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, 2014 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 lustre/doc/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 <lustre_param.h>
73 #include <lustre_fid.h>
74 #include <lustre_lfsck.h>
75 #include <lustre/lustre_idl.h>
76 #include <lustre_dlm.h>
77 #include <lustre_quota.h>
78 #include <lustre_nodemap.h>
80 #include "ofd_internal.h"
82 /* Slab for OFD object allocation */
83 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)
166 const char *dev = lustre_cfg_string(cfg, 0);
168 struct ofd_thread_info *info = ofd_info(env);
169 struct lustre_mount_info *lmi;
175 lmi = server_get_mount(dev);
177 CERROR("Cannot get mount info for %s!\n", dev);
181 /* find bottom osd */
182 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
186 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
187 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
188 OBD_FREE(osdname, MTI_NAME_MAXLEN);
192 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
194 m->ofd_osd = lu2dt_dev(d);
196 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
197 "%s-osd", lustre_cfg_string(cfg, 0));
203 * Finalize the device stack OFD-OSD.
205 * This function cleans OFD-OSD device stack and
206 * disconnects OFD from the OSD.
208 * \param[in] env execution environment
209 * \param[in] m OFD device
210 * \param[in] top top device of stack
212 * \retval 0 if successful
213 * \retval negative value on error
215 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
216 struct lu_device *top)
218 struct obd_device *obd = ofd_obd(m);
219 struct lustre_cfg_bufs bufs;
220 struct lustre_cfg *lcfg;
225 lu_site_purge(env, top->ld_site, ~0);
226 /* process cleanup, pass mdt obd name to get obd umount flags */
227 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
232 lustre_cfg_bufs_set_string(&bufs, 1, flags);
233 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
235 CERROR("Cannot alloc lcfg!\n");
240 top->ld_ops->ldo_process_config(env, top, lcfg);
241 lustre_cfg_free(lcfg);
243 lu_site_purge(env, top->ld_site, ~0);
244 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
245 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
246 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
249 LASSERT(m->ofd_osd_exp);
250 obd_disconnect(m->ofd_osd_exp);
255 /* For interoperability, see mdt_interop_param[]. */
256 static struct cfg_interop_param ofd_interop_param[] = {
257 { "ost.quota_type", NULL },
262 * Check if parameters are symlinks to the OSD.
264 * Some parameters were moved from ofd to osd and only their
265 * symlinks were kept in ofd by LU-3106. They are:
266 * -writehthrough_cache_enable
267 * -readcache_max_filesize
271 * Since they are not included by the static lprocfs var list, a pre-check
272 * is added for them to avoid "unknown param" errors. If they are matched
273 * in this check, they will be passed to the OSD directly.
275 * \param[in] param parameters to check
277 * \retval true if param is symlink to OSD param
280 static bool match_symlink_param(char *param)
285 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
286 sval = strchr(param, '=');
288 paramlen = sval - param;
289 if (strncmp(param, "writethrough_cache_enable",
291 strncmp(param, "readcache_max_filesize",
293 strncmp(param, "read_cache_enable",
295 strncmp(param, "brw_stats", paramlen) == 0)
304 * Process various configuration parameters.
306 * This function is used by MGS to process specific configurations and
307 * pass them through to the next device in server stack, i.e. the OSD.
309 * \param[in] env execution environment
310 * \param[in] d LU device of OFD
311 * \param[in] cfg parameters to process
313 * \retval 0 if successful
314 * \retval negative value on error
316 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
317 struct lustre_cfg *cfg)
319 struct ofd_device *m = ofd_dev(d);
320 struct dt_device *dt_next = m->ofd_osd;
321 struct lu_device *next = &dt_next->dd_lu_dev;
326 switch (cfg->lcfg_command) {
328 struct obd_device *obd = ofd_obd(m);
329 /* For interoperability */
330 struct cfg_interop_param *ptr = NULL;
331 struct lustre_cfg *old_cfg = NULL;
334 param = lustre_cfg_string(cfg, 1);
336 CERROR("param is empty\n");
341 ptr = class_find_old_param(param, ofd_interop_param);
343 if (ptr->new_param == NULL) {
345 CWARN("For interoperability, skip this %s."
346 " It is obsolete.\n", ptr->old_param);
350 CWARN("Found old param %s, changed it to %s.\n",
351 ptr->old_param, ptr->new_param);
354 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
361 if (match_symlink_param(param)) {
362 rc = next->ld_ops->ldo_process_config(env, next, cfg);
366 rc = class_process_proc_param(PARAM_OST, obd->obd_vars, cfg,
368 if (rc > 0 || rc == -ENOSYS) {
369 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
371 /* we don't understand; pass it on */
372 rc = next->ld_ops->ldo_process_config(env, next, cfg);
376 case LCFG_SPTLRPC_CONF: {
381 /* others are passed further */
382 rc = next->ld_ops->ldo_process_config(env, next, cfg);
389 * Implementation of lu_object_operations::loo_object_init for OFD
391 * Allocate just the next object (OSD) in stack.
393 * \param[in] env execution environment
394 * \param[in] o lu_object of OFD object
395 * \param[in] conf additional configuration parameters, not used here
397 * \retval 0 if successful
398 * \retval negative value on error
400 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
401 const struct lu_object_conf *conf)
403 struct ofd_device *d = ofd_dev(o->lo_dev);
404 struct lu_device *under;
405 struct lu_object *below;
410 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
411 PFID(lu_object_fid(o)));
413 under = &d->ofd_osd->dd_lu_dev;
414 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
416 lu_object_add(o, below);
424 * Implementation of lu_object_operations::loo_object_free.
426 * Finish OFD object lifecycle and free its memory.
428 * \param[in] env execution environment
429 * \param[in] o LU object of OFD object
431 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
433 struct ofd_object *of = ofd_obj(o);
434 struct lu_object_header *h;
439 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
440 PFID(lu_object_fid(o)));
443 lu_object_header_fini(h);
444 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
449 * Implementation of lu_object_operations::loo_object_print.
451 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
452 * LU_OBJECT_DEBUG() for more details about the compound object printing.
454 * \param[in] env execution environment
455 * \param[in] cookie opaque data passed to the printer function
456 * \param[in] p printer function to use
457 * \param[in] o LU object of OFD object
459 * \retval 0 if successful
460 * \retval negative value on error
462 static int ofd_object_print(const struct lu_env *env, void *cookie,
463 lu_printer_t p, const struct lu_object *o)
465 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
468 static struct lu_object_operations ofd_obj_ops = {
469 .loo_object_init = ofd_object_init,
470 .loo_object_free = ofd_object_free,
471 .loo_object_print = ofd_object_print
475 * Implementation of lu_device_operations::lod_object_alloc.
477 * This function allocates OFD part of compound OFD-OSD object and
478 * initializes its header, because OFD is the top device in stack
480 * \param[in] env execution environment
481 * \param[in] hdr object header, NULL for OFD
482 * \param[in] d lu_device
484 * \retval allocated object if successful
485 * \retval NULL value on failed allocation
487 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
488 const struct lu_object_header *hdr,
491 struct ofd_object *of;
495 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
498 struct lu_object_header *h;
500 o = &of->ofo_obj.do_lu;
502 lu_object_header_init(h);
503 lu_object_init(o, h, d);
504 lu_object_add_top(h, o);
505 o->lo_ops = &ofd_obj_ops;
513 * Return the result of LFSCK run to the OFD.
515 * Notify OFD about result of LFSCK run. That may block the new object
516 * creation until problem is fixed by LFSCK.
518 * \param[in] env execution environment
519 * \param[in] data pointer to the OFD device
520 * \param[in] event LFSCK event type
522 * \retval 0 if successful
523 * \retval negative value on unknown event
525 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
526 enum lfsck_events event)
528 struct ofd_device *ofd = data;
529 struct obd_device *obd = ofd_obd(ofd);
532 case LE_LASTID_REBUILDING:
533 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
534 "on the device until the LAST_ID rebuilt successfully.\n",
536 down_write(&ofd->ofd_lastid_rwsem);
537 ofd->ofd_lastid_rebuilding = 1;
538 up_write(&ofd->ofd_lastid_rwsem);
540 case LE_LASTID_REBUILT: {
541 down_write(&ofd->ofd_lastid_rwsem);
542 ofd_seqs_free(env, ofd);
543 ofd->ofd_lastid_rebuilding = 0;
544 ofd->ofd_lastid_gen++;
545 up_write(&ofd->ofd_lastid_rwsem);
546 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
551 CERROR("%s: unknown lfsck event: rc = %d\n",
552 ofd_name(ofd), event);
560 * Implementation of lu_device_operations::ldo_prepare.
562 * This method is called after layer has been initialized and before it starts
563 * serving user requests. In OFD it starts lfsk check routines and initializes
566 * \param[in] env execution environment
567 * \param[in] pdev higher device in stack, NULL for OFD
568 * \param[in] dev lu_device of OFD device
570 * \retval 0 if successful
571 * \retval negative value on error
573 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
574 struct lu_device *dev)
576 struct ofd_thread_info *info;
577 struct ofd_device *ofd = ofd_dev(dev);
578 struct obd_device *obd = ofd_obd(ofd);
579 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
584 info = ofd_info_init(env, NULL);
588 /* initialize lower device */
589 rc = next->ld_ops->ldo_prepare(env, dev, next);
593 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
594 ofd_lfsck_out_notify, ofd, false);
596 CERROR("%s: failed to initialize lfsck: rc = %d\n",
601 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
602 /* The LFSCK instance is registered just now, so it must be there when
603 * register the namespace to such instance. */
604 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
606 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
607 LASSERT(obd->obd_no_conn);
608 spin_lock(&obd->obd_dev_lock);
609 obd->obd_no_conn = 0;
610 spin_unlock(&obd->obd_dev_lock);
612 if (obd->obd_recovering == 0)
613 ofd_postrecov(env, ofd);
619 * Implementation of lu_device_operations::ldo_recovery_complete.
621 * This method notifies all layers about 'recovery complete' event. That means
622 * device is in full state and consistent. An OFD calculates available grant
623 * space upon this event.
625 * \param[in] env execution environment
626 * \param[in] dev lu_device of OFD device
628 * \retval 0 if successful
629 * \retval negative value on error
631 static int ofd_recovery_complete(const struct lu_env *env,
632 struct lu_device *dev)
634 struct ofd_device *ofd = ofd_dev(dev);
635 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
636 int rc = 0, max_precreate;
641 * Grant space for object precreation on the self export.
642 * This initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
643 * is enough to create 10k objects. More space is then acquired for
644 * precreation in ofd_grant_create().
646 max_precreate = OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace / 2;
647 ofd_grant_connect(env, dev->ld_obd->obd_self_export, max_precreate,
649 rc = next->ld_ops->ldo_recovery_complete(env, next);
654 * lu_device_operations matrix for OFD device.
656 static struct lu_device_operations ofd_lu_ops = {
657 .ldo_object_alloc = ofd_object_alloc,
658 .ldo_process_config = ofd_process_config,
659 .ldo_recovery_complete = ofd_recovery_complete,
660 .ldo_prepare = ofd_prepare,
663 LPROC_SEQ_FOPS(lprocfs_nid_stats_clear);
666 * Initialize all needed procfs entries for OFD device.
668 * \param[in] ofd OFD device
670 * \retval 0 if successful
671 * \retval negative value on error
673 static int ofd_procfs_init(struct ofd_device *ofd)
675 struct obd_device *obd = ofd_obd(ofd);
676 struct proc_dir_entry *entry;
681 /* lprocfs must be setup before the ofd so state can be safely added
682 * to /proc incrementally as the ofd is setup */
683 obd->obd_vars = lprocfs_ofd_obd_vars;
684 rc = lprocfs_obd_setup(obd);
686 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
691 rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
693 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
695 GOTO(obd_cleanup, rc);
698 obd->obd_uses_nid_stats = 1;
700 entry = lprocfs_seq_register("exports", obd->obd_proc_entry, NULL,
704 CERROR("%s: error %d setting up lprocfs for %s\n",
705 obd->obd_name, rc, "exports");
706 GOTO(obd_cleanup, rc);
708 obd->obd_proc_exports_entry = entry;
710 entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
711 obd, &lprocfs_nid_stats_clear_fops);
714 CERROR("%s: add proc entry 'clear' failed: %d.\n",
716 GOTO(obd_cleanup, rc);
719 ofd_stats_counter_init(obd->obd_stats);
721 rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
722 ofd_stats_counter_init);
724 GOTO(obd_cleanup, rc);
727 lprocfs_obd_cleanup(obd);
728 lprocfs_free_obd_stats(obd);
734 * Expose OSD statistics to OFD layer.
736 * The osd interfaces to the backend file system exposes useful data
737 * such as brw_stats and read or write cache states. This same data
738 * needs to be exposed into the obdfilter (ofd) layer to maintain
739 * backwards compatibility. This function creates the symlinks in the
740 * proc layer to enable this.
742 * \param[in] ofd OFD device
744 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
746 struct obd_device *obd = ofd_obd(ofd);
747 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
749 if (obd->obd_proc_entry == NULL)
752 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
753 "../../%s/%s/brw_stats",
754 osd_obd->obd_type->typ_name, obd->obd_name);
756 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
757 "../../%s/%s/read_cache_enable",
758 osd_obd->obd_type->typ_name, obd->obd_name);
760 lprocfs_add_symlink("readcache_max_filesize",
762 "../../%s/%s/readcache_max_filesize",
763 osd_obd->obd_type->typ_name, obd->obd_name);
765 lprocfs_add_symlink("writethrough_cache_enable",
767 "../../%s/%s/writethrough_cache_enable",
768 osd_obd->obd_type->typ_name, obd->obd_name);
772 * Cleanup all procfs entries in OFD.
774 * \param[in] ofd OFD device
776 static void ofd_procfs_fini(struct ofd_device *ofd)
778 struct obd_device *obd = ofd_obd(ofd);
780 lprocfs_free_per_client_stats(obd);
781 lprocfs_obd_cleanup(obd);
782 lprocfs_free_obd_stats(obd);
783 lprocfs_job_stats_fini(obd);
787 * Stop SEQ/FID server on OFD.
789 * \param[in] env execution environment
790 * \param[in] ofd OFD device
792 * \retval 0 if successful
793 * \retval negative value on error
795 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
797 return seq_site_fini(env, &ofd->ofd_seq_site);
801 * Start SEQ/FID server on OFD.
803 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
804 * It also connects to the master server to get own FID sequence (SEQ) range
805 * to this particular OFD. Typically that happens when the OST is first
806 * formatted or in the rare case that it exhausts the local sequence range.
808 * The sequence range is allocated out to the MDTs for OST object allocations,
809 * and not directly to the clients.
811 * \param[in] env execution environment
812 * \param[in] ofd OFD device
814 * \retval 0 if successful
815 * \retval negative value on error
817 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
819 struct seq_server_site *ss = &ofd->ofd_seq_site;
820 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
821 char *obd_name = ofd_name(ofd);
825 ss = &ofd->ofd_seq_site;
826 lu->ld_site->ld_seq_site = ss;
827 ss->ss_lu = lu->ld_site;
828 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
830 OBD_ALLOC_PTR(ss->ss_server_seq);
831 if (ss->ss_server_seq == NULL)
832 GOTO(out_free, rc = -ENOMEM);
834 OBD_ALLOC(name, strlen(obd_name) + 10);
836 OBD_FREE_PTR(ss->ss_server_seq);
837 ss->ss_server_seq = NULL;
838 GOTO(out_free, rc = -ENOMEM);
841 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
842 LUSTRE_SEQ_SERVER, ss);
844 CERROR("%s : seq server init error %d\n", obd_name, rc);
847 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
849 OBD_ALLOC_PTR(ss->ss_client_seq);
850 if (ss->ss_client_seq == NULL)
851 GOTO(out_free, rc = -ENOMEM);
853 snprintf(name, strlen(obd_name) + 6, "%p-super", obd_name);
854 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
857 CERROR("%s : seq client init error %d\n", obd_name, rc);
860 OBD_FREE(name, strlen(obd_name) + 10);
863 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
867 if (ss->ss_server_seq) {
868 seq_server_fini(ss->ss_server_seq, env);
869 OBD_FREE_PTR(ss->ss_server_seq);
870 ss->ss_server_seq = NULL;
873 if (ss->ss_client_seq) {
874 seq_client_fini(ss->ss_client_seq);
875 OBD_FREE_PTR(ss->ss_client_seq);
876 ss->ss_client_seq = NULL;
880 OBD_FREE(name, strlen(obd_name) + 10);
889 * OFD request handler for OST_SET_INFO RPC.
891 * This is OFD-specific part of request handling
893 * \param[in] tsi target session environment for this request
895 * \retval 0 if successful
896 * \retval negative value on error
898 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
900 struct ptlrpc_request *req = tgt_ses_req(tsi);
901 struct ost_body *body = NULL, *repbody;
902 void *key, *val = NULL;
903 int keylen, vallen, rc = 0;
904 bool is_grant_shrink;
905 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
909 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
911 DEBUG_REQ(D_HA, req, "no set_info key");
912 RETURN(err_serious(-EFAULT));
914 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
917 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
919 DEBUG_REQ(D_HA, req, "no set_info val");
920 RETURN(err_serious(-EFAULT));
922 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
925 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
927 /* In this case the value is actually an RMF_OST_BODY, so we
928 * transmutate the type of this PTLRPC */
929 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
931 rc = req_capsule_server_pack(tsi->tsi_pill);
935 if (is_grant_shrink) {
936 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
938 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
941 /** handle grant shrink, similar to a read request */
942 ofd_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
944 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
946 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
948 } else if (KEY_IS(KEY_CAPA_KEY)) {
949 rc = ofd_update_capa_key(ofd, val);
950 } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
951 rc = tgt_adapt_sptlrpc_conf(tsi->tsi_tgt, 0);
953 CERROR("%s: Unsupported key %s\n",
954 tgt_name(tsi->tsi_tgt), (char *)key);
957 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
964 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
966 * This function returns a list of extents which describes how a file's
967 * blocks are laid out on the disk.
969 * \param[in] env execution environment
970 * \param[in] ofd OFD device
971 * \param[in] fid FID of object
972 * \param[in] fiemap fiemap structure to fill with data
974 * \retval 0 if \a fiemap is filled with data successfully
975 * \retval negative value on error
977 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
978 struct lu_fid *fid, struct ll_user_fiemap *fiemap)
980 struct ofd_object *fo;
983 fo = ofd_object_find(env, ofd, fid);
985 CERROR("%s: error finding object "DFID"\n",
986 ofd_name(ofd), PFID(fid));
990 ofd_read_lock(env, fo);
991 if (ofd_object_exists(fo))
992 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
995 ofd_read_unlock(env, fo);
996 ofd_object_put(env, fo);
1000 struct locked_region {
1001 struct list_head list;
1002 struct lustre_handle lh;
1006 * Lock single extent and save lock handle in the list.
1008 * This is supplemental function for lock_zero_regions(). It allocates
1009 * new locked_region structure and locks it with extent lock, then adds
1010 * it to the list of all such regions.
1012 * \param[in] ns LDLM namespace
1013 * \param[in] res_id resource ID
1014 * \param[in] begin start of region
1015 * \param[in] end end of region
1016 * \param[in] locked list head of regions list
1018 * \retval 0 if successful locking
1019 * \retval negative value on error
1021 static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1022 unsigned long long begin, unsigned long long end,
1023 struct list_head *locked)
1025 struct locked_region *region = NULL;
1029 LASSERT(begin <= end);
1030 OBD_ALLOC_PTR(region);
1034 rc = tgt_extent_lock(ns, res_id, begin, end, ®ion->lh,
1039 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
1041 list_add(®ion->list, locked);
1047 * Lock the sparse areas of given resource.
1049 * The locking of sparse areas will cause dirty data to be flushed back from
1050 * clients. This is used when getting the FIEMAP of an object to make sure
1051 * there is no unaccounted cached data on clients.
1053 * This function goes through \a fiemap list of extents and locks only sparse
1054 * areas between extents.
1056 * \param[in] ns LDLM namespace
1057 * \param[in] res_id resource ID
1058 * \param[in] fiemap file extents mapping on disk
1059 * \param[in] locked list head of regions list
1061 * \retval 0 if successful
1062 * \retval negative value on error
1064 static int lock_zero_regions(struct ldlm_namespace *ns,
1065 struct ldlm_res_id *res_id,
1066 struct ll_user_fiemap *fiemap,
1067 struct list_head *locked)
1069 __u64 begin = fiemap->fm_start;
1072 struct ll_fiemap_extent *fiemap_start = fiemap->fm_extents;
1076 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1077 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1078 if (fiemap_start[i].fe_logical > begin) {
1079 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1080 begin, fiemap_start[i].fe_logical);
1081 rc = lock_region(ns, res_id, begin,
1082 fiemap_start[i].fe_logical, locked);
1087 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1090 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1091 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1092 begin, fiemap->fm_start + fiemap->fm_length);
1093 rc = lock_region(ns, res_id, begin,
1094 fiemap->fm_start + fiemap->fm_length, locked);
1101 * Unlock all previously locked sparse areas for given resource.
1103 * This function goes through list of locked regions, unlocking and freeing
1106 * \param[in] ns LDLM namespace
1107 * \param[in] locked list head of regions list
1110 unlock_zero_regions(struct ldlm_namespace *ns, struct list_head *locked)
1112 struct locked_region *entry, *temp;
1114 list_for_each_entry_safe(entry, temp, locked, list) {
1115 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
1116 tgt_extent_unlock(&entry->lh, LCK_PR);
1117 list_del(&entry->list);
1118 OBD_FREE_PTR(entry);
1123 * OFD request handler for OST_GET_INFO RPC.
1125 * This is OFD-specific part of request handling. The OFD-specific keys are:
1126 * - KEY_LAST_ID (obsolete)
1130 * This function reads needed data from storage and fills reply with it.
1132 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1133 * and is kept for compatibility.
1135 * \param[in] tsi target session environment for this request
1137 * \retval 0 if successful
1138 * \retval negative value on error
1140 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1142 struct obd_export *exp = tsi->tsi_exp;
1143 struct ofd_device *ofd = ofd_exp(exp);
1144 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1147 int replylen, rc = 0;
1151 /* this common part for get_info rpc */
1152 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1154 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1155 RETURN(err_serious(-EPROTO));
1157 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1160 if (KEY_IS(KEY_LAST_ID)) {
1162 struct ofd_seq *oseq;
1164 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1165 rc = req_capsule_server_pack(tsi->tsi_pill);
1167 RETURN(err_serious(rc));
1169 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1171 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1172 (obd_seq)exp->exp_filter_data.fed_group);
1176 *last_id = ofd_seq_last_oid(oseq);
1177 ofd_seq_put(tsi->tsi_env, oseq);
1178 } else if (KEY_IS(KEY_FIEMAP)) {
1179 struct ll_fiemap_info_key *fm_key;
1180 struct ll_user_fiemap *fiemap;
1183 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1185 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1186 rc = tgt_validate_obdo(tsi, &fm_key->oa);
1188 RETURN(err_serious(rc));
1190 fid = &fm_key->oa.o_oi.oi_fid;
1192 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1194 replylen = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1195 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1196 RCL_SERVER, replylen);
1198 rc = req_capsule_server_pack(tsi->tsi_pill);
1200 RETURN(err_serious(rc));
1202 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1206 *fiemap = fm_key->fiemap;
1207 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1209 /* LU-3219: Lock the sparse areas to make sure dirty
1210 * flushed back from client, then call fiemap again. */
1211 if (fm_key->oa.o_valid & OBD_MD_FLFLAGS &&
1212 fm_key->oa.o_flags & OBD_FL_SRVLOCK) {
1213 struct list_head locked;
1215 INIT_LIST_HEAD(&locked);
1216 ost_fid_build_resid(fid, &fti->fti_resid);
1217 rc = lock_zero_regions(ofd->ofd_namespace,
1218 &fti->fti_resid, fiemap,
1220 if (rc == 0 && !list_empty(&locked)) {
1221 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1223 unlock_zero_regions(ofd->ofd_namespace,
1227 } else if (KEY_IS(KEY_LAST_FID)) {
1228 struct ofd_device *ofd = ofd_exp(exp);
1229 struct ofd_seq *oseq;
1233 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1234 rc = req_capsule_server_pack(tsi->tsi_pill);
1236 RETURN(err_serious(rc));
1238 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1240 RETURN(err_serious(-EPROTO));
1242 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1244 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1248 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1249 ostid_seq(&fti->fti_ostid));
1251 RETURN(PTR_ERR(oseq));
1253 rc = ostid_to_fid(fid, &oseq->os_oi,
1254 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1258 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1261 ofd_seq_put(tsi->tsi_env, oseq);
1263 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1267 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1274 * OFD request handler for OST_GETATTR RPC.
1276 * This is OFD-specific part of request handling. It finds the OFD object
1277 * by its FID, gets attributes from storage and packs result to the reply.
1279 * \param[in] tsi target session environment for this request
1281 * \retval 0 if successful
1282 * \retval negative value on error
1284 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1286 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1287 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1288 struct ost_body *repbody;
1289 struct lustre_handle lh = { 0 };
1290 struct ofd_object *fo;
1292 ldlm_mode_t lock_mode = LCK_PR;
1297 LASSERT(tsi->tsi_ost_body != NULL);
1299 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1300 if (repbody == NULL)
1303 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1304 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1306 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1307 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1310 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1313 rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace,
1314 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1320 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1322 GOTO(out, rc = PTR_ERR(fo));
1324 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1328 obdo_from_la(&repbody->oa, &fti->fti_attr,
1329 OFD_VALID_FLAGS | LA_UID | LA_GID);
1330 tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1332 /* Store object version in reply */
1333 curr_version = dt_version_get(tsi->tsi_env,
1334 ofd_object_child(fo));
1335 if ((__s64)curr_version != -EOPNOTSUPP) {
1336 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1337 repbody->oa.o_data_version = curr_version;
1341 ofd_object_put(tsi->tsi_env, fo);
1344 tgt_extent_unlock(&lh, lock_mode);
1346 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1349 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1350 repbody->oa.o_flags = OBD_FL_FLUSH;
1356 * OFD request handler for OST_SETATTR RPC.
1358 * This is OFD-specific part of request handling. It finds the OFD object
1359 * by its FID, sets attributes from request and packs result to the reply.
1361 * \param[in] tsi target session environment for this request
1363 * \retval 0 if successful
1364 * \retval negative value on error
1366 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1368 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1369 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1370 struct ost_body *body = tsi->tsi_ost_body;
1371 struct ost_body *repbody;
1372 struct ldlm_resource *res;
1373 struct ofd_object *fo;
1374 struct filter_fid *ff = NULL;
1379 LASSERT(body != NULL);
1381 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1382 if (repbody == NULL)
1385 repbody->oa.o_oi = body->oa.o_oi;
1386 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1388 /* This would be very bad - accidentally truncating a file when
1389 * changing the time or similar - bug 12203. */
1390 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1391 body->oa.o_size != OBD_OBJECT_EOF) {
1392 static char mdsinum[48];
1394 if (body->oa.o_valid & OBD_MD_FLFID)
1395 snprintf(mdsinum, sizeof(mdsinum) - 1,
1396 "of parent "DFID, body->oa.o_parent_seq,
1397 body->oa.o_parent_oid, 0);
1401 CERROR("%s: setattr from %s is trying to truncate object "DFID
1402 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1403 PFID(&tsi->tsi_fid), mdsinum);
1407 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1409 GOTO(out, rc = PTR_ERR(fo));
1411 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1412 fti->fti_attr.la_valid &= ~LA_TYPE;
1414 if (body->oa.o_valid & OBD_MD_FLFID) {
1415 ff = &fti->fti_mds_fid;
1416 ofd_prepare_fidea(ff, &body->oa);
1419 /* setting objects attributes (including owner/group) */
1420 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff);
1424 obdo_from_la(&repbody->oa, &fti->fti_attr,
1425 OFD_VALID_FLAGS | LA_UID | LA_GID);
1426 tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1428 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1432 ofd_object_put(tsi->tsi_env, fo);
1435 /* we do not call this before to avoid lu_object_find() in
1436 * ->lvbo_update() holding another reference on the object.
1437 * otherwise concurrent destroy can make the object unavailable
1438 * for 2nd lu_object_find() waiting for the first reference
1439 * to go... deadlock! */
1440 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1441 &tsi->tsi_resid, LDLM_EXTENT, 0);
1443 ldlm_res_lvbo_update(res, NULL, 0);
1444 ldlm_resource_putref(res);
1451 * Destroy OST orphans.
1453 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1454 * set then we must destroy possible orphaned objects.
1456 * \param[in] env execution environment
1457 * \param[in] exp OBD export
1458 * \param[in] ofd OFD device
1459 * \param[in] oa obdo structure for reply
1461 * \retval 0 if successful
1462 * \retval negative value on error
1464 static int ofd_orphans_destroy(const struct lu_env *env,
1465 struct obd_export *exp,
1466 struct ofd_device *ofd, struct obdo *oa)
1468 struct ofd_thread_info *info = ofd_info(env);
1469 struct lu_fid *fid = &info->fti_fid;
1470 struct ost_id *oi = &oa->o_oi;
1471 struct ofd_seq *oseq;
1472 obd_seq seq = ostid_seq(oi);
1473 obd_id end_id = ostid_id(oi);
1481 oseq = ofd_seq_get(ofd, seq);
1483 CERROR("%s: Can not find seq for "DOSTID"\n",
1484 ofd_name(ofd), POSTID(oi));
1489 last = ofd_seq_last_oid(oseq);
1492 LASSERT(exp != NULL);
1493 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1495 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1498 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1499 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1501 while (oid > end_id) {
1502 rc = fid_set_id(fid, oid);
1503 if (unlikely(rc != 0))
1506 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1507 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1508 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1509 /* this is pretty fatal... */
1510 CEMERG("%s: error destroying precreated id "
1512 ofd_name(ofd), PFID(fid), rc);
1516 ofd_seq_last_oid_set(oseq, oid);
1517 /* update last_id on disk periodically so that if we
1518 * restart * we don't need to re-scan all of the just
1519 * deleted objects. */
1520 if ((oid & 511) == 0)
1521 ofd_seq_last_oid_write(env, ofd, oseq);
1525 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1526 ofd_name(ofd), seq, oid);
1530 ofd_seq_last_oid_set(oseq, oid);
1531 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1533 /* don't reuse orphan object, return last used objid */
1534 ostid_set_id(oi, last);
1541 ofd_seq_put(env, oseq);
1546 * OFD request handler for OST_CREATE RPC.
1548 * This is OFD-specific part of request handling. Its main purpose is to
1549 * create new data objects on OST, but it also used to destroy orphans.
1551 * \param[in] tsi target session environment for this request
1553 * \retval 0 if successful
1554 * \retval negative value on error
1556 static int ofd_create_hdl(struct tgt_session_info *tsi)
1558 struct ptlrpc_request *req = tgt_ses_req(tsi);
1559 struct ost_body *repbody;
1560 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1561 struct obdo *rep_oa;
1562 struct obd_export *exp = tsi->tsi_exp;
1563 struct ofd_device *ofd = ofd_exp(exp);
1564 obd_seq seq = ostid_seq(&oa->o_oi);
1565 obd_id oid = ostid_id(&oa->o_oi);
1566 struct ofd_seq *oseq;
1572 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1575 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1576 if (repbody == NULL)
1579 down_read(&ofd->ofd_lastid_rwsem);
1580 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1581 * we may do that in the future.
1582 * Return -ENOSPC until the LAST_ID rebuilt. */
1583 if (unlikely(ofd->ofd_lastid_rebuilding))
1584 GOTO(out_sem, rc = -ENOSPC);
1586 rep_oa = &repbody->oa;
1587 rep_oa->o_oi = oa->o_oi;
1589 LASSERT(seq >= FID_SEQ_OST_MDT0);
1590 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1592 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1594 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1596 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1597 ofd_name(ofd), seq, PTR_ERR(oseq));
1598 GOTO(out_sem, rc = -EINVAL);
1601 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1602 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1603 if (!ofd_obd(ofd)->obd_recovering ||
1604 oid > ofd_seq_last_oid(oseq)) {
1605 CERROR("%s: recreate objid "DOSTID" > last id "LPU64
1606 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1607 ofd_seq_last_oid(oseq));
1608 GOTO(out_nolock, rc = -EINVAL);
1610 /* Do nothing here, we re-create objects during recovery
1611 * upon write replay, see ofd_preprw_write() */
1612 GOTO(out_nolock, rc = 0);
1614 /* former ofd_handle_precreate */
1615 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1616 (oa->o_flags & OBD_FL_DELORPHAN)) {
1617 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1619 /* destroy orphans */
1620 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1621 exp->exp_conn_cnt) {
1622 CERROR("%s: dropping old orphan cleanup request\n",
1624 GOTO(out_nolock, rc = 0);
1626 /* This causes inflight precreates to abort and drop lock */
1627 oseq->os_destroys_in_progress = 1;
1628 mutex_lock(&oseq->os_create_lock);
1629 if (!oseq->os_destroys_in_progress) {
1630 CERROR("%s:["LPU64"] destroys_in_progress already"
1631 " cleared\n", ofd_name(ofd), seq);
1632 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1635 diff = oid - ofd_seq_last_oid(oseq);
1636 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1637 ofd_seq_last_oid(oseq), diff);
1638 if (-diff > OST_MAX_PRECREATE) {
1639 /* FIXME: should reset precreate_next_id on MDS */
1641 } else if (diff < 0) {
1642 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1644 oseq->os_destroys_in_progress = 0;
1646 /* XXX: Used by MDS for the first time! */
1647 oseq->os_destroys_in_progress = 0;
1650 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1651 ofd->ofd_lastid_gen)) {
1652 /* Keep the export ref so we can send the reply. */
1653 ofd_obd_disconnect(class_export_get(exp));
1654 GOTO(out_nolock, rc = -ENOTCONN);
1657 mutex_lock(&oseq->os_create_lock);
1658 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1659 exp->exp_conn_cnt) {
1660 CERROR("%s: dropping old precreate request\n",
1664 /* only precreate if seq is 0, IDIF or normal and also o_id
1665 * must be specfied */
1666 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1667 !fid_seq_is_idif(seq)) || oid == 0) {
1668 diff = 1; /* shouldn't we create this right now? */
1670 diff = oid - ofd_seq_last_oid(oseq);
1671 /* Do sync create if the seq is about to used up */
1672 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1673 if (unlikely(oid >= IDIF_MAX_OID - 1))
1675 } else if (fid_seq_is_norm(seq)) {
1677 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1680 CERROR("%s : invalid o_seq "DOSTID"\n",
1681 ofd_name(ofd), POSTID(&oa->o_oi));
1682 GOTO(out, rc = -EINVAL);
1687 cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
1692 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1693 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1694 /* don't enforce grant during orphan recovery */
1695 rc = ofd_grant_create(tsi->tsi_env,
1696 ofd_obd(ofd)->obd_self_export,
1699 CDEBUG(D_HA, "%s: failed to acquire grant "
1700 "space for precreate (%d): rc = %d\n",
1701 ofd_name(ofd), diff, rc);
1706 /* This can happen if a new OST is formatted and installed
1707 * in place of an old one at the same index. Instead of
1708 * precreating potentially millions of deleted old objects
1709 * (possibly filling the OST), only precreate the last batch.
1710 * LFSCK will eventually clean up any orphans. LU-14 */
1711 if (diff > 5 * OST_MAX_PRECREATE) {
1712 diff = OST_MAX_PRECREATE / 2;
1713 LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %u "
1714 "larger than the LAST_ID "DOSTID", only "
1715 "precreating the last %u objects.\n",
1716 ofd_name(ofd), POSTID(&oa->o_oi),
1717 5 * OST_MAX_PRECREATE,
1718 POSTID(&oseq->os_oi), diff);
1719 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1723 next_id = ofd_seq_last_oid(oseq) + 1;
1724 count = ofd_precreate_batch(ofd, diff);
1726 CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1727 " at "LPU64"\n", ofd_name(ofd),
1728 count, seq, next_id);
1730 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1731 && cfs_time_after(jiffies, enough_time)) {
1732 CDEBUG(D_HA, "%s: Slow creates, %d/%d objects"
1733 " created at a rate of %d/s\n",
1734 ofd_name(ofd), created, diff + created,
1735 created / DISK_TIMEOUT);
1739 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1740 oseq, count, sync_trans);
1744 } else if (rc < 0) {
1750 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1751 LCONSOLE_WARN("%s: can't create the same count of"
1752 " objects when replaying the request"
1753 " (diff is %d). see LU-4621\n",
1754 ofd_name(ofd), diff);
1757 /* some objects got created, we can return
1758 * them, even if last creation failed */
1761 CERROR("%s: unable to precreate: rc = %d\n",
1764 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1765 !(oa->o_flags & OBD_FL_DELORPHAN))
1766 ofd_grant_commit(tsi->tsi_env,
1767 ofd_obd(ofd)->obd_self_export, rc);
1769 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1772 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1775 mutex_unlock(&oseq->os_create_lock);
1778 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1779 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1780 struct lu_fid *fid = &info->fti_fid;
1782 /* For compatible purpose, it needs to convert back to
1783 * OST ID before put it on wire. */
1784 *fid = rep_oa->o_oi.oi_fid;
1785 fid_to_ostid(fid, &rep_oa->o_oi);
1787 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1789 ofd_seq_put(tsi->tsi_env, oseq);
1792 up_read(&ofd->ofd_lastid_rwsem);
1797 * OFD request handler for OST_DESTROY RPC.
1799 * This is OFD-specific part of request handling. It destroys data objects
1800 * related to destroyed object on MDT.
1802 * \param[in] tsi target session environment for this request
1804 * \retval 0 if successful
1805 * \retval negative value on error
1807 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1809 const struct ost_body *body = tsi->tsi_ost_body;
1810 struct ost_body *repbody;
1811 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1812 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1813 struct lu_fid *fid = &fti->fti_fid;
1820 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1823 /* This is old case for clients before Lustre 2.4 */
1824 /* If there's a DLM request, cancel the locks mentioned in it */
1825 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1827 struct ldlm_request *dlm;
1829 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1832 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1835 *fid = body->oa.o_oi.oi_fid;
1836 oid = ostid_id(&body->oa.o_oi);
1839 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1841 /* check that o_misc makes sense */
1842 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1843 count = body->oa.o_misc;
1845 count = 1; /* default case - single destroy */
1847 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1848 POSTID(&body->oa.o_oi), count);
1853 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1854 if (lrc == -ENOENT) {
1856 "%s: destroying non-existent object "DFID"\n",
1857 ofd_name(ofd), PFID(fid));
1858 /* rewrite rc with -ENOENT only if it is 0 */
1861 } else if (lrc != 0) {
1862 CERROR("%s: error destroying object "DFID": %d\n",
1863 ofd_name(ofd), PFID(fid), lrc);
1869 lrc = fid_set_id(fid, oid);
1870 if (unlikely(lrc != 0 && count > 0))
1871 GOTO(out, rc = lrc);
1874 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1880 fid_to_ostid(fid, &repbody->oa.o_oi);
1885 * OFD request handler for OST_STATFS RPC.
1887 * This function gets statfs data from storage as part of request
1890 * \param[in] tsi target session environment for this request
1892 * \retval 0 if successful
1893 * \retval negative value on error
1895 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1897 struct obd_statfs *osfs;
1902 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1904 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1905 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
1907 CERROR("%s: statfs failed: rc = %d\n",
1908 tgt_name(tsi->tsi_tgt), rc);
1910 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1913 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1920 * OFD request handler for OST_SYNC RPC.
1922 * Sync object data or all filesystem data to the disk and pack the
1925 * \param[in] tsi target session environment for this request
1927 * \retval 0 if successful
1928 * \retval negative value on error
1930 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1932 struct ost_body *body = tsi->tsi_ost_body;
1933 struct ost_body *repbody;
1934 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1935 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1936 struct ofd_object *fo = NULL;
1941 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1943 /* if no objid is specified, it means "sync whole filesystem" */
1944 if (!fid_is_zero(&tsi->tsi_fid)) {
1945 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1947 RETURN(PTR_ERR(fo));
1950 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1951 fo != NULL ? ofd_object_child(fo) : NULL,
1952 repbody->oa.o_size, repbody->oa.o_blocks);
1956 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1961 repbody->oa.o_oi = body->oa.o_oi;
1962 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1964 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1966 obdo_from_la(&repbody->oa, &fti->fti_attr,
1969 /* don't return rc from getattr */
1974 ofd_object_put(tsi->tsi_env, fo);
1979 * OFD request handler for OST_PUNCH RPC.
1981 * This is part of request processing. Validate request fields,
1982 * punch (truncate) the given OFD object and pack reply.
1984 * \param[in] tsi target session environment for this request
1986 * \retval 0 if successful
1987 * \retval negative value on error
1989 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1991 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1992 struct ost_body *repbody;
1993 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1994 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1995 struct ldlm_resource *res;
1996 struct ofd_object *fo;
1997 struct filter_fid *ff = NULL;
1999 struct lustre_handle lh = { 0, };
2006 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
2007 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
2009 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2010 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2011 RETURN(err_serious(-EPROTO));
2013 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2014 if (repbody == NULL)
2015 RETURN(err_serious(-ENOMEM));
2017 /* punch start,end are passed in o_size,o_blocks throught wire */
2021 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2024 /* standard truncate optimization: if file body is completely
2025 * destroyed, don't send data back to the server. */
2027 flags |= LDLM_FL_AST_DISCARD_DATA;
2029 repbody->oa.o_oi = oa->o_oi;
2030 repbody->oa.o_valid = OBD_MD_FLID;
2032 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2033 oa->o_flags & OBD_FL_SRVLOCK;
2036 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
2042 CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
2043 ", start = "LPD64", end = "LPD64"\n", PFID(&tsi->tsi_fid),
2044 oa->o_valid, start, end);
2046 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2049 GOTO(out, rc = PTR_ERR(fo));
2051 la_from_obdo(&info->fti_attr, oa,
2052 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2053 info->fti_attr.la_size = start;
2054 info->fti_attr.la_valid |= LA_SIZE;
2056 if (oa->o_valid & OBD_MD_FLFID) {
2057 ff = &info->fti_mds_fid;
2058 ofd_prepare_fidea(ff, oa);
2061 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2062 ff, (struct obdo *)oa);
2066 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2070 ofd_object_put(tsi->tsi_env, fo);
2073 tgt_extent_unlock(&lh, LCK_PW);
2075 /* we do not call this before to avoid lu_object_find() in
2076 * ->lvbo_update() holding another reference on the object.
2077 * otherwise concurrent destroy can make the object unavailable
2078 * for 2nd lu_object_find() waiting for the first reference
2079 * to go... deadlock! */
2080 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2083 ldlm_res_lvbo_update(res, NULL, 0);
2084 ldlm_resource_putref(res);
2091 * OFD request handler for OST_QUOTACTL RPC.
2093 * This is part of request processing to validate incoming request fields,
2094 * get the requested data from OSD and pack reply.
2096 * \param[in] tsi target session environment for this request
2098 * \retval 0 if successful
2099 * \retval negative value on error
2101 static int ofd_quotactl(struct tgt_session_info *tsi)
2103 struct obd_quotactl *oqctl, *repoqc;
2104 struct lu_nodemap *nodemap =
2105 tsi->tsi_exp->exp_target_data.ted_nodemap;
2111 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2113 RETURN(err_serious(-EPROTO));
2115 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2117 RETURN(err_serious(-ENOMEM));
2119 /* report success for quota on/off for interoperability with current MDT
2121 if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
2127 if (oqctl->qc_type == USRQUOTA)
2128 id = nodemap_map_id(nodemap, NODEMAP_UID,
2129 NODEMAP_CLIENT_TO_FS,
2131 else if (oqctl->qc_type == GRPQUOTA)
2132 id = nodemap_map_id(nodemap, NODEMAP_GID,
2133 NODEMAP_CLIENT_TO_FS,
2136 if (repoqc->qc_id != id)
2137 swap(repoqc->qc_id, id);
2139 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2141 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2144 if (repoqc->qc_id != id)
2145 swap(repoqc->qc_id, id);
2151 * Calculate the amount of time for lock prolongation.
2153 * This is helper for ofd_prolong_extent_locks() function to get
2154 * the timeout extra time.
2156 * \param[in] req current request
2158 * \retval amount of time to extend the timeout with
2160 static inline int prolong_timeout(struct ptlrpc_request *req,
2161 struct ldlm_lock *lock)
2163 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2166 return obd_timeout / 2;
2168 /* We are in the middle of the process - BL AST is sent, CANCEL
2169 is ahead. Take half of AT + IO process time. */
2170 return at_est2timeout(at_get(&svcpt->scp_at_estimate)) +
2171 (ldlm_bl_timeout(lock) >> 1);
2175 * Prolong single lock timeout.
2177 * This is supplemental function to the ofd_prolong_locks(). It prolongs
2180 * \param[in] tsi target session environment for this request
2181 * \param[in] lock LDLM lock to prolong
2182 * \param[in] extent related extent
2183 * \param[in] timeout timeout value to add
2185 * \retval 0 if lock is not suitable for prolongation
2186 * \retval 1 if lock was prolonged successfully
2188 static int ofd_prolong_one_lock(struct tgt_session_info *tsi,
2189 struct ldlm_lock *lock,
2190 struct ldlm_extent *extent)
2192 int timeout = prolong_timeout(tgt_ses_req(tsi), lock);
2194 if (lock->l_flags & LDLM_FL_DESTROYED) /* lock already cancelled */
2197 /* XXX: never try to grab resource lock here because we're inside
2198 * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
2199 * res lock and then exp_bl_list_lock. */
2201 if (!(lock->l_flags & LDLM_FL_AST_SENT))
2202 /* ignore locks not being cancelled */
2205 LDLM_DEBUG(lock, "refreshed for req x"LPU64" ext("LPU64"->"LPU64") "
2206 "to %ds.\n", tgt_ses_req(tsi)->rq_xid, extent->start,
2207 extent->end, timeout);
2209 /* OK. this is a possible lock the user holds doing I/O
2210 * let's refresh eviction timer for it */
2211 ldlm_refresh_waiting_lock(lock, timeout);
2216 * Prolong lock timeout for the given extent.
2218 * This function finds all locks related with incoming request and
2219 * prolongs their timeout.
2221 * If a client is holding a lock for a long time while it sends
2222 * read or write RPCs to the OST for the object under this lock,
2223 * then we don't want the OST to evict the client. Otherwise,
2224 * if the network or disk is very busy then the client may not
2225 * be able to make any progress to clear out dirty pages under
2226 * the lock and the application will fail.
2228 * Every time a Bulk Read/Write (BRW) request arrives for the object
2229 * covered by the lock, extend the timeout on that lock. The RPC should
2230 * contain a lock handle for the lock it is using, but this
2231 * isn't handled correctly by all client versions, and the
2232 * request may cover multiple locks.
2234 * \param[in] tsi target session environment for this request
2235 * \param[in] start start of extent
2236 * \param[in] end end of extent
2238 * \retval number of prolonged locks
2240 static int ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2241 __u64 start, __u64 end)
2243 struct obd_export *exp = tsi->tsi_exp;
2244 struct obdo *oa = &tsi->tsi_ost_body->oa;
2245 struct ldlm_extent extent = {
2249 struct ldlm_lock *lock;
2254 if (oa->o_valid & OBD_MD_FLHANDLE) {
2255 /* mostly a request should be covered by only one lock, try
2257 lock = ldlm_handle2lock(&oa->o_handle);
2259 /* Fast path to check if the lock covers the whole IO
2260 * region exclusively. */
2261 if (lock->l_granted_mode == LCK_PW &&
2262 ldlm_extent_contain(&lock->l_policy_data.l_extent,
2265 LASSERT(lock->l_export == exp);
2266 lock_count = ofd_prolong_one_lock(tsi, lock,
2268 LDLM_LOCK_PUT(lock);
2271 LDLM_LOCK_PUT(lock);
2275 spin_lock_bh(&exp->exp_bl_list_lock);
2276 list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
2277 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
2278 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
2280 if (!ldlm_res_eq(&tsi->tsi_resid, &lock->l_resource->lr_name))
2283 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
2287 lock_count += ofd_prolong_one_lock(tsi, lock, &extent);
2289 spin_unlock_bh(&exp->exp_bl_list_lock);
2295 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2297 * Determine if \a lock and the lock from request \a req are equivalent
2298 * by comparing their resource names, modes, and extents.
2300 * It is used to give priority to read and write RPCs being done
2301 * under this lock so that the client can drop the contended
2302 * lock more quickly and let other clients use it. This improves
2303 * overall performance in the case where the first client gets a
2304 * very large lock extent that prevents other clients from
2305 * submitting their writes.
2307 * \param[in] req ptlrpc_request being processed
2308 * \param[in] lock contended lock to match
2310 * \retval 1 if lock is matched
2311 * \retval 0 otherwise
2313 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2314 struct ldlm_lock *lock)
2316 struct niobuf_remote *rnb;
2317 struct obd_ioobj *ioo;
2319 struct ldlm_extent ext;
2320 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2324 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2325 LASSERT(ioo != NULL);
2327 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2328 LASSERT(rnb != NULL);
2330 ext.start = rnb->rnb_offset;
2331 rnb += ioo->ioo_bufcnt - 1;
2332 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2334 LASSERT(lock->l_resource != NULL);
2335 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2338 /* a bulk write can only hold a reference on a PW extent lock */
2340 if (opc == OST_READ)
2341 /* whereas a bulk read can be protected by either a PR or PW
2345 if (!(lock->l_granted_mode & mode))
2348 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2352 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2354 * Check for whether the given PTLRPC request (\a req) is blocking
2355 * an LDLM lock cancel.
2357 * \param[in] req the incoming request
2359 * \retval 1 if \a req is blocking an LDLM lock cancel
2360 * \retval 0 if it is not
2362 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2364 struct tgt_session_info *tsi;
2365 struct obd_ioobj *ioo;
2366 struct niobuf_remote *rnb;
2372 /* Don't use tgt_ses_info() to get session info, because lock_match()
2373 * can be called while request has no processing thread yet. */
2374 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2377 * Use LASSERT below because malformed RPCs should have
2378 * been filtered out in tgt_hpreq_handler().
2380 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2381 LASSERT(ioo != NULL);
2383 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2384 LASSERT(rnb != NULL);
2385 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2387 start = rnb->rnb_offset;
2388 rnb += ioo->ioo_bufcnt - 1;
2389 end = rnb->rnb_offset + rnb->rnb_len - 1;
2391 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
2392 " ("LPU64"->"LPU64")\n",
2393 tgt_name(tsi->tsi_tgt), current->comm,
2394 PFID(&tsi->tsi_fid), start, end);
2396 lock_count = ofd_prolong_extent_locks(tsi, start, end);
2398 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2399 tgt_name(tsi->tsi_tgt), lock_count, req);
2401 RETURN(lock_count > 0);
2405 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2407 * Called after the request has been handled. It refreshes lock timeout again
2408 * so that client has more time to send lock cancel RPC.
2410 * \param[in] req request which is being processed.
2412 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2414 ofd_rw_hpreq_check(req);
2418 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2420 * This function checks if the given lock is the same by its resname, mode
2421 * and extent as one taken from the request.
2422 * It is used to give priority to punch/truncate RPCs that might lead to
2423 * the fastest release of that lock when a lock is contended.
2425 * \param[in] req ptlrpc_request being processed
2426 * \param[in] lock contended lock to match
2428 * \retval 1 if lock is matched
2429 * \retval 0 otherwise
2431 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2432 struct ldlm_lock *lock)
2434 struct tgt_session_info *tsi;
2436 /* Don't use tgt_ses_info() to get session info, because lock_match()
2437 * can be called while request has no processing thread yet. */
2438 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2441 * Use LASSERT below because malformed RPCs should have
2442 * been filtered out in tgt_hpreq_handler().
2444 LASSERT(tsi->tsi_ost_body != NULL);
2445 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2446 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2453 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2455 * High-priority queue request check for whether the given punch request
2456 * (\a req) is blocking an LDLM lock cancel.
2458 * \param[in] req the incoming request
2460 * \retval 1 if \a req is blocking an LDLM lock cancel
2461 * \retval 0 if it is not
2463 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2465 struct tgt_session_info *tsi;
2471 /* Don't use tgt_ses_info() to get session info, because lock_match()
2472 * can be called while request has no processing thread yet. */
2473 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2474 LASSERT(tsi != NULL);
2475 oa = &tsi->tsi_ost_body->oa;
2477 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2478 oa->o_flags & OBD_FL_SRVLOCK));
2481 "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
2482 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2483 tsi->tsi_resid.name[1], oa->o_size, oa->o_blocks);
2485 lock_count = ofd_prolong_extent_locks(tsi, oa->o_size, oa->o_blocks);
2487 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2488 tgt_name(tsi->tsi_tgt), lock_count, req);
2490 RETURN(lock_count > 0);
2494 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2496 * Called after the request has been handled. It refreshes lock timeout again
2497 * so that client has more time to send lock cancel RPC.
2499 * \param[in] req request which is being processed.
2501 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2503 ofd_punch_hpreq_check(req);
2506 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2507 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2508 .hpreq_check = ofd_rw_hpreq_check,
2509 .hpreq_fini = ofd_rw_hpreq_fini
2512 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2513 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2514 .hpreq_check = ofd_punch_hpreq_check,
2515 .hpreq_fini = ofd_punch_hpreq_fini
2519 * Assign high priority operations to an IO request.
2521 * Check if the incoming request is a candidate for
2522 * high-priority processing. If it is, assign it a high
2523 * priority operations table.
2525 * \param[in] tsi target session environment for this request
2527 static void ofd_hp_brw(struct tgt_session_info *tsi)
2529 struct niobuf_remote *rnb;
2530 struct obd_ioobj *ioo;
2534 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2535 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2536 if (ioo->ioo_bufcnt > 0) {
2537 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2538 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2540 /* no high priority if server lock is needed */
2541 if (rnb->rnb_flags & OBD_BRW_SRVLOCK)
2544 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2548 * Assign high priority operations to an punch request.
2550 * Check if the incoming request is a candidate for
2551 * high-priority processing. If it is, assign it a high
2552 * priority operations table.
2554 * \param[in] tsi target session environment for this request
2556 static void ofd_hp_punch(struct tgt_session_info *tsi)
2558 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2559 /* no high-priority if server lock is needed */
2560 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2561 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK)
2563 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2566 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2567 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2568 #define OST_BRW_READ OST_READ
2569 #define OST_BRW_WRITE OST_WRITE
2572 * Table of OFD-specific request handlers
2574 * This table contains all opcodes accepted by OFD and
2575 * specifies handlers for them. The tgt_request_handler()
2576 * uses such table from each target to process incoming
2579 static struct tgt_handler ofd_tgt_handlers[] = {
2580 TGT_RPC_HANDLER(OST_FIRST_OPC,
2581 0, OST_CONNECT, tgt_connect,
2582 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2583 TGT_RPC_HANDLER(OST_FIRST_OPC,
2584 0, OST_DISCONNECT, tgt_disconnect,
2585 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2586 TGT_RPC_HANDLER(OST_FIRST_OPC,
2587 0, OST_SET_INFO, ofd_set_info_hdl,
2588 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2589 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2590 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
2591 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2592 OST_SETATTR, ofd_setattr_hdl),
2593 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2594 OST_CREATE, ofd_create_hdl),
2595 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2596 OST_DESTROY, ofd_destroy_hdl),
2597 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
2598 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2599 OST_BRW_READ, tgt_brw_read,
2601 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2602 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
2604 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2605 OST_PUNCH, ofd_punch_hdl,
2607 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
2608 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
2611 static struct tgt_opc_slice ofd_common_slice[] = {
2613 .tos_opc_start = OST_FIRST_OPC,
2614 .tos_opc_end = OST_LAST_OPC,
2615 .tos_hs = ofd_tgt_handlers
2618 .tos_opc_start = OBD_FIRST_OPC,
2619 .tos_opc_end = OBD_LAST_OPC,
2620 .tos_hs = tgt_obd_handlers
2623 .tos_opc_start = LDLM_FIRST_OPC,
2624 .tos_opc_end = LDLM_LAST_OPC,
2625 .tos_hs = tgt_dlm_handlers
2628 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2629 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2630 .tos_hs = tgt_out_handlers
2633 .tos_opc_start = SEQ_FIRST_OPC,
2634 .tos_opc_end = SEQ_LAST_OPC,
2635 .tos_hs = seq_handlers
2638 .tos_opc_start = LFSCK_FIRST_OPC,
2639 .tos_opc_end = LFSCK_LAST_OPC,
2640 .tos_hs = tgt_lfsck_handlers
2647 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2648 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2651 * Implementation of lu_context_key::lct_key_exit.
2653 * Optional method called on lu_context_exit() for all allocated
2655 * It is used in OFD to sanitize context values which may be re-used
2656 * during another request processing by the same thread.
2658 * \param[in] ctx execution context
2659 * \param[in] key context key
2660 * \param[in] data ofd_thread_info
2662 static void ofd_key_exit(const struct lu_context *ctx,
2663 struct lu_context_key *key, void *data)
2665 struct ofd_thread_info *info = data;
2667 info->fti_env = NULL;
2668 info->fti_exp = NULL;
2671 info->fti_pre_version = 0;
2674 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2677 struct lu_context_key ofd_thread_key = {
2678 .lct_tags = LCT_DT_THREAD,
2679 .lct_init = ofd_key_init,
2680 .lct_fini = ofd_key_fini,
2681 .lct_exit = ofd_key_exit
2685 * Initialize OFD device according to parameters in the config log \a cfg.
2687 * This is the main starting point of OFD initialization. It fills all OFD
2688 * parameters with their initial values and calls other initializing functions
2689 * to set up all OFD subsystems.
2691 * \param[in] env execution environment
2692 * \param[in] m OFD device
2693 * \param[in] ldt LU device type of OFD
2694 * \param[in] cfg configuration log
2696 * \retval 0 if successful
2697 * \retval negative value on error
2699 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2700 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2702 const char *dev = lustre_cfg_string(cfg, 0);
2703 struct ofd_thread_info *info = NULL;
2704 struct obd_device *obd;
2705 struct obd_statfs *osfs;
2710 obd = class_name2obd(dev);
2712 CERROR("Cannot find obd with name %s\n", dev);
2716 rc = lu_env_refill((struct lu_env *)env);
2720 obd->u.obt.obt_magic = OBT_MAGIC;
2722 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2723 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2725 spin_lock_init(&m->ofd_flags_lock);
2726 m->ofd_raid_degraded = 0;
2727 m->ofd_syncjournal = 0;
2729 m->ofd_grant_compat_disable = 0;
2730 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2733 spin_lock_init(&m->ofd_osfs_lock);
2734 m->ofd_osfs_age = cfs_time_shift_64(-1000);
2735 m->ofd_osfs_unstable = 0;
2736 m->ofd_statfs_inflight = 0;
2737 m->ofd_osfs_inflight = 0;
2740 spin_lock_init(&m->ofd_grant_lock);
2741 m->ofd_tot_dirty = 0;
2742 m->ofd_tot_granted = 0;
2743 m->ofd_tot_pending = 0;
2744 m->ofd_seq_count = 0;
2745 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2746 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2747 spin_lock_init(&m->ofd_inconsistency_lock);
2749 spin_lock_init(&m->ofd_batch_lock);
2750 init_rwsem(&m->ofd_lastid_rwsem);
2752 obd->u.filter.fo_fl_oss_capa = 0;
2753 INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
2754 obd->u.filter.fo_capa_hash = init_capa_hash();
2755 if (obd->u.filter.fo_capa_hash == NULL)
2758 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2759 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2760 /* set this lu_device to obd, because error handling need it */
2761 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2763 rc = ofd_procfs_init(m);
2765 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2769 /* No connection accepted until configurations will finish */
2770 spin_lock(&obd->obd_dev_lock);
2771 obd->obd_no_conn = 1;
2772 spin_unlock(&obd->obd_dev_lock);
2773 obd->obd_replayable = 1;
2774 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2775 char *str = lustre_cfg_string(cfg, 4);
2777 if (strchr(str, 'n')) {
2778 CWARN("%s: recovery disabled\n", obd->obd_name);
2779 obd->obd_replayable = 0;
2783 info = ofd_info_init(env, NULL);
2787 rc = ofd_stack_init(env, m, cfg);
2789 CERROR("Can't init device stack, rc %d\n", rc);
2790 GOTO(err_fini_proc, rc);
2793 ofd_procfs_add_brw_stats_symlink(m);
2795 /* populate cached statfs data */
2796 osfs = &ofd_info(env)->fti_u.osfs;
2797 rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
2799 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
2800 GOTO(err_fini_stack, rc);
2802 if (!IS_PO2(osfs->os_bsize)) {
2803 CERROR("%s: blocksize (%d) is not a power of 2\n",
2804 obd->obd_name, osfs->os_bsize);
2805 GOTO(err_fini_stack, rc = -EPROTO);
2807 m->ofd_blockbits = fls(osfs->os_bsize) - 1;
2809 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2810 if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
2811 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2813 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2814 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2815 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2816 LDLM_NAMESPACE_SERVER,
2817 LDLM_NAMESPACE_GREEDY,
2819 if (m->ofd_namespace == NULL)
2820 GOTO(err_fini_stack, rc = -ENOMEM);
2821 /* set obd_namespace for compatibility with old code */
2822 obd->obd_namespace = m->ofd_namespace;
2823 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2824 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2825 m->ofd_namespace->ns_lvbp = m;
2827 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2828 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2830 dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
2832 /* Allow at most ddp_grant_reserved% of the available filesystem space
2833 * to be granted to clients, so that any errors in the grant overhead
2834 * calculations do not allow granting more space to clients than can be
2835 * written. Assumes that in aggregate the grant overhead calculations do
2836 * not have more than ddp_grant_reserved% estimation error in them. */
2837 m->ofd_grant_ratio =
2838 ofd_grant_ratio_conv(m->ofd_dt_conf.ddp_grant_reserved);
2840 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2841 OBD_FAIL_OST_ALL_REQUEST_NET,
2842 OBD_FAIL_OST_ALL_REPLY_NET);
2844 GOTO(err_free_ns, rc);
2846 rc = ofd_fs_setup(env, m, obd);
2848 GOTO(err_fini_lut, rc);
2850 rc = ofd_start_inconsistency_verification_thread(m);
2852 GOTO(err_fini_fs, rc);
2857 ofd_fs_cleanup(env, m);
2859 tgt_fini(env, &m->ofd_lut);
2861 ldlm_namespace_free(m->ofd_namespace, 0, obd->obd_force);
2862 obd->obd_namespace = m->ofd_namespace = NULL;
2864 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
2871 * Stop the OFD device
2873 * This function stops the OFD device and all its subsystems.
2874 * This is the end of OFD lifecycle.
2876 * \param[in] env execution environment
2877 * \param[in] m OFD device
2879 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
2881 struct obd_device *obd = ofd_obd(m);
2882 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
2883 struct lfsck_stop stop;
2885 stop.ls_status = LS_PAUSED;
2887 lfsck_stop(env, m->ofd_osd, &stop);
2888 target_recovery_fini(obd);
2889 obd_exports_barrier(obd);
2890 obd_zombie_barrier();
2892 tgt_fini(env, &m->ofd_lut);
2893 ofd_stop_inconsistency_verification_thread(m);
2894 lfsck_degister(env, m->ofd_osd);
2895 ofd_fs_cleanup(env, m);
2897 ofd_free_capa_keys(m);
2898 cleanup_capa_hash(obd->u.filter.fo_capa_hash);
2900 if (m->ofd_namespace != NULL) {
2901 ldlm_namespace_free(m->ofd_namespace, NULL,
2902 d->ld_obd->obd_force);
2903 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
2906 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
2908 LASSERT(atomic_read(&d->ld_ref) == 0);
2909 server_put_mount(obd->obd_name, true);
2914 * Implementation of lu_device_type_operations::ldto_device_fini.
2916 * Finalize device. Dual to ofd_device_init(). It is called from
2917 * obd_precleanup() and stops the current device.
2919 * \param[in] env execution environment
2920 * \param[in] d LU device of OFD
2924 static struct lu_device *ofd_device_fini(const struct lu_env *env,
2925 struct lu_device *d)
2928 ofd_fini(env, ofd_dev(d));
2933 * Implementation of lu_device_type_operations::ldto_device_free.
2935 * Free OFD device. Dual to ofd_device_alloc().
2937 * \param[in] env execution environment
2938 * \param[in] d LU device of OFD
2942 static struct lu_device *ofd_device_free(const struct lu_env *env,
2943 struct lu_device *d)
2945 struct ofd_device *m = ofd_dev(d);
2947 dt_device_fini(&m->ofd_dt_dev);
2953 * Implementation of lu_device_type_operations::ldto_device_alloc.
2955 * This function allocates the new OFD device. It is called from
2956 * obd_setup() if OBD device had lu_device_type defined.
2958 * \param[in] env execution environment
2959 * \param[in] t lu_device_type of OFD device
2960 * \param[in] cfg configuration log
2962 * \retval pointer to the lu_device of just allocated OFD
2963 * \retval ERR_PTR of return value on error
2965 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
2966 struct lu_device_type *t,
2967 struct lustre_cfg *cfg)
2969 struct ofd_device *m;
2970 struct lu_device *l;
2975 return ERR_PTR(-ENOMEM);
2977 l = &m->ofd_dt_dev.dd_lu_dev;
2978 dt_device_init(&m->ofd_dt_dev, t);
2979 rc = ofd_init0(env, m, t, cfg);
2981 ofd_device_free(env, l);
2988 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
2989 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
2991 static struct lu_device_type_operations ofd_device_type_ops = {
2992 .ldto_init = ofd_type_init,
2993 .ldto_fini = ofd_type_fini,
2995 .ldto_start = ofd_type_start,
2996 .ldto_stop = ofd_type_stop,
2998 .ldto_device_alloc = ofd_device_alloc,
2999 .ldto_device_free = ofd_device_free,
3000 .ldto_device_fini = ofd_device_fini
3003 static struct lu_device_type ofd_device_type = {
3004 .ldt_tags = LU_DEVICE_DT,
3005 .ldt_name = LUSTRE_OST_NAME,
3006 .ldt_ops = &ofd_device_type_ops,
3007 .ldt_ctx_tags = LCT_DT_THREAD
3011 * Initialize OFD module.
3013 * This function is called upon module loading. It registers OFD device type
3014 * and prepares all in-memory structures used by all OFD devices.
3016 * \retval 0 if successful
3017 * \retval negative value on error
3019 static int __init ofd_init(void)
3023 rc = lu_kmem_init(ofd_caches);
3027 rc = ofd_fmd_init();
3029 lu_kmem_fini(ofd_caches);
3033 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3034 LUSTRE_OST_NAME, &ofd_device_type);
3041 * This function is called upon OFD module unloading.
3042 * It frees all related structures and unregisters OFD device type.
3044 static void __exit ofd_exit(void)
3047 lu_kmem_fini(ofd_caches);
3048 class_unregister_type(LUSTRE_OST_NAME);
3051 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
3052 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3053 MODULE_LICENSE("GPL");
3055 module_init(ofd_init);
3056 module_exit(ofd_exit);