd_stop_inconsistency_verification_thread(m);
3144 lfsck_degister(env, m->ofd_osd);
3145 ofd_fs_cleanup(env, m);
3146 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3147 obd->u.obt.obt_nodemap_config_file = NULL;
3149 if (m->ofd_namespace != NULL) {
3150 ldlm_namespace_free_post(m->ofd_namespace);
3151 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3154 ofd_access_log_delete(m->ofd_access_log);
3155 m->ofd_access_log = NULL;
3157 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3159 LASSERT(atomic_read(&d->ld_ref) == 0);
3160 server_put_mount(obd->obd_name, true);
3165 * Implementation of lu_device_type_operations::ldto_device_fini.
3167 * Finalize device. Dual to ofd_device_init(). It is called from
3168 * obd_precleanup() and stops the current device.
3170 * \param[in] env execution environment
3171 * \param[in] d LU device of OFD
3175 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3176 struct lu_device *d)
3179 ofd_fini(env, ofd_dev(d));
3184 * Implementation of lu_device_type_operations::ldto_device_free.
3186 * Free OFD device. Dual to ofd_device_alloc().
3188 * \param[in] env execution environment
3189 * \param[in] d LU device of OFD
3193 static struct lu_device *ofd_device_free(const struct lu_env *env,
3194 struct lu_device *d)
3196 struct ofd_device *m = ofd_dev(d);
3198 dt_device_fini(&m->ofd_dt_dev);
3204 * Implementation of lu_device_type_operations::ldto_device_alloc.
3206 * This function allocates the new OFD device. It is called from
3207 * obd_setup() if OBD device had lu_device_type defined.
3209 * \param[in] env execution environment
3210 * \param[in] t lu_device_type of OFD device
3211 * \param[in] cfg configuration log
3213 * \retval pointer to the lu_device of just allocated OFD
3214 * \retval ERR_PTR of return value on error
3216 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3217 struct lu_device_type *t,
3218 struct lustre_cfg *cfg)
3220 struct ofd_device *m;
3221 struct lu_device *l;
3226 return ERR_PTR(-ENOMEM);
3228 l = &m->ofd_dt_dev.dd_lu_dev;
3229 dt_device_init(&m->ofd_dt_dev, t);
3230 rc = ofd_init0(env, m, t, cfg);
3232 ofd_device_free(env, l);
3239 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3240 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3242 static const struct lu_device_type_operations ofd_device_type_ops = {
3243 .ldto_init = ofd_type_init,
3244 .ldto_fini = ofd_type_fini,
3246 .ldto_start = ofd_type_start,
3247 .ldto_stop = ofd_type_stop,
3249 .ldto_device_alloc = ofd_device_alloc,
3250 .ldto_device_free = ofd_device_free,
3251 .ldto_device_fini = ofd_device_fini
3254 static struct lu_device_type ofd_device_type = {
3255 .ldt_tags = LU_DEVICE_DT,
3256 .ldt_name = LUSTRE_OST_NAME,
3257 .ldt_ops = &ofd_device_type_ops,
3258 .ldt_ctx_tags = LCT_DT_THREAD
3262 * Initialize OFD module.
3264 * This function is called upon module loading. It registers OFD device type
3265 * and prepares all in-memory structures used by all OFD devices.
3267 * \retval 0 if successful
3268 * \retval negative value on error
3270 static int __init ofd_init(void)
3274 rc = lu_kmem_init(ofd_caches);
3278 rc = ofd_access_log_module_init();
3282 rc = class_register_type(&ofd_obd_ops, NULL, true,
3283 LUSTRE_OST_NAME, &ofd_device_type);
3285 goto out_ofd_access_log;
3290 ofd_access_log_module_exit();
3292 lu_kmem_fini(ofd_caches);
3300 * This function is called upon OFD module unloading.
3301 * It frees all related structures and unregisters OFD device type.
3303 static void __exit ofd_exit(void)
3305 class_unregister_type(LUSTRE_OST_NAME);
3306 ofd_access_log_module_exit();
3307 lu_kmem_fini(ofd_caches);
3310 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3311 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3312 MODULE_VERSION(LUSTRE_VERSION_STRING);
3313 MODULE_LICENSE("GPL");
3315 module_init(ofd_init);
3316 module_exit(ofd_exit);