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, 2015, 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;
170 struct lustre_mount_data *lmd;
176 lmi = server_get_mount(dev);
178 CERROR("Cannot get mount info for %s!\n", dev);
182 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
183 if (lmd != NULL && lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
184 m->ofd_skip_lfsck = 1;
186 /* find bottom osd */
187 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
191 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
192 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
193 OBD_FREE(osdname, MTI_NAME_MAXLEN);
197 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
199 m->ofd_osd = lu2dt_dev(d);
201 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
202 "%s-osd", lustre_cfg_string(cfg, 0));
208 * Finalize the device stack OFD-OSD.
210 * This function cleans OFD-OSD device stack and
211 * disconnects OFD from the OSD.
213 * \param[in] env execution environment
214 * \param[in] m OFD device
215 * \param[in] top top device of stack
217 * \retval 0 if successful
218 * \retval negative value on error
220 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
221 struct lu_device *top)
223 struct obd_device *obd = ofd_obd(m);
224 struct lustre_cfg_bufs bufs;
225 struct lustre_cfg *lcfg;
230 lu_site_purge(env, top->ld_site, ~0);
231 /* process cleanup, pass mdt obd name to get obd umount flags */
232 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
237 lustre_cfg_bufs_set_string(&bufs, 1, flags);
238 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
243 top->ld_ops->ldo_process_config(env, top, lcfg);
244 lustre_cfg_free(lcfg);
246 lu_site_purge(env, top->ld_site, ~0);
247 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
248 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
249 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
252 LASSERT(m->ofd_osd_exp);
253 obd_disconnect(m->ofd_osd_exp);
258 /* For interoperability, see mdt_interop_param[]. */
259 static struct cfg_interop_param ofd_interop_param[] = {
260 { "ost.quota_type", NULL },
265 * Check if parameters are symlinks to the OSD.
267 * Some parameters were moved from ofd to osd and only their
268 * symlinks were kept in ofd by LU-3106. They are:
269 * -writehthrough_cache_enable
270 * -readcache_max_filesize
274 * Since they are not included by the static lprocfs var list, a pre-check
275 * is added for them to avoid "unknown param" errors. If they are matched
276 * in this check, they will be passed to the OSD directly.
278 * \param[in] param parameters to check
280 * \retval true if param is symlink to OSD param
283 static bool match_symlink_param(char *param)
288 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
289 sval = strchr(param, '=');
291 paramlen = sval - param;
292 if (strncmp(param, "writethrough_cache_enable",
294 strncmp(param, "readcache_max_filesize",
296 strncmp(param, "read_cache_enable",
298 strncmp(param, "brw_stats", paramlen) == 0)
307 * Process various configuration parameters.
309 * This function is used by MGS to process specific configurations and
310 * pass them through to the next device in server stack, i.e. the OSD.
312 * \param[in] env execution environment
313 * \param[in] d LU device of OFD
314 * \param[in] cfg parameters to process
316 * \retval 0 if successful
317 * \retval negative value on error
319 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
320 struct lustre_cfg *cfg)
322 struct ofd_device *m = ofd_dev(d);
323 struct dt_device *dt_next = m->ofd_osd;
324 struct lu_device *next = &dt_next->dd_lu_dev;
329 switch (cfg->lcfg_command) {
331 struct obd_device *obd = ofd_obd(m);
332 /* For interoperability */
333 struct cfg_interop_param *ptr = NULL;
334 struct lustre_cfg *old_cfg = NULL;
337 param = lustre_cfg_string(cfg, 1);
339 CERROR("param is empty\n");
344 ptr = class_find_old_param(param, ofd_interop_param);
346 if (ptr->new_param == NULL) {
348 CWARN("For interoperability, skip this %s."
349 " It is obsolete.\n", ptr->old_param);
353 CWARN("Found old param %s, changed it to %s.\n",
354 ptr->old_param, ptr->new_param);
357 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
364 if (match_symlink_param(param)) {
365 rc = next->ld_ops->ldo_process_config(env, next, cfg);
369 rc = class_process_proc_param(PARAM_OST, obd->obd_vars, cfg,
371 if (rc > 0 || rc == -ENOSYS) {
372 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
374 /* we don't understand; pass it on */
375 rc = next->ld_ops->ldo_process_config(env, next, cfg);
379 case LCFG_SPTLRPC_CONF: {
384 /* others are passed further */
385 rc = next->ld_ops->ldo_process_config(env, next, cfg);
392 * Implementation of lu_object_operations::loo_object_init for OFD
394 * Allocate just the next object (OSD) in stack.
396 * \param[in] env execution environment
397 * \param[in] o lu_object of OFD object
398 * \param[in] conf additional configuration parameters, not used here
400 * \retval 0 if successful
401 * \retval negative value on error
403 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
404 const struct lu_object_conf *conf)
406 struct ofd_device *d = ofd_dev(o->lo_dev);
407 struct lu_device *under;
408 struct lu_object *below;
413 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
414 PFID(lu_object_fid(o)));
416 under = &d->ofd_osd->dd_lu_dev;
417 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
419 lu_object_add(o, below);
427 * Implementation of lu_object_operations::loo_object_free.
429 * Finish OFD object lifecycle and free its memory.
431 * \param[in] env execution environment
432 * \param[in] o LU object of OFD object
434 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
436 struct ofd_object *of = ofd_obj(o);
437 struct lu_object_header *h;
442 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
443 PFID(lu_object_fid(o)));
446 lu_object_header_fini(h);
447 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
452 * Implementation of lu_object_operations::loo_object_print.
454 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
455 * LU_OBJECT_DEBUG() for more details about the compound object printing.
457 * \param[in] env execution environment
458 * \param[in] cookie opaque data passed to the printer function
459 * \param[in] p printer function to use
460 * \param[in] o LU object of OFD object
462 * \retval 0 if successful
463 * \retval negative value on error
465 static int ofd_object_print(const struct lu_env *env, void *cookie,
466 lu_printer_t p, const struct lu_object *o)
468 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
471 static struct lu_object_operations ofd_obj_ops = {
472 .loo_object_init = ofd_object_init,
473 .loo_object_free = ofd_object_free,
474 .loo_object_print = ofd_object_print
478 * Implementation of lu_device_operations::lod_object_alloc.
480 * This function allocates OFD part of compound OFD-OSD object and
481 * initializes its header, because OFD is the top device in stack
483 * \param[in] env execution environment
484 * \param[in] hdr object header, NULL for OFD
485 * \param[in] d lu_device
487 * \retval allocated object if successful
488 * \retval NULL value on failed allocation
490 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
491 const struct lu_object_header *hdr,
494 struct ofd_object *of;
498 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
501 struct lu_object_header *h;
503 o = &of->ofo_obj.do_lu;
505 lu_object_header_init(h);
506 lu_object_init(o, h, d);
507 lu_object_add_top(h, o);
508 o->lo_ops = &ofd_obj_ops;
516 * Return the result of LFSCK run to the OFD.
518 * Notify OFD about result of LFSCK run. That may block the new object
519 * creation until problem is fixed by LFSCK.
521 * \param[in] env execution environment
522 * \param[in] data pointer to the OFD device
523 * \param[in] event LFSCK event type
525 * \retval 0 if successful
526 * \retval negative value on unknown event
528 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
529 enum lfsck_events event)
531 struct ofd_device *ofd = data;
532 struct obd_device *obd = ofd_obd(ofd);
535 case LE_LASTID_REBUILDING:
536 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
537 "on the device until the LAST_ID rebuilt successfully.\n",
539 down_write(&ofd->ofd_lastid_rwsem);
540 ofd->ofd_lastid_rebuilding = 1;
541 up_write(&ofd->ofd_lastid_rwsem);
543 case LE_LASTID_REBUILT: {
544 down_write(&ofd->ofd_lastid_rwsem);
545 ofd_seqs_free(env, ofd);
546 ofd->ofd_lastid_rebuilding = 0;
547 ofd->ofd_lastid_gen++;
548 up_write(&ofd->ofd_lastid_rwsem);
549 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
554 CERROR("%s: unknown lfsck event: rc = %d\n",
555 ofd_name(ofd), event);
563 * Implementation of lu_device_operations::ldo_prepare.
565 * This method is called after layer has been initialized and before it starts
566 * serving user requests. In OFD it starts lfsk check routines and initializes
569 * \param[in] env execution environment
570 * \param[in] pdev higher device in stack, NULL for OFD
571 * \param[in] dev lu_device of OFD device
573 * \retval 0 if successful
574 * \retval negative value on error
576 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
577 struct lu_device *dev)
579 struct ofd_thread_info *info;
580 struct ofd_device *ofd = ofd_dev(dev);
581 struct obd_device *obd = ofd_obd(ofd);
582 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
587 info = ofd_info_init(env, NULL);
591 /* initialize lower device */
592 rc = next->ld_ops->ldo_prepare(env, dev, next);
596 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
597 ofd_lfsck_out_notify, ofd, false);
599 CERROR("%s: failed to initialize lfsck: rc = %d\n",
604 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
605 /* The LFSCK instance is registered just now, so it must be there when
606 * register the namespace to such instance. */
607 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
609 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
610 LASSERT(obd->obd_no_conn);
611 spin_lock(&obd->obd_dev_lock);
612 obd->obd_no_conn = 0;
613 spin_unlock(&obd->obd_dev_lock);
615 if (obd->obd_recovering == 0)
616 ofd_postrecov(env, ofd);
622 * Implementation of lu_device_operations::ldo_recovery_complete.
624 * This method notifies all layers about 'recovery complete' event. That means
625 * device is in full state and consistent. An OFD calculates available grant
626 * space upon this event.
628 * \param[in] env execution environment
629 * \param[in] dev lu_device of OFD device
631 * \retval 0 if successful
632 * \retval negative value on error
634 static int ofd_recovery_complete(const struct lu_env *env,
635 struct lu_device *dev)
637 struct ofd_device *ofd = ofd_dev(dev);
638 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
639 int rc = 0, max_precreate;
644 * Grant space for object precreation on the self export.
645 * This initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
646 * is enough to create 10k objects. More space is then acquired for
647 * precreation in ofd_grant_create().
649 max_precreate = OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace / 2;
650 ofd_grant_connect(env, dev->ld_obd->obd_self_export, max_precreate,
652 rc = next->ld_ops->ldo_recovery_complete(env, next);
657 * lu_device_operations matrix for OFD device.
659 static struct lu_device_operations ofd_lu_ops = {
660 .ldo_object_alloc = ofd_object_alloc,
661 .ldo_process_config = ofd_process_config,
662 .ldo_recovery_complete = ofd_recovery_complete,
663 .ldo_prepare = ofd_prepare,
666 LPROC_SEQ_FOPS(lprocfs_nid_stats_clear);
669 * Initialize all needed procfs entries for OFD device.
671 * \param[in] ofd OFD device
673 * \retval 0 if successful
674 * \retval negative value on error
676 static int ofd_procfs_init(struct ofd_device *ofd)
678 struct obd_device *obd = ofd_obd(ofd);
679 struct proc_dir_entry *entry;
684 /* lprocfs must be setup before the ofd so state can be safely added
685 * to /proc incrementally as the ofd is setup */
686 obd->obd_vars = lprocfs_ofd_obd_vars;
687 rc = lprocfs_obd_setup(obd);
689 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
694 rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
696 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
698 GOTO(obd_cleanup, rc);
701 obd->obd_uses_nid_stats = 1;
703 entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
706 CERROR("%s: error %d setting up lprocfs for %s\n",
707 obd->obd_name, rc, "exports");
708 GOTO(obd_cleanup, rc);
710 obd->obd_proc_exports_entry = entry;
712 entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
713 obd, &lprocfs_nid_stats_clear_fops);
716 CERROR("%s: add proc entry 'clear' failed: %d.\n",
718 GOTO(obd_cleanup, rc);
721 ofd_stats_counter_init(obd->obd_stats);
723 rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
724 ofd_stats_counter_init);
726 GOTO(obd_cleanup, rc);
729 lprocfs_obd_cleanup(obd);
730 lprocfs_free_obd_stats(obd);
736 * Expose OSD statistics to OFD layer.
738 * The osd interfaces to the backend file system exposes useful data
739 * such as brw_stats and read or write cache states. This same data
740 * needs to be exposed into the obdfilter (ofd) layer to maintain
741 * backwards compatibility. This function creates the symlinks in the
742 * proc layer to enable this.
744 * \param[in] ofd OFD device
746 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
748 struct obd_device *obd = ofd_obd(ofd);
749 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
751 if (obd->obd_proc_entry == NULL)
754 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
755 "../../%s/%s/brw_stats",
756 osd_obd->obd_type->typ_name, obd->obd_name);
758 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
759 "../../%s/%s/read_cache_enable",
760 osd_obd->obd_type->typ_name, obd->obd_name);
762 lprocfs_add_symlink("readcache_max_filesize",
764 "../../%s/%s/readcache_max_filesize",
765 osd_obd->obd_type->typ_name, obd->obd_name);
767 lprocfs_add_symlink("writethrough_cache_enable",
769 "../../%s/%s/writethrough_cache_enable",
770 osd_obd->obd_type->typ_name, obd->obd_name);
774 * Cleanup all procfs entries in OFD.
776 * \param[in] ofd OFD device
778 static void ofd_procfs_fini(struct ofd_device *ofd)
780 struct obd_device *obd = ofd_obd(ofd);
782 lprocfs_free_per_client_stats(obd);
783 lprocfs_obd_cleanup(obd);
784 lprocfs_free_obd_stats(obd);
785 lprocfs_job_stats_fini(obd);
789 * Stop SEQ/FID server on OFD.
791 * \param[in] env execution environment
792 * \param[in] ofd OFD device
794 * \retval 0 if successful
795 * \retval negative value on error
797 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
799 return seq_site_fini(env, &ofd->ofd_seq_site);
803 * Start SEQ/FID server on OFD.
805 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
806 * It also connects to the master server to get own FID sequence (SEQ) range
807 * to this particular OFD. Typically that happens when the OST is first
808 * formatted or in the rare case that it exhausts the local sequence range.
810 * The sequence range is allocated out to the MDTs for OST object allocations,
811 * and not directly to the clients.
813 * \param[in] env execution environment
814 * \param[in] ofd OFD device
816 * \retval 0 if successful
817 * \retval negative value on error
819 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
821 struct seq_server_site *ss = &ofd->ofd_seq_site;
822 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
823 char *obd_name = ofd_name(ofd);
827 ss = &ofd->ofd_seq_site;
828 lu->ld_site->ld_seq_site = ss;
829 ss->ss_lu = lu->ld_site;
830 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
832 OBD_ALLOC_PTR(ss->ss_server_seq);
833 if (ss->ss_server_seq == NULL)
834 GOTO(out_free, rc = -ENOMEM);
836 OBD_ALLOC(name, strlen(obd_name) + 10);
838 OBD_FREE_PTR(ss->ss_server_seq);
839 ss->ss_server_seq = NULL;
840 GOTO(out_free, rc = -ENOMEM);
843 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
844 LUSTRE_SEQ_SERVER, ss);
846 CERROR("%s : seq server init error %d\n", obd_name, rc);
849 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
851 OBD_ALLOC_PTR(ss->ss_client_seq);
852 if (ss->ss_client_seq == NULL)
853 GOTO(out_free, rc = -ENOMEM);
855 snprintf(name, strlen(obd_name) + 6, "%p-super", obd_name);
856 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
859 CERROR("%s : seq client init error %d\n", obd_name, rc);
862 OBD_FREE(name, strlen(obd_name) + 10);
865 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
869 if (ss->ss_server_seq) {
870 seq_server_fini(ss->ss_server_seq, env);
871 OBD_FREE_PTR(ss->ss_server_seq);
872 ss->ss_server_seq = NULL;
875 if (ss->ss_client_seq) {
876 seq_client_fini(ss->ss_client_seq);
877 OBD_FREE_PTR(ss->ss_client_seq);
878 ss->ss_client_seq = NULL;
882 OBD_FREE(name, strlen(obd_name) + 10);
891 * OFD request handler for OST_SET_INFO RPC.
893 * This is OFD-specific part of request handling
895 * \param[in] tsi target session environment for this request
897 * \retval 0 if successful
898 * \retval negative value on error
900 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
902 struct ptlrpc_request *req = tgt_ses_req(tsi);
903 struct ost_body *body = NULL, *repbody;
904 void *key, *val = NULL;
905 int keylen, vallen, rc = 0;
906 bool is_grant_shrink;
910 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
912 DEBUG_REQ(D_HA, req, "no set_info key");
913 RETURN(err_serious(-EFAULT));
915 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
918 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
920 DEBUG_REQ(D_HA, req, "no set_info val");
921 RETURN(err_serious(-EFAULT));
923 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
926 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
928 /* In this case the value is actually an RMF_OST_BODY, so we
929 * transmutate the type of this PTLRPC */
930 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
932 rc = req_capsule_server_pack(tsi->tsi_pill);
936 if (is_grant_shrink) {
937 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
939 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
942 /** handle grant shrink, similar to a read request */
943 ofd_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
945 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
947 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
949 } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
950 rc = tgt_adapt_sptlrpc_conf(tsi->tsi_tgt, 0);
952 CERROR("%s: Unsupported key %s\n",
953 tgt_name(tsi->tsi_tgt), (char *)key);
956 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
963 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
965 * This function returns a list of extents which describes how a file's
966 * blocks are laid out on the disk.
968 * \param[in] env execution environment
969 * \param[in] ofd OFD device
970 * \param[in] fid FID of object
971 * \param[in] fiemap fiemap structure to fill with data
973 * \retval 0 if \a fiemap is filled with data successfully
974 * \retval negative value on error
976 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
977 struct lu_fid *fid, struct fiemap *fiemap)
979 struct ofd_object *fo;
982 fo = ofd_object_find(env, ofd, fid);
984 CERROR("%s: error finding object "DFID"\n",
985 ofd_name(ofd), PFID(fid));
989 ofd_read_lock(env, fo);
990 if (ofd_object_exists(fo))
991 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
994 ofd_read_unlock(env, fo);
995 ofd_object_put(env, fo);
999 struct locked_region {
1000 struct list_head list;
1001 struct lustre_handle lh;
1005 * Lock single extent and save lock handle in the list.
1007 * This is supplemental function for lock_zero_regions(). It allocates
1008 * new locked_region structure and locks it with extent lock, then adds
1009 * it to the list of all such regions.
1011 * \param[in] ns LDLM namespace
1012 * \param[in] res_id resource ID
1013 * \param[in] begin start of region
1014 * \param[in] end end of region
1015 * \param[in] locked list head of regions list
1017 * \retval 0 if successful locking
1018 * \retval negative value on error
1020 static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1021 unsigned long long begin, unsigned long long end,
1022 struct list_head *locked)
1024 struct locked_region *region = NULL;
1028 LASSERT(begin <= end);
1029 OBD_ALLOC_PTR(region);
1033 rc = tgt_extent_lock(ns, res_id, begin, end, ®ion->lh,
1038 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
1040 list_add(®ion->list, locked);
1046 * Lock the sparse areas of given resource.
1048 * The locking of sparse areas will cause dirty data to be flushed back from
1049 * clients. This is used when getting the FIEMAP of an object to make sure
1050 * there is no unaccounted cached data on clients.
1052 * This function goes through \a fiemap list of extents and locks only sparse
1053 * areas between extents.
1055 * \param[in] ns LDLM namespace
1056 * \param[in] res_id resource ID
1057 * \param[in] fiemap file extents mapping on disk
1058 * \param[in] locked list head of regions list
1060 * \retval 0 if successful
1061 * \retval negative value on error
1063 static int lock_zero_regions(struct ldlm_namespace *ns,
1064 struct ldlm_res_id *res_id,
1065 struct fiemap *fiemap,
1066 struct list_head *locked)
1068 __u64 begin = fiemap->fm_start;
1071 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1075 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1076 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1077 if (fiemap_start[i].fe_logical > begin) {
1078 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1079 begin, fiemap_start[i].fe_logical);
1080 rc = lock_region(ns, res_id, begin,
1081 fiemap_start[i].fe_logical, locked);
1086 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1089 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1090 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1091 begin, fiemap->fm_start + fiemap->fm_length);
1092 rc = lock_region(ns, res_id, begin,
1093 fiemap->fm_start + fiemap->fm_length, locked);
1100 * Unlock all previously locked sparse areas for given resource.
1102 * This function goes through list of locked regions, unlocking and freeing
1105 * \param[in] ns LDLM namespace
1106 * \param[in] locked list head of regions list
1109 unlock_zero_regions(struct ldlm_namespace *ns, struct list_head *locked)
1111 struct locked_region *entry, *temp;
1113 list_for_each_entry_safe(entry, temp, locked, list) {
1114 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
1115 tgt_extent_unlock(&entry->lh, LCK_PR);
1116 list_del(&entry->list);
1117 OBD_FREE_PTR(entry);
1122 * OFD request handler for OST_GET_INFO RPC.
1124 * This is OFD-specific part of request handling. The OFD-specific keys are:
1125 * - KEY_LAST_ID (obsolete)
1129 * This function reads needed data from storage and fills reply with it.
1131 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1132 * and is kept for compatibility.
1134 * \param[in] tsi target session environment for this request
1136 * \retval 0 if successful
1137 * \retval negative value on error
1139 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1141 struct obd_export *exp = tsi->tsi_exp;
1142 struct ofd_device *ofd = ofd_exp(exp);
1143 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1146 int replylen, rc = 0;
1150 /* this common part for get_info rpc */
1151 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1153 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1154 RETURN(err_serious(-EPROTO));
1156 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1159 if (KEY_IS(KEY_LAST_ID)) {
1161 struct ofd_seq *oseq;
1163 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1164 rc = req_capsule_server_pack(tsi->tsi_pill);
1166 RETURN(err_serious(rc));
1168 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1170 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1171 (u64)exp->exp_filter_data.fed_group);
1175 *last_id = ofd_seq_last_oid(oseq);
1176 ofd_seq_put(tsi->tsi_env, oseq);
1177 } else if (KEY_IS(KEY_FIEMAP)) {
1178 struct ll_fiemap_info_key *fm_key;
1179 struct fiemap *fiemap;
1182 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1184 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1185 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1187 RETURN(err_serious(rc));
1189 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1191 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1193 replylen = fiemap_count_to_size(
1194 fm_key->lfik_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->lfik_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->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1212 fm_key->lfik_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 enum ldlm_mode 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 u64 seq = ostid_seq(oi);
1473 u64 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 u64 seq = ostid_seq(&oa->o_oi);
1565 u64 oid = ostid_id(&oa->o_oi);
1566 struct ofd_seq *oseq;
1573 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1576 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1577 if (repbody == NULL)
1580 down_read(&ofd->ofd_lastid_rwsem);
1581 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1582 * we may do that in the future.
1583 * Return -ENOSPC until the LAST_ID rebuilt. */
1584 if (unlikely(ofd->ofd_lastid_rebuilding))
1585 GOTO(out_sem, rc = -ENOSPC);
1587 rep_oa = &repbody->oa;
1588 rep_oa->o_oi = oa->o_oi;
1590 LASSERT(seq >= FID_SEQ_OST_MDT0);
1591 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1593 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1595 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1597 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1598 ofd_name(ofd), seq, PTR_ERR(oseq));
1599 GOTO(out_sem, rc = -EINVAL);
1602 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1603 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1604 if (!ofd_obd(ofd)->obd_recovering ||
1605 oid > ofd_seq_last_oid(oseq)) {
1606 CERROR("%s: recreate objid "DOSTID" > last id "LPU64
1607 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1608 ofd_seq_last_oid(oseq));
1609 GOTO(out_nolock, rc = -EINVAL);
1611 /* Do nothing here, we re-create objects during recovery
1612 * upon write replay, see ofd_preprw_write() */
1613 GOTO(out_nolock, rc = 0);
1615 /* former ofd_handle_precreate */
1616 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1617 (oa->o_flags & OBD_FL_DELORPHAN)) {
1618 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1620 /* destroy orphans */
1621 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1622 exp->exp_conn_cnt) {
1623 CERROR("%s: dropping old orphan cleanup request\n",
1625 GOTO(out_nolock, rc = 0);
1627 /* This causes inflight precreates to abort and drop lock */
1628 oseq->os_destroys_in_progress = 1;
1629 mutex_lock(&oseq->os_create_lock);
1630 if (!oseq->os_destroys_in_progress) {
1631 CERROR("%s:["LPU64"] destroys_in_progress already"
1632 " cleared\n", ofd_name(ofd), seq);
1633 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1636 diff = oid - ofd_seq_last_oid(oseq);
1637 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1638 ofd_seq_last_oid(oseq), diff);
1639 if (-diff > OST_MAX_PRECREATE) {
1640 /* Let MDS know that we are so far ahead. */
1641 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq) + 1);
1643 } else if (diff < 0) {
1644 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1646 oseq->os_destroys_in_progress = 0;
1648 /* XXX: Used by MDS for the first time! */
1649 oseq->os_destroys_in_progress = 0;
1652 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1653 ofd->ofd_lastid_gen)) {
1654 /* Keep the export ref so we can send the reply. */
1655 ofd_obd_disconnect(class_export_get(exp));
1656 GOTO(out_nolock, rc = -ENOTCONN);
1659 mutex_lock(&oseq->os_create_lock);
1660 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1661 exp->exp_conn_cnt) {
1662 CERROR("%s: dropping old precreate request\n",
1666 /* only precreate if seq is 0, IDIF or normal and also o_id
1667 * must be specfied */
1668 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1669 !fid_seq_is_idif(seq)) || oid == 0) {
1670 diff = 1; /* shouldn't we create this right now? */
1672 diff = oid - ofd_seq_last_oid(oseq);
1673 /* Do sync create if the seq is about to used up */
1674 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1675 if (unlikely(oid >= IDIF_MAX_OID - 1))
1677 } else if (fid_seq_is_norm(seq)) {
1679 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1682 CERROR("%s : invalid o_seq "DOSTID"\n",
1683 ofd_name(ofd), POSTID(&oa->o_oi));
1684 GOTO(out, rc = -EINVAL);
1689 CERROR("%s: invalid precreate request for "
1690 DOSTID", last_id " LPU64 ". "
1691 "Likely MDS last_id corruption\n",
1692 ofd_name(ofd), POSTID(&oa->o_oi),
1693 ofd_seq_last_oid(oseq));
1694 GOTO(out, rc = -EINVAL);
1699 cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
1704 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1705 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1706 /* don't enforce grant during orphan recovery */
1707 granted = ofd_grant_create(tsi->tsi_env,
1708 ofd_obd(ofd)->obd_self_export,
1713 CDEBUG(D_HA, "%s: failed to acquire grant "
1714 "space for precreate (%d): rc = %d\n",
1715 ofd_name(ofd), diff, rc);
1720 /* This can happen if a new OST is formatted and installed
1721 * in place of an old one at the same index. Instead of
1722 * precreating potentially millions of deleted old objects
1723 * (possibly filling the OST), only precreate the last batch.
1724 * LFSCK will eventually clean up any orphans. LU-14 */
1725 if (diff > 5 * OST_MAX_PRECREATE) {
1726 diff = OST_MAX_PRECREATE / 2;
1727 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1728 "OST replaced or reformatted: "
1729 "LFSCK will clean up",
1732 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1733 "%u larger than the LAST_ID "DOSTID", only "
1734 "precreating the last %u objects.\n",
1735 ofd_name(ofd), POSTID(&oa->o_oi),
1736 5 * OST_MAX_PRECREATE,
1737 POSTID(&oseq->os_oi), diff);
1738 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1742 next_id = ofd_seq_last_oid(oseq) + 1;
1743 count = ofd_precreate_batch(ofd, diff);
1745 CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1746 " at "LPU64"\n", ofd_name(ofd),
1747 count, seq, next_id);
1749 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1750 && cfs_time_after(jiffies, enough_time)) {
1751 CDEBUG(D_HA, "%s: Slow creates, %d/%d objects"
1752 " created at a rate of %d/s\n",
1753 ofd_name(ofd), created, diff + created,
1754 created / DISK_TIMEOUT);
1758 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1759 oseq, count, sync_trans);
1763 } else if (rc < 0) {
1769 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1770 LCONSOLE_WARN("%s: can't create the same count of"
1771 " objects when replaying the request"
1772 " (diff is %d). see LU-4621\n",
1773 ofd_name(ofd), diff);
1776 /* some objects got created, we can return
1777 * them, even if last creation failed */
1780 CERROR("%s: unable to precreate: rc = %d\n",
1783 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1784 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1785 ofd_grant_commit(ofd_obd(ofd)->obd_self_export, granted,
1790 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1793 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1796 mutex_unlock(&oseq->os_create_lock);
1799 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1800 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1801 struct lu_fid *fid = &info->fti_fid;
1803 /* For compatible purpose, it needs to convert back to
1804 * OST ID before put it on wire. */
1805 *fid = rep_oa->o_oi.oi_fid;
1806 fid_to_ostid(fid, &rep_oa->o_oi);
1808 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1810 ofd_seq_put(tsi->tsi_env, oseq);
1813 up_read(&ofd->ofd_lastid_rwsem);
1818 * OFD request handler for OST_DESTROY RPC.
1820 * This is OFD-specific part of request handling. It destroys data objects
1821 * related to destroyed object on MDT.
1823 * \param[in] tsi target session environment for this request
1825 * \retval 0 if successful
1826 * \retval negative value on error
1828 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1830 const struct ost_body *body = tsi->tsi_ost_body;
1831 struct ost_body *repbody;
1832 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1833 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1834 struct lu_fid *fid = &fti->fti_fid;
1841 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1844 /* This is old case for clients before Lustre 2.4 */
1845 /* If there's a DLM request, cancel the locks mentioned in it */
1846 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1848 struct ldlm_request *dlm;
1850 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1853 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1856 *fid = body->oa.o_oi.oi_fid;
1857 oid = ostid_id(&body->oa.o_oi);
1860 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1862 /* check that o_misc makes sense */
1863 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1864 count = body->oa.o_misc;
1866 count = 1; /* default case - single destroy */
1868 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1869 POSTID(&body->oa.o_oi), count);
1874 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1875 if (lrc == -ENOENT) {
1877 "%s: destroying non-existent object "DFID"\n",
1878 ofd_name(ofd), PFID(fid));
1879 /* rewrite rc with -ENOENT only if it is 0 */
1882 } else if (lrc != 0) {
1883 CERROR("%s: error destroying object "DFID": %d\n",
1884 ofd_name(ofd), PFID(fid), lrc);
1890 lrc = fid_set_id(fid, oid);
1891 if (unlikely(lrc != 0 && count > 0))
1892 GOTO(out, rc = lrc);
1895 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1901 fid_to_ostid(fid, &repbody->oa.o_oi);
1906 * OFD request handler for OST_STATFS RPC.
1908 * This function gets statfs data from storage as part of request
1911 * \param[in] tsi target session environment for this request
1913 * \retval 0 if successful
1914 * \retval negative value on error
1916 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1918 struct obd_statfs *osfs;
1923 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1925 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1926 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
1928 CERROR("%s: statfs failed: rc = %d\n",
1929 tgt_name(tsi->tsi_tgt), rc);
1931 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1934 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1941 * OFD request handler for OST_SYNC RPC.
1943 * Sync object data or all filesystem data to the disk and pack the
1946 * \param[in] tsi target session environment for this request
1948 * \retval 0 if successful
1949 * \retval negative value on error
1951 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1953 struct ost_body *body = tsi->tsi_ost_body;
1954 struct ost_body *repbody;
1955 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1956 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1957 struct ofd_object *fo = NULL;
1962 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1964 /* if no objid is specified, it means "sync whole filesystem" */
1965 if (!fid_is_zero(&tsi->tsi_fid)) {
1966 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1968 RETURN(PTR_ERR(fo));
1971 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1972 fo != NULL ? ofd_object_child(fo) : NULL,
1973 repbody->oa.o_size, repbody->oa.o_blocks);
1977 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1982 repbody->oa.o_oi = body->oa.o_oi;
1983 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1985 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1987 obdo_from_la(&repbody->oa, &fti->fti_attr,
1990 /* don't return rc from getattr */
1995 ofd_object_put(tsi->tsi_env, fo);
2000 * OFD request handler for OST_PUNCH RPC.
2002 * This is part of request processing. Validate request fields,
2003 * punch (truncate) the given OFD object and pack reply.
2005 * \param[in] tsi target session environment for this request
2007 * \retval 0 if successful
2008 * \retval negative value on error
2010 static int ofd_punch_hdl(struct tgt_session_info *tsi)
2012 const struct obdo *oa = &tsi->tsi_ost_body->oa;
2013 struct ost_body *repbody;
2014 struct ofd_thread_info *info = tsi2ofd_info(tsi);
2015 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
2016 struct ldlm_resource *res;
2017 struct ofd_object *fo;
2018 struct filter_fid *ff = NULL;
2020 struct lustre_handle lh = { 0, };
2027 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
2028 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
2030 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2031 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2032 RETURN(err_serious(-EPROTO));
2034 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2035 if (repbody == NULL)
2036 RETURN(err_serious(-ENOMEM));
2038 /* punch start,end are passed in o_size,o_blocks throught wire */
2042 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2045 /* standard truncate optimization: if file body is completely
2046 * destroyed, don't send data back to the server. */
2048 flags |= LDLM_FL_AST_DISCARD_DATA;
2050 repbody->oa.o_oi = oa->o_oi;
2051 repbody->oa.o_valid = OBD_MD_FLID;
2053 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2054 oa->o_flags & OBD_FL_SRVLOCK;
2057 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
2063 CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
2064 ", start = "LPD64", end = "LPD64"\n", PFID(&tsi->tsi_fid),
2065 oa->o_valid, start, end);
2067 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2070 GOTO(out, rc = PTR_ERR(fo));
2072 la_from_obdo(&info->fti_attr, oa,
2073 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2074 info->fti_attr.la_size = start;
2075 info->fti_attr.la_valid |= LA_SIZE;
2077 if (oa->o_valid & OBD_MD_FLFID) {
2078 ff = &info->fti_mds_fid;
2079 ofd_prepare_fidea(ff, oa);
2082 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2083 ff, (struct obdo *)oa);
2087 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2091 ofd_object_put(tsi->tsi_env, fo);
2094 tgt_extent_unlock(&lh, LCK_PW);
2096 /* we do not call this before to avoid lu_object_find() in
2097 * ->lvbo_update() holding another reference on the object.
2098 * otherwise concurrent destroy can make the object unavailable
2099 * for 2nd lu_object_find() waiting for the first reference
2100 * to go... deadlock! */
2101 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2104 ldlm_res_lvbo_update(res, NULL, 0);
2105 ldlm_resource_putref(res);
2112 * OFD request handler for OST_QUOTACTL RPC.
2114 * This is part of request processing to validate incoming request fields,
2115 * get the requested data from OSD and pack reply.
2117 * \param[in] tsi target session environment for this request
2119 * \retval 0 if successful
2120 * \retval negative value on error
2122 static int ofd_quotactl(struct tgt_session_info *tsi)
2124 struct obd_quotactl *oqctl, *repoqc;
2125 struct lu_nodemap *nodemap =
2126 tsi->tsi_exp->exp_target_data.ted_nodemap;
2132 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2134 RETURN(err_serious(-EPROTO));
2136 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2138 RETURN(err_serious(-ENOMEM));
2143 if (oqctl->qc_type == USRQUOTA)
2144 id = nodemap_map_id(nodemap, NODEMAP_UID,
2145 NODEMAP_CLIENT_TO_FS,
2147 else if (oqctl->qc_type == GRPQUOTA)
2148 id = nodemap_map_id(nodemap, NODEMAP_GID,
2149 NODEMAP_CLIENT_TO_FS,
2152 if (repoqc->qc_id != id)
2153 swap(repoqc->qc_id, id);
2155 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2157 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2160 if (repoqc->qc_id != id)
2161 swap(repoqc->qc_id, id);
2167 * Calculate the amount of time for lock prolongation.
2169 * This is helper for ofd_prolong_extent_locks() function to get
2170 * the timeout extra time.
2172 * \param[in] req current request
2174 * \retval amount of time to extend the timeout with
2176 static inline int prolong_timeout(struct ptlrpc_request *req,
2177 struct ldlm_lock *lock)
2179 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2182 return obd_timeout / 2;
2184 /* We are in the middle of the process - BL AST is sent, CANCEL
2185 is ahead. Take half of AT + IO process time. */
2186 return at_est2timeout(at_get(&svcpt->scp_at_estimate)) +
2187 (ldlm_bl_timeout(lock) >> 1);
2191 * Prolong single lock timeout.
2193 * This is supplemental function to the ofd_prolong_locks(). It prolongs
2196 * \param[in] tsi target session environment for this request
2197 * \param[in] lock LDLM lock to prolong
2198 * \param[in] extent related extent
2199 * \param[in] timeout timeout value to add
2201 * \retval 0 if lock is not suitable for prolongation
2202 * \retval 1 if lock was prolonged successfully
2204 static int ofd_prolong_one_lock(struct tgt_session_info *tsi,
2205 struct ldlm_lock *lock,
2206 struct ldlm_extent *extent)
2208 int timeout = prolong_timeout(tgt_ses_req(tsi), lock);
2210 if (lock->l_flags & LDLM_FL_DESTROYED) /* lock already cancelled */
2213 /* XXX: never try to grab resource lock here because we're inside
2214 * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
2215 * res lock and then exp_bl_list_lock. */
2217 if (!(lock->l_flags & LDLM_FL_AST_SENT))
2218 /* ignore locks not being cancelled */
2221 LDLM_DEBUG(lock, "refreshed for req x"LPU64" ext("LPU64"->"LPU64") "
2222 "to %ds.\n", tgt_ses_req(tsi)->rq_xid, extent->start,
2223 extent->end, timeout);
2225 /* OK. this is a possible lock the user holds doing I/O
2226 * let's refresh eviction timer for it */
2227 ldlm_refresh_waiting_lock(lock, timeout);
2232 * Prolong lock timeout for the given extent.
2234 * This function finds all locks related with incoming request and
2235 * prolongs their timeout.
2237 * If a client is holding a lock for a long time while it sends
2238 * read or write RPCs to the OST for the object under this lock,
2239 * then we don't want the OST to evict the client. Otherwise,
2240 * if the network or disk is very busy then the client may not
2241 * be able to make any progress to clear out dirty pages under
2242 * the lock and the application will fail.
2244 * Every time a Bulk Read/Write (BRW) request arrives for the object
2245 * covered by the lock, extend the timeout on that lock. The RPC should
2246 * contain a lock handle for the lock it is using, but this
2247 * isn't handled correctly by all client versions, and the
2248 * request may cover multiple locks.
2250 * \param[in] tsi target session environment for this request
2251 * \param[in] start start of extent
2252 * \param[in] end end of extent
2254 * \retval number of prolonged locks
2256 static int ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2257 __u64 start, __u64 end)
2259 struct obd_export *exp = tsi->tsi_exp;
2260 struct obdo *oa = &tsi->tsi_ost_body->oa;
2261 struct ldlm_extent extent = {
2265 struct ldlm_lock *lock;
2270 if (oa->o_valid & OBD_MD_FLHANDLE) {
2271 /* mostly a request should be covered by only one lock, try
2273 lock = ldlm_handle2lock(&oa->o_handle);
2275 /* Fast path to check if the lock covers the whole IO
2276 * region exclusively. */
2277 if (lock->l_granted_mode == LCK_PW &&
2278 ldlm_extent_contain(&lock->l_policy_data.l_extent,
2281 LASSERT(lock->l_export == exp);
2282 lock_count = ofd_prolong_one_lock(tsi, lock,
2284 LDLM_LOCK_PUT(lock);
2287 lock->l_last_used = cfs_time_current();
2288 LDLM_LOCK_PUT(lock);
2292 spin_lock_bh(&exp->exp_bl_list_lock);
2293 list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
2294 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
2295 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
2297 /* ignore waiting locks, no more granted locks in the list */
2298 if (lock->l_granted_mode != lock->l_req_mode)
2301 if (!ldlm_res_eq(&tsi->tsi_resid, &lock->l_resource->lr_name))
2304 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
2308 lock_count += ofd_prolong_one_lock(tsi, lock, &extent);
2310 spin_unlock_bh(&exp->exp_bl_list_lock);
2316 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2318 * Determine if \a lock and the lock from request \a req are equivalent
2319 * by comparing their resource names, modes, and extents.
2321 * It is used to give priority to read and write RPCs being done
2322 * under this lock so that the client can drop the contended
2323 * lock more quickly and let other clients use it. This improves
2324 * overall performance in the case where the first client gets a
2325 * very large lock extent that prevents other clients from
2326 * submitting their writes.
2328 * \param[in] req ptlrpc_request being processed
2329 * \param[in] lock contended lock to match
2331 * \retval 1 if lock is matched
2332 * \retval 0 otherwise
2334 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2335 struct ldlm_lock *lock)
2337 struct niobuf_remote *rnb;
2338 struct obd_ioobj *ioo;
2339 enum ldlm_mode mode;
2340 struct ldlm_extent ext;
2341 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2345 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2346 LASSERT(ioo != NULL);
2348 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2349 LASSERT(rnb != NULL);
2351 ext.start = rnb->rnb_offset;
2352 rnb += ioo->ioo_bufcnt - 1;
2353 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2355 LASSERT(lock->l_resource != NULL);
2356 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2359 /* a bulk write can only hold a reference on a PW extent lock */
2361 if (opc == OST_READ)
2362 /* whereas a bulk read can be protected by either a PR or PW
2366 if (!(lock->l_granted_mode & mode))
2369 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2373 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2375 * Check for whether the given PTLRPC request (\a req) is blocking
2376 * an LDLM lock cancel.
2378 * \param[in] req the incoming request
2380 * \retval 1 if \a req is blocking an LDLM lock cancel
2381 * \retval 0 if it is not
2383 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2385 struct tgt_session_info *tsi;
2386 struct obd_ioobj *ioo;
2387 struct niobuf_remote *rnb;
2393 /* Don't use tgt_ses_info() to get session info, because lock_match()
2394 * can be called while request has no processing thread yet. */
2395 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2398 * Use LASSERT below because malformed RPCs should have
2399 * been filtered out in tgt_hpreq_handler().
2401 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2402 LASSERT(ioo != NULL);
2404 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2405 LASSERT(rnb != NULL);
2406 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2408 start = rnb->rnb_offset;
2409 rnb += ioo->ioo_bufcnt - 1;
2410 end = rnb->rnb_offset + rnb->rnb_len - 1;
2412 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
2413 " ("LPU64"->"LPU64")\n",
2414 tgt_name(tsi->tsi_tgt), current->comm,
2415 PFID(&tsi->tsi_fid), start, end);
2417 lock_count = ofd_prolong_extent_locks(tsi, start, end);
2419 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2420 tgt_name(tsi->tsi_tgt), lock_count, req);
2422 RETURN(lock_count > 0);
2426 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2428 * Called after the request has been handled. It refreshes lock timeout again
2429 * so that client has more time to send lock cancel RPC.
2431 * \param[in] req request which is being processed.
2433 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2435 ofd_rw_hpreq_check(req);
2439 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2441 * This function checks if the given lock is the same by its resname, mode
2442 * and extent as one taken from the request.
2443 * It is used to give priority to punch/truncate RPCs that might lead to
2444 * the fastest release of that lock when a lock is contended.
2446 * \param[in] req ptlrpc_request being processed
2447 * \param[in] lock contended lock to match
2449 * \retval 1 if lock is matched
2450 * \retval 0 otherwise
2452 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2453 struct ldlm_lock *lock)
2455 struct tgt_session_info *tsi;
2457 /* Don't use tgt_ses_info() to get session info, because lock_match()
2458 * can be called while request has no processing thread yet. */
2459 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2462 * Use LASSERT below because malformed RPCs should have
2463 * been filtered out in tgt_hpreq_handler().
2465 LASSERT(tsi->tsi_ost_body != NULL);
2466 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2467 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2474 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2476 * High-priority queue request check for whether the given punch request
2477 * (\a req) is blocking an LDLM lock cancel.
2479 * \param[in] req the incoming request
2481 * \retval 1 if \a req is blocking an LDLM lock cancel
2482 * \retval 0 if it is not
2484 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2486 struct tgt_session_info *tsi;
2492 /* Don't use tgt_ses_info() to get session info, because lock_match()
2493 * can be called while request has no processing thread yet. */
2494 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2495 LASSERT(tsi != NULL);
2496 oa = &tsi->tsi_ost_body->oa;
2498 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2499 oa->o_flags & OBD_FL_SRVLOCK));
2502 "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
2503 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2504 tsi->tsi_resid.name[1], oa->o_size, oa->o_blocks);
2506 lock_count = ofd_prolong_extent_locks(tsi, oa->o_size, oa->o_blocks);
2508 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2509 tgt_name(tsi->tsi_tgt), lock_count, req);
2511 RETURN(lock_count > 0);
2515 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2517 * Called after the request has been handled. It refreshes lock timeout again
2518 * so that client has more time to send lock cancel RPC.
2520 * \param[in] req request which is being processed.
2522 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2524 ofd_punch_hpreq_check(req);
2527 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2528 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2529 .hpreq_check = ofd_rw_hpreq_check,
2530 .hpreq_fini = ofd_rw_hpreq_fini
2533 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2534 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2535 .hpreq_check = ofd_punch_hpreq_check,
2536 .hpreq_fini = ofd_punch_hpreq_fini
2540 * Assign high priority operations to an IO request.
2542 * Check if the incoming request is a candidate for
2543 * high-priority processing. If it is, assign it a high
2544 * priority operations table.
2546 * \param[in] tsi target session environment for this request
2548 static void ofd_hp_brw(struct tgt_session_info *tsi)
2550 struct niobuf_remote *rnb;
2551 struct obd_ioobj *ioo;
2555 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2556 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2557 if (ioo->ioo_bufcnt > 0) {
2558 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2559 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2561 /* no high priority if server lock is needed */
2562 if (rnb->rnb_flags & OBD_BRW_SRVLOCK)
2565 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2569 * Assign high priority operations to an punch request.
2571 * Check if the incoming request is a candidate for
2572 * high-priority processing. If it is, assign it a high
2573 * priority operations table.
2575 * \param[in] tsi target session environment for this request
2577 static void ofd_hp_punch(struct tgt_session_info *tsi)
2579 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2580 /* no high-priority if server lock is needed */
2581 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2582 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK)
2584 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2587 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2588 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2589 #define OST_BRW_READ OST_READ
2590 #define OST_BRW_WRITE OST_WRITE
2593 * Table of OFD-specific request handlers
2595 * This table contains all opcodes accepted by OFD and
2596 * specifies handlers for them. The tgt_request_handler()
2597 * uses such table from each target to process incoming
2600 static struct tgt_handler ofd_tgt_handlers[] = {
2601 TGT_RPC_HANDLER(OST_FIRST_OPC,
2602 0, OST_CONNECT, tgt_connect,
2603 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2604 TGT_RPC_HANDLER(OST_FIRST_OPC,
2605 0, OST_DISCONNECT, tgt_disconnect,
2606 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2607 TGT_RPC_HANDLER(OST_FIRST_OPC,
2608 0, OST_SET_INFO, ofd_set_info_hdl,
2609 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2610 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2611 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
2612 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2613 OST_SETATTR, ofd_setattr_hdl),
2614 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2615 OST_CREATE, ofd_create_hdl),
2616 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2617 OST_DESTROY, ofd_destroy_hdl),
2618 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
2619 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2620 OST_BRW_READ, tgt_brw_read,
2622 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2623 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
2625 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2626 OST_PUNCH, ofd_punch_hdl,
2628 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
2629 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
2632 static struct tgt_opc_slice ofd_common_slice[] = {
2634 .tos_opc_start = OST_FIRST_OPC,
2635 .tos_opc_end = OST_LAST_OPC,
2636 .tos_hs = ofd_tgt_handlers
2639 .tos_opc_start = OBD_FIRST_OPC,
2640 .tos_opc_end = OBD_LAST_OPC,
2641 .tos_hs = tgt_obd_handlers
2644 .tos_opc_start = LDLM_FIRST_OPC,
2645 .tos_opc_end = LDLM_LAST_OPC,
2646 .tos_hs = tgt_dlm_handlers
2649 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2650 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2651 .tos_hs = tgt_out_handlers
2654 .tos_opc_start = SEQ_FIRST_OPC,
2655 .tos_opc_end = SEQ_LAST_OPC,
2656 .tos_hs = seq_handlers
2659 .tos_opc_start = LFSCK_FIRST_OPC,
2660 .tos_opc_end = LFSCK_LAST_OPC,
2661 .tos_hs = tgt_lfsck_handlers
2664 .tos_opc_start = SEC_FIRST_OPC,
2665 .tos_opc_end = SEC_LAST_OPC,
2666 .tos_hs = tgt_sec_ctx_handlers
2673 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2674 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2677 * Implementation of lu_context_key::lct_key_exit.
2679 * Optional method called on lu_context_exit() for all allocated
2681 * It is used in OFD to sanitize context values which may be re-used
2682 * during another request processing by the same thread.
2684 * \param[in] ctx execution context
2685 * \param[in] key context key
2686 * \param[in] data ofd_thread_info
2688 static void ofd_key_exit(const struct lu_context *ctx,
2689 struct lu_context_key *key, void *data)
2691 struct ofd_thread_info *info = data;
2693 info->fti_env = NULL;
2694 info->fti_exp = NULL;
2697 info->fti_pre_version = 0;
2700 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2703 struct lu_context_key ofd_thread_key = {
2704 .lct_tags = LCT_DT_THREAD,
2705 .lct_init = ofd_key_init,
2706 .lct_fini = ofd_key_fini,
2707 .lct_exit = ofd_key_exit
2711 * Initialize OFD device according to parameters in the config log \a cfg.
2713 * This is the main starting point of OFD initialization. It fills all OFD
2714 * parameters with their initial values and calls other initializing functions
2715 * to set up all OFD subsystems.
2717 * \param[in] env execution environment
2718 * \param[in] m OFD device
2719 * \param[in] ldt LU device type of OFD
2720 * \param[in] cfg configuration log
2722 * \retval 0 if successful
2723 * \retval negative value on error
2725 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2726 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2728 const char *dev = lustre_cfg_string(cfg, 0);
2729 struct ofd_thread_info *info = NULL;
2730 struct obd_device *obd;
2731 struct obd_statfs *osfs;
2736 obd = class_name2obd(dev);
2738 CERROR("Cannot find obd with name %s\n", dev);
2742 rc = lu_env_refill((struct lu_env *)env);
2746 obd->u.obt.obt_magic = OBT_MAGIC;
2748 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2749 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2751 spin_lock_init(&m->ofd_flags_lock);
2752 m->ofd_raid_degraded = 0;
2753 m->ofd_syncjournal = 0;
2755 m->ofd_grant_compat_disable = 0;
2756 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2759 spin_lock_init(&m->ofd_osfs_lock);
2760 m->ofd_osfs_age = cfs_time_shift_64(-1000);
2761 m->ofd_osfs_unstable = 0;
2762 m->ofd_statfs_inflight = 0;
2763 m->ofd_osfs_inflight = 0;
2766 spin_lock_init(&m->ofd_grant_lock);
2767 m->ofd_tot_dirty = 0;
2768 m->ofd_tot_granted = 0;
2769 m->ofd_tot_pending = 0;
2770 m->ofd_seq_count = 0;
2771 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2772 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2773 spin_lock_init(&m->ofd_inconsistency_lock);
2775 spin_lock_init(&m->ofd_batch_lock);
2776 init_rwsem(&m->ofd_lastid_rwsem);
2778 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2779 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2780 /* set this lu_device to obd, because error handling need it */
2781 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2783 rc = ofd_procfs_init(m);
2785 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2789 /* No connection accepted until configurations will finish */
2790 spin_lock(&obd->obd_dev_lock);
2791 obd->obd_no_conn = 1;
2792 spin_unlock(&obd->obd_dev_lock);
2793 obd->obd_replayable = 1;
2794 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2795 char *str = lustre_cfg_string(cfg, 4);
2797 if (strchr(str, 'n')) {
2798 CWARN("%s: recovery disabled\n", obd->obd_name);
2799 obd->obd_replayable = 0;
2803 info = ofd_info_init(env, NULL);
2805 GOTO(err_fini_proc, rc = -EFAULT);
2807 rc = ofd_stack_init(env, m, cfg);
2809 CERROR("Can't init device stack, rc %d\n", rc);
2810 GOTO(err_fini_proc, rc);
2813 ofd_procfs_add_brw_stats_symlink(m);
2815 /* populate cached statfs data */
2816 osfs = &ofd_info(env)->fti_u.osfs;
2817 rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
2819 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
2820 GOTO(err_fini_stack, rc);
2822 if (!IS_PO2(osfs->os_bsize)) {
2823 CERROR("%s: blocksize (%d) is not a power of 2\n",
2824 obd->obd_name, osfs->os_bsize);
2825 GOTO(err_fini_stack, rc = -EPROTO);
2827 m->ofd_blockbits = fls(osfs->os_bsize) - 1;
2829 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2830 if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
2831 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2833 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2834 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2835 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2836 LDLM_NAMESPACE_SERVER,
2837 LDLM_NAMESPACE_GREEDY,
2839 if (m->ofd_namespace == NULL)
2840 GOTO(err_fini_stack, rc = -ENOMEM);
2841 /* set obd_namespace for compatibility with old code */
2842 obd->obd_namespace = m->ofd_namespace;
2843 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2844 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2845 m->ofd_namespace->ns_lvbp = m;
2847 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2848 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2850 dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
2852 /* Allow at most ddp_grant_reserved% of the available filesystem space
2853 * to be granted to clients, so that any errors in the grant overhead
2854 * calculations do not allow granting more space to clients than can be
2855 * written. Assumes that in aggregate the grant overhead calculations do
2856 * not have more than ddp_grant_reserved% estimation error in them. */
2857 m->ofd_grant_ratio =
2858 ofd_grant_ratio_conv(m->ofd_dt_conf.ddp_grant_reserved);
2860 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2861 OBD_FAIL_OST_ALL_REQUEST_NET,
2862 OBD_FAIL_OST_ALL_REPLY_NET);
2864 GOTO(err_free_ns, rc);
2866 rc = ofd_fs_setup(env, m, obd);
2868 GOTO(err_fini_lut, rc);
2870 rc = ofd_start_inconsistency_verification_thread(m);
2872 GOTO(err_fini_fs, rc);
2874 tgt_adapt_sptlrpc_conf(&m->ofd_lut, 1);
2879 ofd_fs_cleanup(env, m);
2881 tgt_fini(env, &m->ofd_lut);
2883 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
2884 obd->obd_namespace = m->ofd_namespace = NULL;
2886 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
2893 * Stop the OFD device
2895 * This function stops the OFD device and all its subsystems.
2896 * This is the end of OFD lifecycle.
2898 * \param[in] env execution environment
2899 * \param[in] m OFD device
2901 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
2903 struct obd_device *obd = ofd_obd(m);
2904 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
2905 struct lfsck_stop stop;
2907 stop.ls_status = LS_PAUSED;
2909 lfsck_stop(env, m->ofd_osd, &stop);
2910 target_recovery_fini(obd);
2911 if (m->ofd_namespace != NULL)
2912 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
2913 d->ld_obd->obd_force);
2915 obd_exports_barrier(obd);
2916 obd_zombie_barrier();
2918 tgt_fini(env, &m->ofd_lut);
2919 ofd_stop_inconsistency_verification_thread(m);
2920 lfsck_degister(env, m->ofd_osd);
2921 ofd_fs_cleanup(env, m);
2923 if (m->ofd_namespace != NULL) {
2924 ldlm_namespace_free_post(m->ofd_namespace);
2925 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
2928 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
2930 LASSERT(atomic_read(&d->ld_ref) == 0);
2931 server_put_mount(obd->obd_name, true);
2936 * Implementation of lu_device_type_operations::ldto_device_fini.
2938 * Finalize device. Dual to ofd_device_init(). It is called from
2939 * obd_precleanup() and stops the current device.
2941 * \param[in] env execution environment
2942 * \param[in] d LU device of OFD
2946 static struct lu_device *ofd_device_fini(const struct lu_env *env,
2947 struct lu_device *d)
2950 ofd_fini(env, ofd_dev(d));
2955 * Implementation of lu_device_type_operations::ldto_device_free.
2957 * Free OFD device. Dual to ofd_device_alloc().
2959 * \param[in] env execution environment
2960 * \param[in] d LU device of OFD
2964 static struct lu_device *ofd_device_free(const struct lu_env *env,
2965 struct lu_device *d)
2967 struct ofd_device *m = ofd_dev(d);
2969 dt_device_fini(&m->ofd_dt_dev);
2975 * Implementation of lu_device_type_operations::ldto_device_alloc.
2977 * This function allocates the new OFD device. It is called from
2978 * obd_setup() if OBD device had lu_device_type defined.
2980 * \param[in] env execution environment
2981 * \param[in] t lu_device_type of OFD device
2982 * \param[in] cfg configuration log
2984 * \retval pointer to the lu_device of just allocated OFD
2985 * \retval ERR_PTR of return value on error
2987 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
2988 struct lu_device_type *t,
2989 struct lustre_cfg *cfg)
2991 struct ofd_device *m;
2992 struct lu_device *l;
2997 return ERR_PTR(-ENOMEM);
2999 l = &m->ofd_dt_dev.dd_lu_dev;
3000 dt_device_init(&m->ofd_dt_dev, t);
3001 rc = ofd_init0(env, m, t, cfg);
3003 ofd_device_free(env, l);
3010 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3011 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3013 static struct lu_device_type_operations ofd_device_type_ops = {
3014 .ldto_init = ofd_type_init,
3015 .ldto_fini = ofd_type_fini,
3017 .ldto_start = ofd_type_start,
3018 .ldto_stop = ofd_type_stop,
3020 .ldto_device_alloc = ofd_device_alloc,
3021 .ldto_device_free = ofd_device_free,
3022 .ldto_device_fini = ofd_device_fini
3025 static struct lu_device_type ofd_device_type = {
3026 .ldt_tags = LU_DEVICE_DT,
3027 .ldt_name = LUSTRE_OST_NAME,
3028 .ldt_ops = &ofd_device_type_ops,
3029 .ldt_ctx_tags = LCT_DT_THREAD
3033 * Initialize OFD module.
3035 * This function is called upon module loading. It registers OFD device type
3036 * and prepares all in-memory structures used by all OFD devices.
3038 * \retval 0 if successful
3039 * \retval negative value on error
3041 static int __init ofd_init(void)
3045 rc = lu_kmem_init(ofd_caches);
3049 rc = ofd_fmd_init();
3051 lu_kmem_fini(ofd_caches);
3055 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3056 LUSTRE_OST_NAME, &ofd_device_type);
3063 * This function is called upon OFD module unloading.
3064 * It frees all related structures and unregisters OFD device type.
3066 static void __exit ofd_exit(void)
3069 lu_kmem_fini(ofd_caches);
3070 class_unregister_type(LUSTRE_OST_NAME);
3073 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3074 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3075 MODULE_VERSION(LUSTRE_VERSION_STRING);
3076 MODULE_LICENSE("GPL");
3078 module_init(ofd_init);
3079 module_exit(ofd_exit);