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/
31 * lustre/ofd/ofd_dev.c
33 * This file contains OSD API methods for OBD Filter Device (OFD),
34 * request handlers and supplemental functions to set OFD up and clean it up.
36 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
37 * Author: Mike Pershin <mike.pershin@intel.com>
38 * Author: Johann Lombardi <johann.lombardi@intel.com>
41 * The OBD Filter Device (OFD) module belongs to the Object Storage
42 * Server stack and connects the RPC oriented Unified Target (TGT)
43 * layer (see lustre/include/lu_target.h) to the storage oriented OSD
44 * layer (see Documentation/osd-api.txt).
52 * OFD implements the LU and OBD device APIs and is responsible for:
54 * - Handling client requests (create, destroy, bulk IO, setattr,
55 * get_info, set_info, statfs) for the objects belonging to the OST
56 * (together with TGT).
58 * - Providing grant space management which allows clients to reserve
59 * disk space for data writeback. OFD tracks grants on global and
62 * - Handling object precreation requests from MDTs.
64 * - Operating the LDLM service that allows clients to maintain object
65 * data cache coherence.
68 #define DEBUG_SUBSYSTEM S_FILTER
70 #include <obd_class.h>
71 #include <obd_cksum.h>
72 #include <uapi/linux/lustre/lustre_param.h>
73 #include <lustre_fid.h>
74 #include <lustre_lfsck.h>
75 #include <lustre_dlm.h>
76 #include <lustre_quota.h>
77 #include <lustre_nodemap.h>
78 #include <lustre_log.h>
79 #include <linux/falloc.h>
81 #include "ofd_internal.h"
83 /* Slab for OFD object allocation */
84 static struct kmem_cache *ofd_object_kmem;
85 static struct lu_kmem_descr ofd_caches[] = {
87 .ckd_cache = &ofd_object_kmem,
88 .ckd_name = "ofd_obj",
89 .ckd_size = sizeof(struct ofd_object)
97 * Connect OFD to the next device in the stack.
99 * This function is used for device stack configuration and links OFD
100 * device with bottom OSD device.
102 * \param[in] env execution environment
103 * \param[in] m OFD device
104 * \param[in] next name of next device in the stack
105 * \param[out] exp export to return
107 * \retval 0 and export in \a exp if successful
108 * \retval negative value on error
110 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
111 const char *next, struct obd_export **exp)
113 struct obd_connect_data *data = NULL;
114 struct obd_device *obd;
120 GOTO(out, rc = -ENOMEM);
122 obd = class_name2obd(next);
124 CERROR("%s: can't locate next device: %s\n",
126 GOTO(out, rc = -ENOTCONN);
129 data->ocd_connect_flags = OBD_CONNECT_VERSION;
130 data->ocd_version = LUSTRE_VERSION_CODE;
132 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
134 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
135 ofd_name(m), next, rc);
139 m->ofd_dt_dev.dd_lu_dev.ld_site =
140 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
141 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
142 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
143 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
152 * Initialize stack of devices.
154 * This function initializes OFD-OSD device stack to serve OST requests
156 * \param[in] env execution environment
157 * \param[in] m OFD device
158 * \param[in] cfg Lustre config for this server
160 * \retval 0 if successful
161 * \retval negative value on error
163 static int ofd_stack_init(const struct lu_env *env,
164 struct ofd_device *m, struct lustre_cfg *cfg,
167 const char *dev = lustre_cfg_string(cfg, 0);
169 struct ofd_thread_info *info = ofd_info(env);
170 struct lustre_mount_info *lmi;
171 struct lustre_mount_data *lmd;
177 lmi = server_get_mount(dev);
179 CERROR("Cannot get mount info for %s!\n", dev);
183 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
185 if (lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
186 m->ofd_skip_lfsck = 1;
187 if (lmd->lmd_flags & LMD_FLG_NO_PRECREATE)
188 m->ofd_no_precreate = 1;
189 *lmd_flags = lmd->lmd_flags;
192 /* find bottom osd */
193 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
197 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
198 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
199 OBD_FREE(osdname, MTI_NAME_MAXLEN);
203 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
205 m->ofd_osd = lu2dt_dev(d);
207 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
208 "%s-osd", lustre_cfg_string(cfg, 0));
214 * Finalize the device stack OFD-OSD.
216 * This function cleans OFD-OSD device stack and
217 * disconnects OFD from the OSD.
219 * \param[in] env execution environment
220 * \param[in] m OFD device
221 * \param[in] top top device of stack
223 * \retval 0 if successful
224 * \retval negative value on error
226 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
227 struct lu_device *top)
229 struct obd_device *obd = ofd_obd(m);
230 struct lustre_cfg_bufs bufs;
231 struct lustre_cfg *lcfg;
236 lu_site_purge(env, top->ld_site, ~0);
237 /* process cleanup, pass mdt obd name to get obd umount flags */
238 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
243 lustre_cfg_bufs_set_string(&bufs, 1, flags);
244 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
247 lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs);
250 top->ld_ops->ldo_process_config(env, top, lcfg);
251 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
253 if (m->ofd_los != NULL) {
254 local_oid_storage_fini(env, m->ofd_los);
258 lu_site_purge(env, top->ld_site, ~0);
259 lu_site_print(env, top->ld_site, &top->ld_site->ls_obj_hash.nelems,
260 D_OTHER, lu_cdebug_printer);
261 LASSERT(m->ofd_osd_exp);
262 obd_disconnect(m->ofd_osd_exp);
267 static void ofd_stack_pre_fini(const struct lu_env *env, struct ofd_device *m,
268 struct lu_device *top)
270 struct lustre_cfg_bufs bufs;
271 struct lustre_cfg *lcfg;
276 lustre_cfg_bufs_reset(&bufs, ofd_name(m));
277 lustre_cfg_bufs_set_string(&bufs, 1, NULL);
278 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
280 CERROR("%s: failed to trigger LCFG_PRE_CLEANUP\n", ofd_name(m));
282 lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, &bufs);
283 top->ld_ops->ldo_process_config(env, top, lcfg);
284 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
285 lcfg->lcfg_buflens));
291 /* For interoperability, see mdt_interop_param[]. */
292 static struct cfg_interop_param ofd_interop_param[] = {
293 { "ost.quota_type", NULL },
298 * Check if parameters are symlinks to the OSD.
300 * Some parameters were moved from ofd to osd and only their
301 * symlinks were kept in ofd by LU-3106. They are:
302 * -writehthrough_cache_enable
303 * -readcache_max_filesize
307 * Since they are not included by the static lprocfs var list, a pre-check
308 * is added for them to avoid "unknown param" errors. If they are matched
309 * in this check, they will be passed to the OSD directly.
311 * \param[in] param parameters to check
313 * \retval true if param is symlink to OSD param
316 static bool match_symlink_param(char *param)
321 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
322 sval = strchr(param, '=');
324 paramlen = sval - param;
325 if (strncmp(param, "brw_stats", paramlen) == 0)
334 * Process various configuration parameters.
336 * This function is used by MGS to process specific configurations and
337 * pass them through to the next device in server stack, i.e. the OSD.
339 * \param[in] env execution environment
340 * \param[in] d LU device of OFD
341 * \param[in] cfg parameters to process
343 * \retval 0 if successful
344 * \retval negative value on error
346 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
347 struct lustre_cfg *cfg)
349 struct ofd_device *m = ofd_dev(d);
350 struct dt_device *dt_next = m->ofd_osd;
351 struct lu_device *next = &dt_next->dd_lu_dev;
356 switch (cfg->lcfg_command) {
358 /* For interoperability */
359 struct cfg_interop_param *ptr = NULL;
360 struct lustre_cfg *old_cfg = NULL;
364 param = lustre_cfg_string(cfg, 1);
366 CERROR("param is empty\n");
371 ptr = class_find_old_param(param, ofd_interop_param);
373 if (ptr->new_param == NULL) {
375 CWARN("For interoperability, skip this %s."
376 " It is obsolete.\n", ptr->old_param);
380 CWARN("Found old param %s, changed it to %s.\n",
381 ptr->old_param, ptr->new_param);
384 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
391 if (match_symlink_param(param)) {
392 rc = next->ld_ops->ldo_process_config(env, next, cfg);
396 count = class_modify_config(cfg, PARAM_OST,
397 &d->ld_obd->obd_kset.kobj);
402 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
404 /* we don't understand; pass it on */
405 rc = next->ld_ops->ldo_process_config(env, next, cfg);
408 case LCFG_SPTLRPC_CONF: {
413 /* others are passed further */
414 rc = next->ld_ops->ldo_process_config(env, next, cfg);
421 * Implementation of lu_object_operations::loo_object_init for OFD
423 * Allocate just the next object (OSD) in stack.
425 * \param[in] env execution environment
426 * \param[in] o lu_object of OFD object
427 * \param[in] conf additional configuration parameters, not used here
429 * \retval 0 if successful
430 * \retval negative value on error
432 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
433 const struct lu_object_conf *conf)
435 struct ofd_device *d = ofd_dev(o->lo_dev);
436 struct lu_device *under;
437 struct lu_object *below;
442 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
443 PFID(lu_object_fid(o)));
445 under = &d->ofd_osd->dd_lu_dev;
446 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
448 lu_object_add(o, below);
455 static void ofd_object_free_rcu(struct rcu_head *head)
457 struct ofd_object *of = container_of(head, struct ofd_object,
460 kmem_cache_free(ofd_object_kmem, of);
464 * Implementation of lu_object_operations::loo_object_free.
466 * Finish OFD object lifecycle and free its memory.
468 * \param[in] env execution environment
469 * \param[in] o LU object of OFD object
471 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
473 struct ofd_object *of = ofd_obj(o);
474 struct lu_object_header *h;
479 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
480 PFID(lu_object_fid(o)));
483 lu_object_header_fini(h);
484 OBD_FREE_PRE(of, sizeof(*of), "slab-freed");
485 call_rcu(&of->ofo_header.loh_rcu, ofd_object_free_rcu);
490 * Implementation of lu_object_operations::loo_object_print.
492 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
493 * LU_OBJECT_DEBUG() for more details about the compound object printing.
495 * \param[in] env execution environment
496 * \param[in] cookie opaque data passed to the printer function
497 * \param[in] p printer function to use
498 * \param[in] o LU object of OFD object
500 * \retval 0 if successful
501 * \retval negative value on error
503 static int ofd_object_print(const struct lu_env *env, void *cookie,
504 lu_printer_t p, const struct lu_object *o)
506 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
509 static const struct lu_object_operations ofd_obj_ops = {
510 .loo_object_init = ofd_object_init,
511 .loo_object_free = ofd_object_free,
512 .loo_object_print = ofd_object_print
516 * Implementation of lu_device_operations::lod_object_alloc.
518 * This function allocates OFD part of compound OFD-OSD object and
519 * initializes its header, because OFD is the top device in stack
521 * \param[in] env execution environment
522 * \param[in] hdr object header, NULL for OFD
523 * \param[in] d lu_device
525 * \retval allocated object if successful
526 * \retval NULL value on failed allocation
528 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
529 const struct lu_object_header *hdr,
532 struct ofd_object *of;
536 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
539 struct lu_object_header *h;
541 o = &of->ofo_obj.do_lu;
543 lu_object_header_init(h);
544 lu_object_init(o, h, d);
545 lu_object_add_top(h, o);
546 o->lo_ops = &ofd_obj_ops;
547 range_lock_tree_init(&of->ofo_write_tree);
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 OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_OST_PREPARE_DELAY, OBD_FAIL_ONCE,
650 (OBD_TIMEOUT_DEFAULT + 1) / 4);
651 LASSERT(obd->obd_no_conn);
652 spin_lock(&obd->obd_dev_lock);
653 obd->obd_no_conn = 0;
654 spin_unlock(&obd->obd_dev_lock);
656 if (obd->obd_recovering == 0)
657 ofd_postrecov(env, ofd);
663 * Implementation of lu_device_operations::ldo_recovery_complete.
665 * This method notifies all layers about 'recovery complete' event. That means
666 * device is in full state and consistent. An OFD calculates available grant
667 * space upon this event.
669 * \param[in] env execution environment
670 * \param[in] dev lu_device of OFD device
672 * \retval 0 if successful
673 * \retval negative value on error
675 static int ofd_recovery_complete(const struct lu_env *env,
676 struct lu_device *dev)
678 struct ofd_thread_info *oti = ofd_info(env);
679 struct ofd_device *ofd = ofd_dev(dev);
680 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
686 * Grant space for object precreation on the self export.
687 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
688 * is enough to create 10k objects. More space is then acquired for
689 * precreation in tgt_grant_create().
691 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
692 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
693 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
694 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
695 OBD_CONNECT_GRANT_PARAM;
696 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
698 rc = next->ld_ops->ldo_recovery_complete(env, next);
703 * lu_device_operations matrix for OFD device.
705 static const struct lu_device_operations ofd_lu_ops = {
706 .ldo_object_alloc = ofd_object_alloc,
707 .ldo_process_config = ofd_process_config,
708 .ldo_recovery_complete = ofd_recovery_complete,
709 .ldo_prepare = ofd_prepare,
712 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
714 * Expose OSD statistics to OFD layer.
716 * The osd interfaces to the backend file system exposes useful data
717 * such as brw_stats and read or write cache states. This same data
718 * needs to be exposed into the obdfilter (ofd) layer to maintain
719 * backwards compatibility. This function creates the symlinks in the
720 * proc layer to enable this.
722 * \param[in] ofd OFD device
724 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
726 struct obd_device *obd = ofd_obd(ofd);
727 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
728 struct kobj_type *osd_type;
731 osd_type = get_ktype(&ofd->ofd_osd->dd_kobj);
732 for (i = 0; osd_type->default_attrs[i]; i++) {
733 if (strcmp(osd_type->default_attrs[i]->name,
734 "read_cache_enable") == 0) {
735 ofd->ofd_read_cache_enable =
736 osd_type->default_attrs[i];
739 if (strcmp(osd_type->default_attrs[i]->name,
740 "readcache_max_filesize") == 0) {
741 ofd->ofd_read_cache_max_filesize =
742 osd_type->default_attrs[i];
745 if (strcmp(osd_type->default_attrs[i]->name,
746 "writethrough_cache_enable") == 0) {
747 ofd->ofd_write_cache_enable =
748 osd_type->default_attrs[i];
752 if (obd->obd_proc_entry == NULL)
755 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
756 "../../%s/%s/brw_stats",
757 osd_obd->obd_type->typ_name, obd->obd_name);
762 * Cleanup all procfs entries in OFD.
764 * \param[in] ofd OFD device
766 static void ofd_procfs_fini(struct ofd_device *ofd)
768 struct obd_device *obd = ofd_obd(ofd);
770 tgt_tunables_fini(&ofd->ofd_lut);
771 lprocfs_free_per_client_stats(obd);
772 lprocfs_obd_cleanup(obd);
773 lprocfs_free_obd_stats(obd);
774 lprocfs_job_stats_fini(obd);
778 * Stop SEQ/FID server on OFD.
780 * \param[in] env execution environment
781 * \param[in] ofd OFD device
783 * \retval 0 if successful
784 * \retval negative value on error
786 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
788 return seq_site_fini(env, &ofd->ofd_seq_site);
792 * Start SEQ/FID server on OFD.
794 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
795 * It also connects to the master server to get own FID sequence (SEQ) range
796 * to this particular OFD. Typically that happens when the OST is first
797 * formatted or in the rare case that it exhausts the local sequence range.
799 * The sequence range is allocated out to the MDTs for OST object allocations,
800 * and not directly to the clients.
802 * \param[in] env execution environment
803 * \param[in] ofd OFD device
805 * \retval 0 if successful
806 * \retval negative value on error
808 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
810 struct seq_server_site *ss = &ofd->ofd_seq_site;
811 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
812 char *obd_name = ofd_name(ofd);
814 int len = strlen(obd_name) + 7;
817 ss = &ofd->ofd_seq_site;
818 lu->ld_site->ld_seq_site = ss;
819 ss->ss_lu = lu->ld_site;
820 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
822 OBD_ALLOC(name, len);
826 OBD_ALLOC_PTR(ss->ss_server_seq);
827 if (ss->ss_server_seq == NULL)
828 GOTO(out_name, rc = -ENOMEM);
830 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
831 LUSTRE_SEQ_SERVER, ss);
833 CERROR("%s: seq server init error: rc = %d\n", obd_name, rc);
834 GOTO(out_server, rc);
836 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
838 OBD_ALLOC_PTR(ss->ss_client_seq);
839 if (ss->ss_client_seq == NULL)
840 GOTO(out_server, rc = -ENOMEM);
842 snprintf(name, len, "%s-super", obd_name);
843 seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
846 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
849 seq_client_fini(ss->ss_client_seq);
850 OBD_FREE_PTR(ss->ss_client_seq);
851 ss->ss_client_seq = NULL;
853 seq_server_fini(ss->ss_server_seq, env);
854 OBD_FREE_PTR(ss->ss_server_seq);
855 ss->ss_server_seq = NULL;
864 * OFD request handler for OST_SET_INFO RPC.
866 * This is OFD-specific part of request handling
868 * \param[in] tsi target session environment for this request
870 * \retval 0 if successful
871 * \retval negative value on error
873 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
875 struct ptlrpc_request *req = tgt_ses_req(tsi);
876 struct ost_body *body = NULL, *repbody;
877 void *key, *val = NULL;
878 int keylen, vallen, rc = 0;
879 bool is_grant_shrink;
880 ktime_t kstart = ktime_get();
884 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
886 DEBUG_REQ(D_HA, req, "no set_info key");
887 RETURN(err_serious(-EFAULT));
889 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
892 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
894 DEBUG_REQ(D_HA, req, "no set_info val");
895 RETURN(err_serious(-EFAULT));
897 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
900 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
902 /* In this case the value is actually an RMF_OST_BODY, so we
903 * transmutate the type of this PTLRPC */
904 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
906 rc = req_capsule_server_pack(tsi->tsi_pill);
910 if (is_grant_shrink) {
911 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
913 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
916 /** handle grant shrink, similar to a read request */
917 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
919 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
921 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
924 CERROR("%s: Unsupported key %s\n",
925 tgt_name(tsi->tsi_tgt), (char *)key);
928 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
929 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
935 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
937 * This function returns a list of extents which describes how a file's
938 * blocks are laid out on the disk.
940 * \param[in] env execution environment
941 * \param[in] ofd OFD device
942 * \param[in] fid FID of object
943 * \param[in] fiemap fiemap structure to fill with data
945 * \retval 0 if \a fiemap is filled with data successfully
946 * \retval negative value on error
948 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
949 struct lu_fid *fid, struct fiemap *fiemap)
951 struct ofd_object *fo;
954 fo = ofd_object_find(env, ofd, fid);
956 CERROR("%s: error finding object "DFID"\n",
957 ofd_name(ofd), PFID(fid));
961 ofd_read_lock(env, fo);
962 if (ofd_object_exists(fo))
963 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
966 ofd_read_unlock(env, fo);
967 ofd_object_put(env, fo);
972 static int ofd_lock_unlock_region(const struct lu_env *env,
973 struct ldlm_namespace *ns,
974 struct ldlm_res_id *res_id,
975 unsigned long long begin,
976 unsigned long long end)
980 struct lustre_handle lh = { 0 };
982 LASSERT(begin <= end);
984 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
988 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
989 tgt_data_unlock(&lh, LCK_PR);
995 * Lock the sparse areas of given resource.
997 * The locking of sparse areas will cause dirty data to be flushed back from
998 * clients. This is used when getting the FIEMAP of an object to make sure
999 * there is no unaccounted cached data on clients.
1001 * This function goes through \a fiemap list of extents and locks only sparse
1002 * areas between extents.
1004 * \param[in] ns LDLM namespace
1005 * \param[in] res_id resource ID
1006 * \param[in] fiemap file extents mapping on disk
1007 * \param[in] locked list head of regions list
1009 * \retval 0 if successful
1010 * \retval negative value on error
1012 static int lock_zero_regions(const struct lu_env *env,
1013 struct ldlm_namespace *ns,
1014 struct ldlm_res_id *res_id,
1015 struct fiemap *fiemap)
1017 __u64 begin = fiemap->fm_start;
1020 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1024 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1025 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1026 if (fiemap_start[i].fe_logical > begin) {
1027 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1028 begin, fiemap_start[i].fe_logical);
1029 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1030 fiemap_start[i].fe_logical);
1035 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1038 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1039 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1040 begin, fiemap->fm_start + fiemap->fm_length);
1041 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1042 fiemap->fm_start + fiemap->fm_length);
1050 * OFD request handler for OST_GET_INFO RPC.
1052 * This is OFD-specific part of request handling. The OFD-specific keys are:
1053 * - KEY_LAST_ID (obsolete)
1057 * This function reads needed data from storage and fills reply with it.
1059 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1060 * and is kept for compatibility.
1062 * \param[in] tsi target session environment for this request
1064 * \retval 0 if successful
1065 * \retval negative value on error
1067 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1069 struct obd_export *exp = tsi->tsi_exp;
1070 struct ofd_device *ofd = ofd_exp(exp);
1071 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1074 int replylen, rc = 0;
1075 ktime_t kstart = ktime_get();
1079 /* this common part for get_info rpc */
1080 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1082 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1083 RETURN(err_serious(-EPROTO));
1085 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1088 if (KEY_IS(KEY_LAST_ID)) {
1090 struct ofd_seq *oseq;
1092 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1093 rc = req_capsule_server_pack(tsi->tsi_pill);
1095 RETURN(err_serious(rc));
1097 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1099 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1100 (u64)exp->exp_filter_data.fed_group);
1104 *last_id = ofd_seq_last_oid(oseq);
1105 ofd_seq_put(tsi->tsi_env, oseq);
1106 } else if (KEY_IS(KEY_FIEMAP)) {
1107 struct ll_fiemap_info_key *fm_key;
1108 struct fiemap *fiemap;
1111 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1113 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1114 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1116 RETURN(err_serious(rc));
1118 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1120 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1122 replylen = fiemap_count_to_size(
1123 fm_key->lfik_fiemap.fm_extent_count);
1124 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1125 RCL_SERVER, replylen);
1127 rc = req_capsule_server_pack(tsi->tsi_pill);
1129 RETURN(err_serious(rc));
1131 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1135 *fiemap = fm_key->lfik_fiemap;
1136 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1138 /* LU-3219: Lock the sparse areas to make sure dirty
1139 * flushed back from client, then call fiemap again. */
1140 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1141 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1142 ost_fid_build_resid(fid, &fti->fti_resid);
1143 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1144 &fti->fti_resid, fiemap);
1146 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1149 } else if (KEY_IS(KEY_LAST_FID)) {
1150 struct ofd_device *ofd = ofd_exp(exp);
1151 struct ofd_seq *oseq;
1155 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1156 rc = req_capsule_server_pack(tsi->tsi_pill);
1158 RETURN(err_serious(rc));
1160 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1162 RETURN(err_serious(-EPROTO));
1164 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1166 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1170 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1171 ostid_seq(&fti->fti_ostid));
1173 RETURN(PTR_ERR(oseq));
1175 rc = ostid_to_fid(fid, &oseq->os_oi,
1176 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1180 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1183 ofd_seq_put(tsi->tsi_env, oseq);
1185 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1189 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1190 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1196 * OFD request handler for OST_GETATTR RPC.
1198 * This is OFD-specific part of request handling. It finds the OFD object
1199 * by its FID, gets attributes from storage and packs result to the reply.
1201 * \param[in] tsi target session environment for this request
1203 * \retval 0 if successful
1204 * \retval negative value on error
1206 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1208 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1209 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1210 struct ost_body *repbody;
1211 struct lustre_handle lh = { 0 };
1212 struct ofd_object *fo;
1214 enum ldlm_mode lock_mode = LCK_PR;
1215 ktime_t kstart = ktime_get();
1220 LASSERT(tsi->tsi_ost_body != NULL);
1222 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1223 if (repbody == NULL)
1226 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1227 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1229 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1230 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1233 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1236 rc = tgt_extent_lock(tsi->tsi_env,
1237 tsi->tsi_tgt->lut_obd->obd_namespace,
1238 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1244 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1246 GOTO(out, rc = PTR_ERR(fo));
1248 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1252 obdo_from_la(&repbody->oa, &fti->fti_attr,
1253 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1255 /* Store object version in reply */
1256 curr_version = dt_version_get(tsi->tsi_env,
1257 ofd_object_child(fo));
1258 if ((__s64)curr_version != -EOPNOTSUPP) {
1259 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1260 repbody->oa.o_data_version = curr_version;
1263 if (fo->ofo_ff.ff_layout_version > 0) {
1264 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1265 repbody->oa.o_layout_version =
1266 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1268 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1269 PFID(&tsi->tsi_fid),
1270 repbody->oa.o_layout_version);
1274 ofd_object_put(tsi->tsi_env, fo);
1277 tgt_data_unlock(&lh, lock_mode);
1279 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1280 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1282 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1283 repbody->oa.o_flags = OBD_FL_FLUSH;
1289 * OFD request handler for OST_SETATTR RPC.
1291 * This is OFD-specific part of request handling. It finds the OFD object
1292 * by its FID, sets attributes from request and packs result to the reply.
1294 * \param[in] tsi target session environment for this request
1296 * \retval 0 if successful
1297 * \retval negative value on error
1299 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1301 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1302 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1303 struct ost_body *body = tsi->tsi_ost_body;
1304 struct ost_body *repbody;
1305 struct ldlm_resource *res;
1306 struct ofd_object *fo;
1307 ktime_t kstart = ktime_get();
1312 LASSERT(body != NULL);
1314 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1315 if (repbody == NULL)
1318 repbody->oa.o_oi = body->oa.o_oi;
1319 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1321 /* This would be very bad - accidentally truncating a file when
1322 * changing the time or similar - bug 12203. */
1323 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1324 body->oa.o_size != OBD_OBJECT_EOF) {
1325 static char mdsinum[48];
1327 if (body->oa.o_valid & OBD_MD_FLFID)
1328 snprintf(mdsinum, sizeof(mdsinum) - 1,
1329 "of parent "DFID, body->oa.o_parent_seq,
1330 body->oa.o_parent_oid, 0);
1334 CERROR("%s: setattr from %s is trying to truncate object "DFID
1335 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1336 PFID(&tsi->tsi_fid), mdsinum);
1340 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1342 GOTO(out, rc = PTR_ERR(fo));
1344 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1345 fti->fti_attr.la_valid &= ~LA_TYPE;
1347 /* setting objects attributes (including owner/group) */
1348 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1352 obdo_from_la(&repbody->oa, &fti->fti_attr,
1353 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1355 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1356 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1359 ofd_object_put(tsi->tsi_env, fo);
1362 /* we do not call this before to avoid lu_object_find() in
1363 * ->lvbo_update() holding another reference on the object.
1364 * otherwise concurrent destroy can make the object unavailable
1365 * for 2nd lu_object_find() waiting for the first reference
1366 * to go... deadlock! */
1367 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1368 &tsi->tsi_resid, LDLM_EXTENT, 0);
1370 ldlm_res_lvbo_update(res, NULL, 0);
1371 ldlm_resource_putref(res);
1378 * Destroy OST orphans.
1380 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1381 * set then we must destroy possible orphaned objects.
1383 * \param[in] env execution environment
1384 * \param[in] exp OBD export
1385 * \param[in] ofd OFD device
1386 * \param[in] oa obdo structure for reply
1388 * \retval 0 if successful
1389 * \retval negative value on error
1391 static int ofd_orphans_destroy(const struct lu_env *env,
1392 struct obd_export *exp,
1393 struct ofd_device *ofd, struct obdo *oa)
1395 struct ofd_thread_info *info = ofd_info(env);
1396 struct lu_fid *fid = &info->fti_fid;
1397 struct ost_id *oi = &oa->o_oi;
1398 struct ofd_seq *oseq;
1399 u64 seq = ostid_seq(oi);
1400 u64 end_id = ostid_id(oi);
1408 oseq = ofd_seq_get(ofd, seq);
1410 CERROR("%s: Can not find seq for "DOSTID"\n",
1411 ofd_name(ofd), POSTID(oi));
1416 last = ofd_seq_last_oid(oseq);
1419 LASSERT(exp != NULL);
1420 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1422 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1425 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1426 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1428 while (oid > end_id) {
1429 rc = fid_set_id(fid, oid);
1430 if (unlikely(rc != 0))
1433 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1434 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1435 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1436 /* this is pretty fatal... */
1437 CEMERG("%s: error destroying precreated id "
1439 ofd_name(ofd), PFID(fid), rc);
1443 ofd_seq_last_oid_set(oseq, oid);
1444 /* update last_id on disk periodically so that if we
1445 * restart * we don't need to re-scan all of the just
1446 * deleted objects. */
1447 if ((oid & 511) == 0)
1448 ofd_seq_last_oid_write(env, ofd, oseq);
1452 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1453 ofd_name(ofd), seq, oid);
1457 ofd_seq_last_oid_set(oseq, oid);
1458 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1460 /* don't reuse orphan object, return last used objid */
1461 rc = ostid_set_id(oi, last);
1467 ofd_seq_put(env, oseq);
1472 * OFD request handler for OST_CREATE RPC.
1474 * This is OFD-specific part of request handling. Its main purpose is to
1475 * create new data objects on OST, but it also used to destroy orphans.
1477 * \param[in] tsi target session environment for this request
1479 * \retval 0 if successful
1480 * \retval negative value on error
1482 static int ofd_create_hdl(struct tgt_session_info *tsi)
1484 struct ptlrpc_request *req = tgt_ses_req(tsi);
1485 struct ost_body *repbody;
1486 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1487 struct obdo *rep_oa;
1488 struct obd_export *exp = tsi->tsi_exp;
1489 struct ofd_device *ofd = ofd_exp(exp);
1490 u64 seq = ostid_seq(&oa->o_oi);
1491 u64 oid = ostid_id(&oa->o_oi);
1492 struct ofd_seq *oseq;
1495 ktime_t kstart = ktime_get();
1501 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1504 if (ofd->ofd_no_precreate)
1507 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1508 if (repbody == NULL)
1511 down_read(&ofd->ofd_lastid_rwsem);
1512 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1513 * we may do that in the future.
1514 * Return -ENOSPC until the LAST_ID rebuilt. */
1515 if (unlikely(ofd->ofd_lastid_rebuilding))
1516 GOTO(out_sem, rc = -ENOSPC);
1518 rep_oa = &repbody->oa;
1519 rep_oa->o_oi = oa->o_oi;
1521 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1523 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1525 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1527 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1528 ofd_name(ofd), seq, PTR_ERR(oseq));
1529 GOTO(out_sem, rc = -EINVAL);
1532 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1533 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1534 if (!ofd_obd(ofd)->obd_recovering ||
1535 oid > ofd_seq_last_oid(oseq)) {
1536 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1537 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1538 ofd_seq_last_oid(oseq));
1539 GOTO(out_nolock, rc = -EINVAL);
1541 /* Do nothing here, we re-create objects during recovery
1542 * upon write replay, see ofd_preprw_write() */
1543 GOTO(out_nolock, rc = 0);
1545 /* former ofd_handle_precreate */
1546 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1547 (oa->o_flags & OBD_FL_DELORPHAN)) {
1548 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1550 /* destroy orphans */
1551 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1552 exp->exp_conn_cnt) {
1553 CERROR("%s: dropping old orphan cleanup request\n",
1555 GOTO(out_nolock, rc = 0);
1557 /* This causes inflight precreates to abort and drop lock */
1558 oseq->os_destroys_in_progress = 1;
1559 mutex_lock(&oseq->os_create_lock);
1560 if (!oseq->os_destroys_in_progress) {
1562 "%s:[%llu] destroys_in_progress already cleared\n",
1563 ofd_name(ofd), seq);
1564 rc = ostid_set_id(&rep_oa->o_oi,
1565 ofd_seq_last_oid(oseq));
1568 diff = oid - ofd_seq_last_oid(oseq);
1569 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1570 ofd_seq_last_oid(oseq), diff);
1571 if (-diff > OST_MAX_PRECREATE) {
1572 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1573 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1574 "(%llu), trust the OST\n",
1575 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1576 PFID(&oseq->os_oi.oi_fid),
1577 ofd_seq_last_oid(oseq));
1579 /* Let MDS know that we are so far ahead. */
1580 rc = ostid_set_id(&rep_oa->o_oi,
1581 ofd_seq_last_oid(oseq) + 1);
1582 } else if (diff < 0) {
1583 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1585 oseq->os_destroys_in_progress = 0;
1587 /* XXX: Used by MDS for the first time! */
1588 oseq->os_destroys_in_progress = 0;
1591 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1592 ofd->ofd_lastid_gen)) {
1593 /* Keep the export ref so we can send the reply. */
1594 ofd_obd_disconnect(class_export_get(exp));
1595 GOTO(out_nolock, rc = -ENOTCONN);
1598 mutex_lock(&oseq->os_create_lock);
1599 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1600 exp->exp_conn_cnt) {
1601 CERROR("%s: dropping old precreate request\n",
1605 /* only precreate if seq is 0, IDIF or normal and also o_id
1606 * must be specfied */
1607 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1608 !fid_seq_is_idif(seq)) || oid == 0) {
1609 diff = 1; /* shouldn't we create this right now? */
1611 diff = oid - ofd_seq_last_oid(oseq);
1612 /* Do sync create if the seq is about to used up */
1613 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1614 if (unlikely(oid >= IDIF_MAX_OID - 1))
1616 } else if (fid_seq_is_norm(seq)) {
1618 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1621 CERROR("%s : invalid o_seq "DOSTID"\n",
1622 ofd_name(ofd), POSTID(&oa->o_oi));
1623 GOTO(out, rc = -EINVAL);
1628 CERROR("%s: invalid precreate request for "
1629 DOSTID", last_id %llu. "
1630 "Likely MDS last_id corruption\n",
1631 ofd_name(ofd), POSTID(&oa->o_oi),
1632 ofd_seq_last_oid(oseq));
1633 GOTO(out, rc = -EINVAL);
1638 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1644 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1645 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1646 /* don't enforce grant during orphan recovery */
1647 granted = tgt_grant_create(tsi->tsi_env,
1648 ofd_obd(ofd)->obd_self_export,
1653 CDEBUG(D_HA, "%s: failed to acquire grant "
1654 "space for precreate (%lld): rc = %d\n",
1655 ofd_name(ofd), diff, rc);
1660 /* This can happen if a new OST is formatted and installed
1661 * in place of an old one at the same index. Instead of
1662 * precreating potentially millions of deleted old objects
1663 * (possibly filling the OST), only precreate the last batch.
1664 * LFSCK will eventually clean up any orphans. LU-14 */
1665 if (diff > 5 * OST_MAX_PRECREATE) {
1666 diff = OST_MAX_PRECREATE / 2;
1667 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1668 "OST replaced or reformatted: "
1669 "LFSCK will clean up",
1672 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1673 "%u larger than the LAST_ID "DOSTID", only "
1674 "precreating the last %lld objects.\n",
1675 ofd_name(ofd), POSTID(&oa->o_oi),
1676 5 * OST_MAX_PRECREATE,
1677 POSTID(&oseq->os_oi), diff);
1678 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1682 next_id = ofd_seq_last_oid(oseq) + 1;
1683 count = ofd_precreate_batch(ofd, (int)diff);
1685 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1686 " at %llu\n", ofd_name(ofd),
1687 count, seq, next_id);
1689 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1690 && ktime_get_seconds() > enough_time) {
1691 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1692 " created at a rate of %d/s\n",
1693 ofd_name(ofd), created, diff + created,
1694 created / DISK_TIMEOUT);
1698 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1699 oseq, count, sync_trans);
1703 } else if (rc < 0) {
1709 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1710 LCONSOLE_WARN("%s: can't create the same count of"
1711 " objects when replaying the request"
1712 " (diff is %lld). see LU-4621\n",
1713 ofd_name(ofd), diff);
1716 /* some objects got created, we can return
1717 * them, even if last creation failed */
1720 CERROR("%s: unable to precreate: rc = %d\n",
1723 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1724 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1725 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1730 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1734 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1735 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1736 if (unlikely(!oseq->os_last_id_synced))
1737 oseq->os_last_id_synced = 1;
1739 mutex_unlock(&oseq->os_create_lock);
1742 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1743 ofd_seq_put(tsi->tsi_env, oseq);
1746 up_read(&ofd->ofd_lastid_rwsem);
1751 * OFD request handler for OST_DESTROY RPC.
1753 * This is OFD-specific part of request handling. It destroys data objects
1754 * related to destroyed object on MDT.
1756 * \param[in] tsi target session environment for this request
1758 * \retval 0 if successful
1759 * \retval negative value on error
1761 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1763 const struct ost_body *body = tsi->tsi_ost_body;
1764 struct ost_body *repbody;
1765 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1766 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1767 struct lu_fid *fid = &fti->fti_fid;
1768 ktime_t kstart = ktime_get();
1775 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1778 /* This is old case for clients before Lustre 2.4 */
1779 /* If there's a DLM request, cancel the locks mentioned in it */
1780 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1782 struct ldlm_request *dlm;
1784 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1787 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1790 *fid = body->oa.o_oi.oi_fid;
1791 oid = ostid_id(&body->oa.o_oi);
1794 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1796 /* check that o_misc makes sense */
1797 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1798 count = body->oa.o_misc;
1800 count = 1; /* default case - single destroy */
1802 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1803 POSTID(&body->oa.o_oi), count);
1808 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1809 if (lrc == -ENOENT) {
1811 "%s: destroying non-existent object "DFID"\n",
1812 ofd_name(ofd), PFID(fid));
1813 /* rewrite rc with -ENOENT only if it is 0 */
1816 } else if (lrc != 0) {
1817 CERROR("%s: error destroying object "DFID": %d\n",
1818 ofd_name(ofd), PFID(fid), lrc);
1824 lrc = fid_set_id(fid, oid);
1825 if (unlikely(lrc != 0 && count > 0))
1826 GOTO(out, rc = lrc);
1829 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1830 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1835 fid_to_ostid(fid, &repbody->oa.o_oi);
1840 * OFD request handler for OST_STATFS RPC.
1842 * This function gets statfs data from storage as part of request
1845 * \param[in] tsi target session environment for this request
1847 * \retval 0 if successful
1848 * \retval negative value on error
1850 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1852 ktime_t kstart = ktime_get();
1853 struct obd_statfs *osfs;
1858 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1860 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1862 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1863 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1865 CERROR("%s: statfs failed: rc = %d\n",
1866 tgt_name(tsi->tsi_tgt), rc);
1868 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1871 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1872 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1878 * OFD request handler for OST_SYNC RPC.
1880 * Sync object data or all filesystem data to the disk and pack the
1883 * \param[in] tsi target session environment for this request
1885 * \retval 0 if successful
1886 * \retval negative value on error
1888 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1890 struct ost_body *body = tsi->tsi_ost_body;
1891 struct ost_body *repbody;
1892 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1893 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1894 struct ofd_object *fo = NULL;
1895 ktime_t kstart = ktime_get();
1900 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1902 /* if no objid is specified, it means "sync whole filesystem" */
1903 if (!fid_is_zero(&tsi->tsi_fid)) {
1904 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1906 RETURN(PTR_ERR(fo));
1909 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1910 fo != NULL ? ofd_object_child(fo) : NULL,
1911 repbody->oa.o_size, repbody->oa.o_blocks);
1915 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1916 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1920 repbody->oa.o_oi = body->oa.o_oi;
1921 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1923 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1925 obdo_from_la(&repbody->oa, &fti->fti_attr,
1928 /* don't return rc from getattr */
1933 ofd_object_put(tsi->tsi_env, fo);
1938 * OFD request handler for OST_FALLOCATE RPC.
1940 * This is part of request processing. Validate request fields,
1941 * preallocate the given OFD object and pack reply.
1943 * \param[in] tsi target session environment for this request
1945 * \retval 0 if successful
1946 * \retval negative value on error
1948 static int ofd_fallocate_hdl(struct tgt_session_info *tsi)
1950 struct obdo *oa = &tsi->tsi_ost_body->oa;
1951 struct ost_body *repbody;
1952 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1953 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1954 struct ldlm_resource *res;
1955 struct ofd_object *fo;
1957 struct lustre_handle lh = { 0, };
1961 ktime_t kstart = ktime_get();
1963 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1964 if (repbody == NULL)
1965 RETURN(err_serious(-ENOMEM));
1968 * fallocate start and end are passed in o_size, o_blocks
1971 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1972 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1973 RETURN(err_serious(-EPROTO));
1977 mode = oa->o_falloc_mode;
1979 * mode == 0 (which is standard prealloc) and PUNCH is supported
1980 * Rest of mode options are not supported yet.
1982 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1983 RETURN(-EOPNOTSUPP);
1985 /* PUNCH_HOLE mode should always be accompanied with KEEP_SIZE flag
1986 * Check that and add the missing flag for such invalid call with
1989 if (mode & FALLOC_FL_PUNCH_HOLE && !(mode & FALLOC_FL_KEEP_SIZE)) {
1990 CWARN("%s: PUNCH mode misses KEEP_SIZE flag, setting it\n",
1991 tsi->tsi_tgt->lut_obd->obd_name);
1992 mode |= FALLOC_FL_KEEP_SIZE;
1995 repbody->oa.o_oi = oa->o_oi;
1996 repbody->oa.o_valid = OBD_MD_FLID;
1998 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1999 oa->o_flags & OBD_FL_SRVLOCK;
2002 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid,
2003 start, end, &lh, LCK_PW, &flags);
2008 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2011 GOTO(out, rc = PTR_ERR(fo));
2013 la_from_obdo(&info->fti_attr, oa,
2014 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2016 rc = ofd_object_fallocate(tsi->tsi_env, fo, start, end, mode,
2017 &info->fti_attr, oa);
2021 rc = ofd_attr_get(tsi->tsi_env, fo, &info->fti_attr);
2023 obdo_from_la(&repbody->oa, &info->fti_attr,
2028 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PREALLOC,
2029 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2033 ofd_object_put(tsi->tsi_env, fo);
2036 tgt_data_unlock(&lh, LCK_PW);
2038 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2041 struct ost_lvb *res_lvb;
2043 ldlm_res_lvbo_update(res, NULL, 0);
2044 res_lvb = res->lr_lvb_data;
2046 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2047 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2049 repbody->oa.o_valid |= OBD_MD_FLSIZE;
2050 repbody->oa.o_size = res_lvb->lvb_size;
2052 ldlm_resource_putref(res);
2060 * OFD request handler for OST_PUNCH RPC.
2062 * This is part of request processing. Validate request fields,
2063 * punch (truncate) the given OFD object and pack reply.
2065 * \param[in] tsi target session environment for this request
2067 * \retval 0 if successful
2068 * \retval negative value on error
2070 static int ofd_punch_hdl(struct tgt_session_info *tsi)
2072 const struct obdo *oa = &tsi->tsi_ost_body->oa;
2073 struct ost_body *repbody;
2074 struct ofd_thread_info *info = tsi2ofd_info(tsi);
2075 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
2076 struct ldlm_resource *res;
2077 struct ofd_object *fo;
2079 struct lustre_handle lh = { 0, };
2082 ktime_t kstart = ktime_get();
2087 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
2089 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
2090 BUILD_BUG_ON(!(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK));
2092 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2093 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2094 RETURN(err_serious(-EPROTO));
2096 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2097 if (repbody == NULL)
2098 RETURN(err_serious(-ENOMEM));
2100 /* punch start,end are passed in o_size,o_blocks throught wire */
2104 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2107 /* standard truncate optimization: if file body is completely
2108 * destroyed, don't send data back to the server. */
2110 flags |= LDLM_FL_AST_DISCARD_DATA;
2112 repbody->oa.o_oi = oa->o_oi;
2113 repbody->oa.o_valid = OBD_MD_FLID;
2115 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2116 oa->o_flags & OBD_FL_SRVLOCK;
2119 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
2120 end, &lh, LCK_PW, &flags);
2125 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2126 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2127 oa->o_valid, start, end);
2129 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2132 GOTO(out, rc = PTR_ERR(fo));
2134 la_from_obdo(&info->fti_attr, oa,
2135 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2136 info->fti_attr.la_size = start;
2137 info->fti_attr.la_valid |= LA_SIZE;
2139 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2144 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2145 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2148 ofd_object_put(tsi->tsi_env, fo);
2151 tgt_data_unlock(&lh, LCK_PW);
2153 /* we do not call this before to avoid lu_object_find() in
2154 * ->lvbo_update() holding another reference on the object.
2155 * otherwise concurrent destroy can make the object unavailable
2156 * for 2nd lu_object_find() waiting for the first reference
2157 * to go... deadlock! */
2158 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2161 struct ost_lvb *res_lvb;
2163 ldlm_res_lvbo_update(res, NULL, 0);
2164 res_lvb = res->lr_lvb_data;
2165 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2166 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2167 ldlm_resource_putref(res);
2173 static int ofd_ladvise_prefetch(const struct lu_env *env,
2174 struct ofd_object *fo,
2175 struct niobuf_local *lnb,
2176 __u64 start, __u64 end, enum dt_bufs_type dbt)
2178 struct ofd_thread_info *info = ofd_info(env);
2179 pgoff_t start_index, end_index, pages;
2180 struct niobuf_remote rnb;
2181 unsigned long nr_local;
2187 ofd_read_lock(env, fo);
2188 if (!ofd_object_exists(fo))
2189 GOTO(out_unlock, rc = -ENOENT);
2191 rc = ofd_attr_get(env, fo, &info->fti_attr);
2193 GOTO(out_unlock, rc);
2195 if (end > info->fti_attr.la_size)
2196 end = info->fti_attr.la_size;
2199 GOTO(out_unlock, rc);
2201 /* We need page aligned offset and length */
2202 start_index = start >> PAGE_SHIFT;
2203 end_index = (end - 1) >> PAGE_SHIFT;
2204 pages = end_index - start_index + 1;
2206 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2207 PTLRPC_MAX_BRW_PAGES;
2208 rnb.rnb_offset = start_index << PAGE_SHIFT;
2209 rnb.rnb_len = nr_local << PAGE_SHIFT;
2210 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb,
2211 PTLRPC_MAX_BRW_PAGES, dbt);
2212 if (unlikely(rc < 0))
2215 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2216 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2219 start_index += nr_local;
2224 ofd_read_unlock(env, fo);
2229 * OFD request handler for OST_LADVISE RPC.
2231 * Tune cache or perfetch policies according to advices.
2233 * \param[in] tsi target session environment for this request
2235 * \retval 0 if successful
2236 * \retval negative errno on error
2238 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2240 struct ptlrpc_request *req = tgt_ses_req(tsi);
2241 struct obd_export *exp = tsi->tsi_exp;
2242 struct ofd_device *ofd = ofd_exp(exp);
2243 struct ost_body *body, *repbody;
2244 struct ofd_thread_info *info;
2245 struct ofd_object *fo;
2246 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2247 const struct lu_env *env = svc_thread->t_env;
2248 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2249 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2250 struct lu_ladvise *ladvise;
2252 struct ladvise_hdr *ladvise_hdr;
2253 struct obd_ioobj ioo;
2254 struct lustre_handle lockh = { 0 };
2257 struct dt_object *dob;
2263 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2264 body = tsi->tsi_ost_body;
2266 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2267 RETURN(err_serious(-EPROTO));
2269 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2270 &RMF_OST_LADVISE_HDR);
2271 if (ladvise_hdr == NULL)
2272 RETURN(err_serious(-EPROTO));
2274 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2275 ladvise_hdr->lah_count < 1)
2276 RETURN(err_serious(-EPROTO));
2278 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2279 RETURN(err_serious(-EPROTO));
2281 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2282 if (ladvise == NULL)
2283 RETURN(err_serious(-EPROTO));
2285 num_advise = req_capsule_get_size(&req->rq_pill,
2286 &RMF_OST_LADVISE, RCL_CLIENT) /
2288 if (num_advise < ladvise_hdr->lah_count)
2289 RETURN(err_serious(-EPROTO));
2291 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2292 repbody->oa = body->oa;
2294 info = ofd_info_init(env, exp);
2296 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2297 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2301 fo = ofd_object_find(env, ofd, &info->fti_fid);
2306 LASSERT(fo != NULL);
2307 dob = ofd_object_child(fo);
2309 if (ptlrpc_connection_is_local(exp->exp_connection))
2310 dbt |= DT_BUFS_TYPE_LOCAL;
2312 for (i = 0; i < num_advise; i++, ladvise++) {
2313 start = ladvise->lla_start;
2314 end = ladvise->lla_end;
2316 rc = err_serious(-EPROTO);
2320 /* Handle different advice types */
2321 switch (ladvise->lla_advice) {
2325 case LU_LADVISE_WILLREAD:
2329 ioo.ioo_oid = body->oa.o_oi;
2331 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2332 &tsi->tsi_resid, start, end - 1,
2333 &lockh, LCK_PR, &flags);
2337 req->rq_status = ofd_ladvise_prefetch(env, fo,
2340 tgt_data_unlock(&lockh, LCK_PR);
2342 case LU_LADVISE_DONTNEED:
2343 rc = dt_ladvise(env, dob, ladvise->lla_start,
2344 ladvise->lla_end, LU_LADVISE_DONTNEED);
2351 ofd_object_put(env, fo);
2352 req->rq_status = rc;
2357 * OFD request handler for OST_QUOTACTL RPC.
2359 * This is part of request processing to validate incoming request fields,
2360 * get the requested data from OSD and pack reply.
2362 * \param[in] tsi target session environment for this request
2364 * \retval 0 if successful
2365 * \retval negative value on error
2367 static int ofd_quotactl(struct tgt_session_info *tsi)
2369 struct obd_quotactl *oqctl, *repoqc;
2370 struct lu_nodemap *nodemap;
2371 ktime_t kstart = ktime_get();
2377 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2379 RETURN(err_serious(-EPROTO));
2381 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2383 RETURN(err_serious(-ENOMEM));
2387 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2388 if (IS_ERR(nodemap))
2389 RETURN(PTR_ERR(nodemap));
2392 if (oqctl->qc_type == USRQUOTA)
2393 id = nodemap_map_id(nodemap, NODEMAP_UID,
2394 NODEMAP_CLIENT_TO_FS,
2396 else if (oqctl->qc_type == GRPQUOTA)
2397 id = nodemap_map_id(nodemap, NODEMAP_GID,
2398 NODEMAP_CLIENT_TO_FS,
2401 nodemap_putref(nodemap);
2403 if (repoqc->qc_id != id)
2404 swap(repoqc->qc_id, id);
2406 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2408 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2409 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2411 if (repoqc->qc_id != id)
2412 swap(repoqc->qc_id, id);
2418 * Prolong lock timeout for the given extent.
2420 * This function finds all locks related with incoming request and
2421 * prolongs their timeout.
2423 * If a client is holding a lock for a long time while it sends
2424 * read or write RPCs to the OST for the object under this lock,
2425 * then we don't want the OST to evict the client. Otherwise,
2426 * if the network or disk is very busy then the client may not
2427 * be able to make any progress to clear out dirty pages under
2428 * the lock and the application will fail.
2430 * Every time a Bulk Read/Write (BRW) request arrives for the object
2431 * covered by the lock, extend the timeout on that lock. The RPC should
2432 * contain a lock handle for the lock it is using, but this
2433 * isn't handled correctly by all client versions, and the
2434 * request may cover multiple locks.
2436 * \param[in] tsi target session environment for this request
2437 * \param[in] data struct of data to prolong locks
2440 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2441 struct ldlm_prolong_args *data)
2443 struct obdo *oa = &tsi->tsi_ost_body->oa;
2444 struct ldlm_lock *lock;
2448 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2449 data->lpa_export = tsi->tsi_exp;
2450 data->lpa_resid = tsi->tsi_resid;
2452 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2453 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2454 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2455 data->lpa_extent.end);
2457 if (oa->o_valid & OBD_MD_FLHANDLE) {
2458 /* mostly a request should be covered by only one lock, try
2460 lock = ldlm_handle2lock(&oa->o_handle);
2462 /* Fast path to check if the lock covers the whole IO
2463 * region exclusively. */
2464 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2465 &data->lpa_extent)) {
2467 LASSERT(lock->l_export == data->lpa_export);
2468 ldlm_lock_prolong_one(lock, data);
2469 LDLM_LOCK_PUT(lock);
2470 if (data->lpa_locks_cnt > 0)
2472 /* The lock was destroyed probably lets try
2475 lock->l_last_used = ktime_get();
2476 LDLM_LOCK_PUT(lock);
2481 ldlm_resource_prolong(data);
2486 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2488 * Determine if \a lock and the lock from request \a req are equivalent
2489 * by comparing their resource names, modes, and extents.
2491 * It is used to give priority to read and write RPCs being done
2492 * under this lock so that the client can drop the contended
2493 * lock more quickly and let other clients use it. This improves
2494 * overall performance in the case where the first client gets a
2495 * very large lock extent that prevents other clients from
2496 * submitting their writes.
2498 * \param[in] req ptlrpc_request being processed
2499 * \param[in] lock contended lock to match
2501 * \retval 1 if lock is matched
2502 * \retval 0 otherwise
2504 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2505 struct ldlm_lock *lock)
2507 struct niobuf_remote *rnb;
2508 struct obd_ioobj *ioo;
2509 enum ldlm_mode mode;
2510 struct ldlm_extent ext;
2511 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2515 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2516 LASSERT(ioo != NULL);
2518 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2519 LASSERT(rnb != NULL);
2521 ext.start = rnb->rnb_offset;
2522 rnb += ioo->ioo_bufcnt - 1;
2523 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2525 LASSERT(lock->l_resource != NULL);
2526 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2529 /* a bulk write can only hold a reference on a PW extent lock
2532 mode = LCK_PW | LCK_GROUP;
2533 if (opc == OST_READ)
2534 /* whereas a bulk read can be protected by either a PR or PW
2538 if (!(lock->l_granted_mode & mode))
2541 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2545 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2547 * Check for whether the given PTLRPC request (\a req) is blocking
2548 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2551 * \param[in] req the incoming request
2553 * \retval 1 if \a req is blocking an LDLM lock cancel
2554 * \retval 0 if it is not
2555 * \retval -ESTALE if lock is not found
2557 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2559 struct tgt_session_info *tsi;
2560 struct obd_ioobj *ioo;
2561 struct niobuf_remote *rnb;
2563 struct ldlm_prolong_args pa = { 0 };
2567 /* Don't use tgt_ses_info() to get session info, because lock_match()
2568 * can be called while request has no processing thread yet. */
2569 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2572 * Use LASSERT below because malformed RPCs should have
2573 * been filtered out in tgt_hpreq_handler().
2575 opc = lustre_msg_get_opc(req->rq_reqmsg);
2576 LASSERT(opc == OST_READ || opc == OST_WRITE);
2578 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2579 LASSERT(ioo != NULL);
2581 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2582 LASSERT(rnb != NULL);
2583 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2585 pa.lpa_mode = LCK_PW | LCK_GROUP;
2586 if (opc == OST_READ)
2587 pa.lpa_mode |= LCK_PR;
2589 pa.lpa_extent.start = rnb->rnb_offset;
2590 rnb += ioo->ioo_bufcnt - 1;
2591 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2593 DEBUG_REQ(D_RPCTRACE, req,
2594 "%s %s: refresh rw locks for "DFID" (%llu->%llu)",
2595 tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid),
2596 pa.lpa_extent.start, pa.lpa_extent.end);
2598 ofd_prolong_extent_locks(tsi, &pa);
2600 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n",
2601 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2603 if (pa.lpa_blocks_cnt > 0)
2606 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2610 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2612 * Called after the request has been handled. It refreshes lock timeout again
2613 * so that client has more time to send lock cancel RPC.
2615 * \param[in] req request which is being processed.
2617 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2619 ofd_rw_hpreq_check(req);
2623 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2625 * This function checks if the given lock is the same by its resname, mode
2626 * and extent as one taken from the request.
2627 * It is used to give priority to punch/truncate RPCs that might lead to
2628 * the fastest release of that lock when a lock is contended.
2630 * \param[in] req ptlrpc_request being processed
2631 * \param[in] lock contended lock to match
2633 * \retval 1 if lock is matched
2634 * \retval 0 otherwise
2636 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2637 struct ldlm_lock *lock)
2639 struct tgt_session_info *tsi;
2641 struct ldlm_extent ext;
2645 /* Don't use tgt_ses_info() to get session info, because lock_match()
2646 * can be called while request has no processing thread yet. */
2647 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2650 * Use LASSERT below because malformed RPCs should have
2651 * been filtered out in tgt_hpreq_handler().
2653 LASSERT(tsi->tsi_ost_body != NULL);
2654 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2655 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2658 oa = &tsi->tsi_ost_body->oa;
2659 ext.start = oa->o_size;
2660 ext.end = oa->o_blocks;
2662 LASSERT(lock->l_resource != NULL);
2663 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2666 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2669 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2673 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2675 * High-priority queue request check for whether the given punch request
2676 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2677 * covered by an LDLM lock.
2681 * \param[in] req the incoming request
2683 * \retval 1 if \a req is blocking an LDLM lock cancel
2684 * \retval 0 if it is not
2685 * \retval -ESTALE if lock is not found
2687 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2689 struct tgt_session_info *tsi;
2691 struct ldlm_prolong_args pa = { 0 };
2695 /* Don't use tgt_ses_info() to get session info, because lock_match()
2696 * can be called while request has no processing thread yet. */
2697 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2698 LASSERT(tsi != NULL);
2699 oa = &tsi->tsi_ost_body->oa;
2701 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2702 oa->o_flags & OBD_FL_SRVLOCK));
2704 pa.lpa_mode = LCK_PW | LCK_GROUP;
2705 pa.lpa_extent.start = oa->o_size;
2706 pa.lpa_extent.end = oa->o_blocks;
2709 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2710 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2711 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2713 ofd_prolong_extent_locks(tsi, &pa);
2715 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2716 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2718 if (pa.lpa_blocks_cnt > 0)
2721 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2725 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2727 * Called after the request has been handled. It refreshes lock timeout again
2728 * so that client has more time to send lock cancel RPC.
2730 * \param[in] req request which is being processed.
2732 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2734 ofd_punch_hpreq_check(req);
2737 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2738 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2739 .hpreq_check = ofd_rw_hpreq_check,
2740 .hpreq_fini = ofd_rw_hpreq_fini
2743 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2744 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2745 .hpreq_check = ofd_punch_hpreq_check,
2746 .hpreq_fini = ofd_punch_hpreq_fini
2750 * Assign high priority operations to an IO request.
2752 * Check if the incoming request is a candidate for
2753 * high-priority processing. If it is, assign it a high
2754 * priority operations table.
2756 * \param[in] tsi target session environment for this request
2758 static void ofd_hp_brw(struct tgt_session_info *tsi)
2760 struct niobuf_remote *rnb;
2761 struct obd_ioobj *ioo;
2765 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2766 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2767 if (ioo->ioo_bufcnt > 0) {
2768 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2769 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2771 /* no high priority if server lock is needed */
2772 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2773 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2777 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2781 * Assign high priority operations to an punch request.
2783 * Check if the incoming request is a candidate for
2784 * high-priority processing. If it is, assign it a high
2785 * priority operations table.
2787 * \param[in] tsi target session environment for this request
2789 static void ofd_hp_punch(struct tgt_session_info *tsi)
2791 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2792 /* no high-priority if server lock is needed */
2793 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2794 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2795 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2796 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2798 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2801 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2802 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2803 #define OST_BRW_READ OST_READ
2804 #define OST_BRW_WRITE OST_WRITE
2807 * Table of OFD-specific request handlers
2809 * This table contains all opcodes accepted by OFD and
2810 * specifies handlers for them. The tgt_request_handler()
2811 * uses such table from each target to process incoming
2814 static struct tgt_handler ofd_tgt_handlers[] = {
2815 TGT_RPC_HANDLER(OST_FIRST_OPC,
2816 0, OST_CONNECT, tgt_connect,
2817 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2818 TGT_RPC_HANDLER(OST_FIRST_OPC,
2819 0, OST_DISCONNECT, tgt_disconnect,
2820 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2821 TGT_RPC_HANDLER(OST_FIRST_OPC,
2822 0, OST_SET_INFO, ofd_set_info_hdl,
2823 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2824 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2825 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl),
2826 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2827 OST_SETATTR, ofd_setattr_hdl),
2828 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2829 OST_CREATE, ofd_create_hdl),
2830 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2831 OST_DESTROY, ofd_destroy_hdl),
2832 TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl),
2833 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
2835 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2836 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write,
2838 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2839 OST_PUNCH, ofd_punch_hdl,
2841 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl),
2842 TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl),
2843 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl),
2844 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE, ofd_fallocate_hdl),
2845 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
2848 static struct tgt_opc_slice ofd_common_slice[] = {
2850 .tos_opc_start = OST_FIRST_OPC,
2851 .tos_opc_end = OST_LAST_OPC,
2852 .tos_hs = ofd_tgt_handlers
2855 .tos_opc_start = OBD_FIRST_OPC,
2856 .tos_opc_end = OBD_LAST_OPC,
2857 .tos_hs = tgt_obd_handlers
2860 .tos_opc_start = LDLM_FIRST_OPC,
2861 .tos_opc_end = LDLM_LAST_OPC,
2862 .tos_hs = tgt_dlm_handlers
2865 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2866 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2867 .tos_hs = tgt_out_handlers
2870 .tos_opc_start = SEQ_FIRST_OPC,
2871 .tos_opc_end = SEQ_LAST_OPC,
2872 .tos_hs = seq_handlers
2875 .tos_opc_start = LFSCK_FIRST_OPC,
2876 .tos_opc_end = LFSCK_LAST_OPC,
2877 .tos_hs = tgt_lfsck_handlers
2880 .tos_opc_start = SEC_FIRST_OPC,
2881 .tos_opc_end = SEC_LAST_OPC,
2882 .tos_hs = tgt_sec_ctx_handlers
2889 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2890 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2893 * Implementation of lu_context_key::lct_key_exit.
2895 * Optional method called on lu_context_exit() for all allocated
2897 * It is used in OFD to sanitize context values which may be re-used
2898 * during another request processing by the same thread.
2900 * \param[in] ctx execution context
2901 * \param[in] key context key
2902 * \param[in] data ofd_thread_info
2904 static void ofd_key_exit(const struct lu_context *ctx,
2905 struct lu_context_key *key, void *data)
2907 struct ofd_thread_info *info = data;
2909 info->fti_env = NULL;
2910 info->fti_exp = NULL;
2913 info->fti_pre_version = 0;
2915 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2918 struct lu_context_key ofd_thread_key = {
2919 .lct_tags = LCT_DT_THREAD,
2920 .lct_init = ofd_key_init,
2921 .lct_fini = ofd_key_fini,
2922 .lct_exit = ofd_key_exit
2926 * Initialize OFD device according to parameters in the config log \a cfg.
2928 * This is the main starting point of OFD initialization. It fills all OFD
2929 * parameters with their initial values and calls other initializing functions
2930 * to set up all OFD subsystems.
2932 * \param[in] env execution environment
2933 * \param[in] m OFD device
2934 * \param[in] ldt LU device type of OFD
2935 * \param[in] cfg configuration log
2937 * \retval 0 if successful
2938 * \retval negative value on error
2940 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2941 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2943 const char *dev = lustre_cfg_string(cfg, 0);
2944 struct ofd_thread_info *info = NULL;
2945 struct obd_device *obd;
2946 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2948 struct nm_config_file *nodemap_config;
2949 struct obd_device_target *obt;
2955 obd = class_name2obd(dev);
2957 CERROR("Cannot find obd with name %s\n", dev);
2961 rc = lu_env_refill((struct lu_env *)env);
2966 obt->obt_magic = OBT_MAGIC;
2968 spin_lock_init(&m->ofd_flags_lock);
2969 m->ofd_raid_degraded = 0;
2970 m->ofd_sync_journal = 0;
2972 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2974 m->ofd_seq_count = 0;
2975 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2976 spin_lock_init(&m->ofd_inconsistency_lock);
2978 m->ofd_access_log_mask = -1; /* Log all accesses if enabled. */
2980 spin_lock_init(&m->ofd_batch_lock);
2981 init_rwsem(&m->ofd_lastid_rwsem);
2983 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2984 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2985 /* set this lu_device to obd, because error handling need it */
2986 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2988 /* No connection accepted until configurations will finish */
2989 spin_lock(&obd->obd_dev_lock);
2990 obd->obd_no_conn = 1;
2991 spin_unlock(&obd->obd_dev_lock);
2992 obd->obd_replayable = 1;
2993 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2994 char *str = lustre_cfg_string(cfg, 4);
2996 if (strchr(str, 'n')) {
2997 CWARN("%s: recovery disabled\n", obd->obd_name);
2998 obd->obd_replayable = 0;
3002 info = ofd_info_init(env, NULL);
3006 rc = ofd_stack_init(env, m, cfg, &lmd_flags);
3008 CERROR("%s: can't init device stack, rc %d\n",
3013 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
3014 ofd_procfs_add_brw_stats_symlink(m);
3017 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
3018 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
3019 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
3020 LDLM_NAMESPACE_SERVER,
3021 LDLM_NAMESPACE_GREEDY,
3023 if (IS_ERR(m->ofd_namespace)) {
3024 rc = PTR_ERR(m->ofd_namespace);
3025 CERROR("%s: unable to create server namespace: rc = %d\n",
3027 m->ofd_namespace = NULL;
3028 GOTO(err_fini_stack, rc);
3030 /* set obd_namespace for compatibility with old code */
3031 obd->obd_namespace = m->ofd_namespace;
3032 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
3033 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
3034 m->ofd_namespace->ns_lvbp = m;
3036 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3037 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
3039 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
3040 OBD_FAIL_OST_ALL_REQUEST_NET,
3041 OBD_FAIL_OST_ALL_REPLY_NET);
3043 GOTO(err_free_ns, rc);
3045 if (lmd_flags & LMD_FLG_SKIP_LFSCK)
3046 m->ofd_skip_lfsck = 1;
3047 if (lmd_flags & LMD_FLG_LOCAL_RECOV)
3048 m->ofd_lut.lut_local_recovery = 1;
3050 rc = ofd_tunables_init(m);
3052 GOTO(err_fini_lut, rc);
3054 tgd->tgd_reserved_pcnt = 0;
3056 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
3057 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
3058 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
3059 OFD_PRECREATE_SMALL_FS)
3060 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
3061 m->ofd_atime_diff = OFD_DEF_ATIME_DIFF;
3063 rc = ofd_fs_setup(env, m, obd);
3065 GOTO(err_fini_proc, rc);
3067 fid.f_seq = FID_SEQ_LOCAL_NAME;
3070 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
3073 GOTO(err_fini_fs, rc);
3075 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
3077 if (IS_ERR(nodemap_config)) {
3078 rc = PTR_ERR(nodemap_config);
3080 GOTO(err_fini_los, rc);
3082 obt->obt_nodemap_config_file = nodemap_config;
3085 rc = ofd_start_inconsistency_verification_thread(m);
3087 GOTO(err_fini_nm, rc);
3089 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
3094 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
3095 obt->obt_nodemap_config_file = NULL;
3097 local_oid_storage_fini(env, m->ofd_los);
3100 ofd_fs_cleanup(env, m);
3104 tgt_fini(env, &m->ofd_lut);
3106 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3107 obd->obd_namespace = m->ofd_namespace = NULL;
3109 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3114 * Stop the OFD device
3116 * This function stops the OFD device and all its subsystems.
3117 * This is the end of OFD lifecycle.
3119 * \param[in] env execution environment
3120 * \param[in] m OFD device
3122 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3124 struct obd_device *obd = ofd_obd(m);
3125 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3126 struct lfsck_stop stop;
3128 stop.ls_status = LS_PAUSED;
3130 lfsck_stop(env, m->ofd_osd, &stop);
3131 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3132 target_recovery_fini(obd);
3133 if (m->ofd_namespace != NULL)
3134 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3135 d->ld_obd->obd_force);
3137 obd_exports_barrier(obd);
3138 obd_zombie_barrier();
3141 tgt_fini(env, &m->ofd_lut);
3142 ofd_stop_inconsistency_verification_thread(m);
3143 lfsck_degister(env, m->ofd_osd);
3144 ofd_fs_cleanup(env, m);
3145 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3146 obd->u.obt.obt_nodemap_config_file = NULL;
3148 if (m->ofd_namespace != NULL) {
3149 ldlm_namespace_free_post(m->ofd_namespace);
3150 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3153 ofd_access_log_delete(m->ofd_access_log);
3154 m->ofd_access_log = NULL;
3156 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3158 LASSERT(atomic_read(&d->ld_ref) == 0);
3159 server_put_mount(obd->obd_name, true);
3164 * Implementation of lu_device_type_operations::ldto_device_fini.
3166 * Finalize device. Dual to ofd_device_init(). It is called from
3167 * obd_precleanup() and stops the current device.
3169 * \param[in] env execution environment
3170 * \param[in] d LU device of OFD
3174 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3175 struct lu_device *d)
3178 ofd_fini(env, ofd_dev(d));
3183 * Implementation of lu_device_type_operations::ldto_device_free.
3185 * Free OFD device. Dual to ofd_device_alloc().
3187 * \param[in] env execution environment
3188 * \param[in] d LU device of OFD
3192 static struct lu_device *ofd_device_free(const struct lu_env *env,
3193 struct lu_device *d)
3195 struct ofd_device *m = ofd_dev(d);
3197 dt_device_fini(&m->ofd_dt_dev);
3203 * Implementation of lu_device_type_operations::ldto_device_alloc.
3205 * This function allocates the new OFD device. It is called from
3206 * obd_setup() if OBD device had lu_device_type defined.
3208 * \param[in] env execution environment
3209 * \param[in] t lu_device_type of OFD device
3210 * \param[in] cfg configuration log
3212 * \retval pointer to the lu_device of just allocated OFD
3213 * \retval ERR_PTR of return value on error
3215 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3216 struct lu_device_type *t,
3217 struct lustre_cfg *cfg)
3219 struct ofd_device *m;
3220 struct lu_device *l;
3225 return ERR_PTR(-ENOMEM);
3227 l = &m->ofd_dt_dev.dd_lu_dev;
3228 dt_device_init(&m->ofd_dt_dev, t);
3229 rc = ofd_init0(env, m, t, cfg);
3231 ofd_device_free(env, l);
3238 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3239 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3241 static const struct lu_device_type_operations ofd_device_type_ops = {
3242 .ldto_init = ofd_type_init,
3243 .ldto_fini = ofd_type_fini,
3245 .ldto_start = ofd_type_start,
3246 .ldto_stop = ofd_type_stop,
3248 .ldto_device_alloc = ofd_device_alloc,
3249 .ldto_device_free = ofd_device_free,
3250 .ldto_device_fini = ofd_device_fini
3253 static struct lu_device_type ofd_device_type = {
3254 .ldt_tags = LU_DEVICE_DT,
3255 .ldt_name = LUSTRE_OST_NAME,
3256 .ldt_ops = &ofd_device_type_ops,
3257 .ldt_ctx_tags = LCT_DT_THREAD
3261 * Initialize OFD module.
3263 * This function is called upon module loading. It registers OFD device type
3264 * and prepares all in-memory structures used by all OFD devices.
3266 * \retval 0 if successful
3267 * \retval negative value on error
3269 static int __init ofd_init(void)
3273 rc = lu_kmem_init(ofd_caches);
3277 rc = ofd_access_log_module_init();
3281 rc = class_register_type(&ofd_obd_ops, NULL, true,
3282 LUSTRE_OST_NAME, &ofd_device_type);
3284 goto out_ofd_access_log;
3289 ofd_access_log_module_exit();
3291 lu_kmem_fini(ofd_caches);
3299 * This function is called upon OFD module unloading.
3300 * It frees all related structures and unregisters OFD device type.
3302 static void __exit ofd_exit(void)
3304 class_unregister_type(LUSTRE_OST_NAME);
3305 ofd_access_log_module_exit();
3306 lu_kmem_fini(ofd_caches);
3309 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3310 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3311 MODULE_VERSION(LUSTRE_VERSION_STRING);
3312 MODULE_LICENSE("GPL");
3314 module_init(ofd_init);
3315 module_exit(ofd_exit);