4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/ofd/ofd_dev.c
34 * This file contains OSD API methods for OBD Filter Device (OFD),
35 * request handlers and supplemental functions to set OFD up and clean it up.
37 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
38 * Author: Mike Pershin <mike.pershin@intel.com>
39 * Author: Johann Lombardi <johann.lombardi@intel.com>
42 * The OBD Filter Device (OFD) module belongs to the Object Storage
43 * Server stack and connects the RPC oriented Unified Target (TGT)
44 * layer (see lustre/include/lu_target.h) to the storage oriented OSD
45 * layer (see Documentation/osd-api.txt).
53 * OFD implements the LU and OBD device APIs and is responsible for:
55 * - Handling client requests (create, destroy, bulk IO, setattr,
56 * get_info, set_info, statfs) for the objects belonging to the OST
57 * (together with TGT).
59 * - Providing grant space management which allows clients to reserve
60 * disk space for data writeback. OFD tracks grants on global and
63 * - Handling object precreation requests from MDTs.
65 * - Operating the LDLM service that allows clients to maintain object
66 * data cache coherence.
69 #define DEBUG_SUBSYSTEM S_FILTER
71 #include <obd_class.h>
72 #include <obd_cksum.h>
73 #include <uapi/linux/lustre/lustre_param.h>
74 #include <lustre_fid.h>
75 #include <lustre_lfsck.h>
76 #include <lustre_dlm.h>
77 #include <lustre_quota.h>
78 #include <lustre_nodemap.h>
79 #include <lustre_log.h>
81 #include "ofd_internal.h"
83 /* Slab for OFD object allocation */
84 static struct kmem_cache *ofd_object_kmem;
86 static struct lu_kmem_descr ofd_caches[] = {
88 .ckd_cache = &ofd_object_kmem,
89 .ckd_name = "ofd_obj",
90 .ckd_size = sizeof(struct ofd_object)
98 * Connect OFD to the next device in the stack.
100 * This function is used for device stack configuration and links OFD
101 * device with bottom OSD device.
103 * \param[in] env execution environment
104 * \param[in] m OFD device
105 * \param[in] next name of next device in the stack
106 * \param[out] exp export to return
108 * \retval 0 and export in \a exp if successful
109 * \retval negative value on error
111 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
112 const char *next, struct obd_export **exp)
114 struct obd_connect_data *data = NULL;
115 struct obd_device *obd;
121 GOTO(out, rc = -ENOMEM);
123 obd = class_name2obd(next);
125 CERROR("%s: can't locate next device: %s\n",
127 GOTO(out, rc = -ENOTCONN);
130 data->ocd_connect_flags = OBD_CONNECT_VERSION;
131 data->ocd_version = LUSTRE_VERSION_CODE;
133 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
135 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
136 ofd_name(m), next, rc);
140 m->ofd_dt_dev.dd_lu_dev.ld_site =
141 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
142 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
143 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
144 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
153 * Initialize stack of devices.
155 * This function initializes OFD-OSD device stack to serve OST requests
157 * \param[in] env execution environment
158 * \param[in] m OFD device
159 * \param[in] cfg Lustre config for this server
161 * \retval 0 if successful
162 * \retval negative value on error
164 static int ofd_stack_init(const struct lu_env *env,
165 struct ofd_device *m, struct lustre_cfg *cfg)
167 const char *dev = lustre_cfg_string(cfg, 0);
169 struct ofd_thread_info *info = ofd_info(env);
170 struct lustre_mount_info *lmi;
171 struct lustre_mount_data *lmd;
177 lmi = server_get_mount(dev);
179 CERROR("Cannot get mount info for %s!\n", dev);
183 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
184 if (lmd != NULL && lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
185 m->ofd_skip_lfsck = 1;
187 /* find bottom osd */
188 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
192 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
193 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
194 OBD_FREE(osdname, MTI_NAME_MAXLEN);
198 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
200 m->ofd_osd = lu2dt_dev(d);
202 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
203 "%s-osd", lustre_cfg_string(cfg, 0));
209 * Finalize the device stack OFD-OSD.
211 * This function cleans OFD-OSD device stack and
212 * disconnects OFD from the OSD.
214 * \param[in] env execution environment
215 * \param[in] m OFD device
216 * \param[in] top top device of stack
218 * \retval 0 if successful
219 * \retval negative value on error
221 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
222 struct lu_device *top)
224 struct obd_device *obd = ofd_obd(m);
225 struct lustre_cfg_bufs bufs;
226 struct lustre_cfg *lcfg;
231 lu_site_purge(env, top->ld_site, ~0);
232 /* process cleanup, pass mdt obd name to get obd umount flags */
233 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
238 lustre_cfg_bufs_set_string(&bufs, 1, flags);
239 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
242 lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs);
245 top->ld_ops->ldo_process_config(env, top, lcfg);
246 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
248 if (m->ofd_los != NULL) {
249 local_oid_storage_fini(env, m->ofd_los);
253 lu_site_purge(env, top->ld_site, ~0);
254 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
255 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
256 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
259 LASSERT(m->ofd_osd_exp);
260 obd_disconnect(m->ofd_osd_exp);
265 static void ofd_stack_pre_fini(const struct lu_env *env, struct ofd_device *m,
266 struct lu_device *top)
268 struct lustre_cfg_bufs bufs;
269 struct lustre_cfg *lcfg;
274 lustre_cfg_bufs_reset(&bufs, ofd_name(m));
275 lustre_cfg_bufs_set_string(&bufs, 1, NULL);
276 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
278 CERROR("%s: failed to trigger LCFG_PRE_CLEANUP\n", ofd_name(m));
280 lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, &bufs);
281 top->ld_ops->ldo_process_config(env, top, lcfg);
282 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
283 lcfg->lcfg_buflens));
289 /* For interoperability, see mdt_interop_param[]. */
290 static struct cfg_interop_param ofd_interop_param[] = {
291 { "ost.quota_type", NULL },
296 * Check if parameters are symlinks to the OSD.
298 * Some parameters were moved from ofd to osd and only their
299 * symlinks were kept in ofd by LU-3106. They are:
300 * -writehthrough_cache_enable
301 * -readcache_max_filesize
305 * Since they are not included by the static lprocfs var list, a pre-check
306 * is added for them to avoid "unknown param" errors. If they are matched
307 * in this check, they will be passed to the OSD directly.
309 * \param[in] param parameters to check
311 * \retval true if param is symlink to OSD param
314 static bool match_symlink_param(char *param)
319 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
320 sval = strchr(param, '=');
322 paramlen = sval - param;
323 if (strncmp(param, "writethrough_cache_enable",
325 strncmp(param, "readcache_max_filesize",
327 strncmp(param, "read_cache_enable",
329 strncmp(param, "brw_stats", paramlen) == 0)
338 * Process various configuration parameters.
340 * This function is used by MGS to process specific configurations and
341 * pass them through to the next device in server stack, i.e. the OSD.
343 * \param[in] env execution environment
344 * \param[in] d LU device of OFD
345 * \param[in] cfg parameters to process
347 * \retval 0 if successful
348 * \retval negative value on error
350 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
351 struct lustre_cfg *cfg)
353 struct ofd_device *m = ofd_dev(d);
354 struct dt_device *dt_next = m->ofd_osd;
355 struct lu_device *next = &dt_next->dd_lu_dev;
360 switch (cfg->lcfg_command) {
362 /* For interoperability */
363 struct cfg_interop_param *ptr = NULL;
364 struct lustre_cfg *old_cfg = NULL;
368 param = lustre_cfg_string(cfg, 1);
370 CERROR("param is empty\n");
375 ptr = class_find_old_param(param, ofd_interop_param);
377 if (ptr->new_param == NULL) {
379 CWARN("For interoperability, skip this %s."
380 " It is obsolete.\n", ptr->old_param);
384 CWARN("Found old param %s, changed it to %s.\n",
385 ptr->old_param, ptr->new_param);
388 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
395 if (match_symlink_param(param)) {
396 rc = next->ld_ops->ldo_process_config(env, next, cfg);
400 count = class_modify_config(cfg, PARAM_OST,
401 &d->ld_obd->obd_kset.kobj);
406 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
408 /* we don't understand; pass it on */
409 rc = next->ld_ops->ldo_process_config(env, next, cfg);
412 case LCFG_SPTLRPC_CONF: {
417 /* others are passed further */
418 rc = next->ld_ops->ldo_process_config(env, next, cfg);
425 * Implementation of lu_object_operations::loo_object_init for OFD
427 * Allocate just the next object (OSD) in stack.
429 * \param[in] env execution environment
430 * \param[in] o lu_object of OFD object
431 * \param[in] conf additional configuration parameters, not used here
433 * \retval 0 if successful
434 * \retval negative value on error
436 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
437 const struct lu_object_conf *conf)
439 struct ofd_device *d = ofd_dev(o->lo_dev);
440 struct lu_device *under;
441 struct lu_object *below;
446 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
447 PFID(lu_object_fid(o)));
449 under = &d->ofd_osd->dd_lu_dev;
450 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
452 lu_object_add(o, below);
460 * Implementation of lu_object_operations::loo_object_free.
462 * Finish OFD object lifecycle and free its memory.
464 * \param[in] env execution environment
465 * \param[in] o LU object of OFD object
467 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
469 struct ofd_object *of = ofd_obj(o);
470 struct lu_object_header *h;
475 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
476 PFID(lu_object_fid(o)));
479 lu_object_header_fini(h);
480 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
485 * Implementation of lu_object_operations::loo_object_print.
487 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
488 * LU_OBJECT_DEBUG() for more details about the compound object printing.
490 * \param[in] env execution environment
491 * \param[in] cookie opaque data passed to the printer function
492 * \param[in] p printer function to use
493 * \param[in] o LU object of OFD object
495 * \retval 0 if successful
496 * \retval negative value on error
498 static int ofd_object_print(const struct lu_env *env, void *cookie,
499 lu_printer_t p, const struct lu_object *o)
501 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
504 static struct lu_object_operations ofd_obj_ops = {
505 .loo_object_init = ofd_object_init,
506 .loo_object_free = ofd_object_free,
507 .loo_object_print = ofd_object_print
511 * Implementation of lu_device_operations::lod_object_alloc.
513 * This function allocates OFD part of compound OFD-OSD object and
514 * initializes its header, because OFD is the top device in stack
516 * \param[in] env execution environment
517 * \param[in] hdr object header, NULL for OFD
518 * \param[in] d lu_device
520 * \retval allocated object if successful
521 * \retval NULL value on failed allocation
523 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
524 const struct lu_object_header *hdr,
527 struct ofd_object *of;
531 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
534 struct lu_object_header *h;
536 o = &of->ofo_obj.do_lu;
538 lu_object_header_init(h);
539 lu_object_init(o, h, d);
540 lu_object_add_top(h, o);
541 o->lo_ops = &ofd_obj_ops;
549 * Return the result of LFSCK run to the OFD.
551 * Notify OFD about result of LFSCK run. That may block the new object
552 * creation until problem is fixed by LFSCK.
554 * \param[in] env execution environment
555 * \param[in] data pointer to the OFD device
556 * \param[in] event LFSCK event type
558 * \retval 0 if successful
559 * \retval negative value on unknown event
561 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
562 enum lfsck_events event)
564 struct ofd_device *ofd = data;
565 struct obd_device *obd = ofd_obd(ofd);
568 case LE_LASTID_REBUILDING:
569 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
570 "on the device until the LAST_ID rebuilt successfully.\n",
572 down_write(&ofd->ofd_lastid_rwsem);
573 ofd->ofd_lastid_rebuilding = 1;
574 up_write(&ofd->ofd_lastid_rwsem);
576 case LE_LASTID_REBUILT: {
577 down_write(&ofd->ofd_lastid_rwsem);
578 ofd_seqs_free(env, ofd);
579 ofd->ofd_lastid_rebuilding = 0;
580 ofd->ofd_lastid_gen++;
581 up_write(&ofd->ofd_lastid_rwsem);
582 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
587 CERROR("%s: unknown lfsck event: rc = %d\n",
588 ofd_name(ofd), event);
596 * Implementation of lu_device_operations::ldo_prepare.
598 * This method is called after layer has been initialized and before it starts
599 * serving user requests. In OFD it starts lfsk check routines and initializes
602 * \param[in] env execution environment
603 * \param[in] pdev higher device in stack, NULL for OFD
604 * \param[in] dev lu_device of OFD device
606 * \retval 0 if successful
607 * \retval negative value on error
609 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
610 struct lu_device *dev)
612 struct ofd_thread_info *info;
613 struct ofd_device *ofd = ofd_dev(dev);
614 struct obd_device *obd = ofd_obd(ofd);
615 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
620 info = ofd_info_init(env, NULL);
624 /* initialize lower device */
625 rc = next->ld_ops->ldo_prepare(env, dev, next);
629 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
630 ofd_lfsck_out_notify, ofd, false);
632 CERROR("%s: failed to initialize lfsck: rc = %d\n",
637 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
638 /* The LFSCK instance is registered just now, so it must be there when
639 * register the namespace to such instance. */
640 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
642 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
643 LASSERT(obd->obd_no_conn);
644 spin_lock(&obd->obd_dev_lock);
645 obd->obd_no_conn = 0;
646 spin_unlock(&obd->obd_dev_lock);
648 if (obd->obd_recovering == 0)
649 ofd_postrecov(env, ofd);
655 * Implementation of lu_device_operations::ldo_recovery_complete.
657 * This method notifies all layers about 'recovery complete' event. That means
658 * device is in full state and consistent. An OFD calculates available grant
659 * space upon this event.
661 * \param[in] env execution environment
662 * \param[in] dev lu_device of OFD device
664 * \retval 0 if successful
665 * \retval negative value on error
667 static int ofd_recovery_complete(const struct lu_env *env,
668 struct lu_device *dev)
670 struct ofd_thread_info *oti = ofd_info(env);
671 struct ofd_device *ofd = ofd_dev(dev);
672 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
678 * Grant space for object precreation on the self export.
679 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
680 * is enough to create 10k objects. More space is then acquired for
681 * precreation in tgt_grant_create().
683 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
684 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
685 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
686 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
687 OBD_CONNECT_GRANT_PARAM;
688 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
690 rc = next->ld_ops->ldo_recovery_complete(env, next);
695 * lu_device_operations matrix for OFD device.
697 static struct lu_device_operations ofd_lu_ops = {
698 .ldo_object_alloc = ofd_object_alloc,
699 .ldo_process_config = ofd_process_config,
700 .ldo_recovery_complete = ofd_recovery_complete,
701 .ldo_prepare = ofd_prepare,
704 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
706 * Expose OSD statistics to OFD layer.
708 * The osd interfaces to the backend file system exposes useful data
709 * such as brw_stats and read or write cache states. This same data
710 * needs to be exposed into the obdfilter (ofd) layer to maintain
711 * backwards compatibility. This function creates the symlinks in the
712 * proc layer to enable this.
714 * \param[in] ofd OFD device
716 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
718 struct obd_device *obd = ofd_obd(ofd);
719 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
721 if (obd->obd_proc_entry == NULL)
724 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
725 "../../%s/%s/brw_stats",
726 osd_obd->obd_type->typ_name, obd->obd_name);
728 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
729 "../../%s/%s/read_cache_enable",
730 osd_obd->obd_type->typ_name, obd->obd_name);
732 lprocfs_add_symlink("readcache_max_filesize",
734 "../../%s/%s/readcache_max_filesize",
735 osd_obd->obd_type->typ_name, obd->obd_name);
737 lprocfs_add_symlink("writethrough_cache_enable",
739 "../../%s/%s/writethrough_cache_enable",
740 osd_obd->obd_type->typ_name, obd->obd_name);
745 * Cleanup all procfs entries in OFD.
747 * \param[in] ofd OFD device
749 static void ofd_procfs_fini(struct ofd_device *ofd)
751 struct obd_device *obd = ofd_obd(ofd);
753 lprocfs_free_per_client_stats(obd);
754 lprocfs_obd_cleanup(obd);
755 lprocfs_free_obd_stats(obd);
756 lprocfs_job_stats_fini(obd);
760 * Stop SEQ/FID server on OFD.
762 * \param[in] env execution environment
763 * \param[in] ofd OFD device
765 * \retval 0 if successful
766 * \retval negative value on error
768 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
770 return seq_site_fini(env, &ofd->ofd_seq_site);
774 * Start SEQ/FID server on OFD.
776 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
777 * It also connects to the master server to get own FID sequence (SEQ) range
778 * to this particular OFD. Typically that happens when the OST is first
779 * formatted or in the rare case that it exhausts the local sequence range.
781 * The sequence range is allocated out to the MDTs for OST object allocations,
782 * and not directly to the clients.
784 * \param[in] env execution environment
785 * \param[in] ofd OFD device
787 * \retval 0 if successful
788 * \retval negative value on error
790 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
792 struct seq_server_site *ss = &ofd->ofd_seq_site;
793 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
794 char *obd_name = ofd_name(ofd);
798 ss = &ofd->ofd_seq_site;
799 lu->ld_site->ld_seq_site = ss;
800 ss->ss_lu = lu->ld_site;
801 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
803 OBD_ALLOC(name, sizeof(obd_name) * 2 + 10);
807 OBD_ALLOC_PTR(ss->ss_server_seq);
808 if (ss->ss_server_seq == NULL)
809 GOTO(out_name, rc = -ENOMEM);
811 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
812 LUSTRE_SEQ_SERVER, ss);
814 CERROR("%s : seq server init error %d\n", obd_name, rc);
815 GOTO(out_server, rc);
817 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
819 OBD_ALLOC_PTR(ss->ss_client_seq);
820 if (ss->ss_client_seq == NULL)
821 GOTO(out_server, rc = -ENOMEM);
824 * It always printed as "%p", so that the name is unique in the kernel,
825 * even if the filesystem is mounted twice. So sizeof(.) * 2 is enough.
827 snprintf(name, sizeof(obd_name) * 2 + 7, "%p-super", obd_name);
828 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
831 CERROR("%s : seq client init error %d\n", obd_name, rc);
832 GOTO(out_client, rc);
835 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
839 seq_client_fini(ss->ss_client_seq);
840 OBD_FREE_PTR(ss->ss_client_seq);
841 ss->ss_client_seq = NULL;
843 seq_server_fini(ss->ss_server_seq, env);
844 OBD_FREE_PTR(ss->ss_server_seq);
845 ss->ss_server_seq = NULL;
848 OBD_FREE(name, sizeof(obd_name) * 2 + 10);
854 * OFD request handler for OST_SET_INFO RPC.
856 * This is OFD-specific part of request handling
858 * \param[in] tsi target session environment for this request
860 * \retval 0 if successful
861 * \retval negative value on error
863 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
865 struct ptlrpc_request *req = tgt_ses_req(tsi);
866 struct ost_body *body = NULL, *repbody;
867 void *key, *val = NULL;
868 int keylen, vallen, rc = 0;
869 bool is_grant_shrink;
873 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
875 DEBUG_REQ(D_HA, req, "no set_info key");
876 RETURN(err_serious(-EFAULT));
878 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
881 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
883 DEBUG_REQ(D_HA, req, "no set_info val");
884 RETURN(err_serious(-EFAULT));
886 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
889 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
891 /* In this case the value is actually an RMF_OST_BODY, so we
892 * transmutate the type of this PTLRPC */
893 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
895 rc = req_capsule_server_pack(tsi->tsi_pill);
899 if (is_grant_shrink) {
900 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
902 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
905 /** handle grant shrink, similar to a read request */
906 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
908 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
910 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
913 CERROR("%s: Unsupported key %s\n",
914 tgt_name(tsi->tsi_tgt), (char *)key);
917 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
924 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
926 * This function returns a list of extents which describes how a file's
927 * blocks are laid out on the disk.
929 * \param[in] env execution environment
930 * \param[in] ofd OFD device
931 * \param[in] fid FID of object
932 * \param[in] fiemap fiemap structure to fill with data
934 * \retval 0 if \a fiemap is filled with data successfully
935 * \retval negative value on error
937 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
938 struct lu_fid *fid, struct fiemap *fiemap)
940 struct ofd_object *fo;
943 fo = ofd_object_find(env, ofd, fid);
945 CERROR("%s: error finding object "DFID"\n",
946 ofd_name(ofd), PFID(fid));
950 ofd_read_lock(env, fo);
951 if (ofd_object_exists(fo))
952 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
955 ofd_read_unlock(env, fo);
956 ofd_object_put(env, fo);
961 static int ofd_lock_unlock_region(const struct lu_env *env,
962 struct ldlm_namespace *ns,
963 struct ldlm_res_id *res_id,
964 unsigned long long begin,
965 unsigned long long end)
969 struct lustre_handle lh = { 0 };
971 LASSERT(begin <= end);
973 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
977 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
978 tgt_extent_unlock(&lh, LCK_PR);
984 * Lock the sparse areas of given resource.
986 * The locking of sparse areas will cause dirty data to be flushed back from
987 * clients. This is used when getting the FIEMAP of an object to make sure
988 * there is no unaccounted cached data on clients.
990 * This function goes through \a fiemap list of extents and locks only sparse
991 * areas between extents.
993 * \param[in] ns LDLM namespace
994 * \param[in] res_id resource ID
995 * \param[in] fiemap file extents mapping on disk
996 * \param[in] locked list head of regions list
998 * \retval 0 if successful
999 * \retval negative value on error
1001 static int lock_zero_regions(const struct lu_env *env,
1002 struct ldlm_namespace *ns,
1003 struct ldlm_res_id *res_id,
1004 struct fiemap *fiemap)
1006 __u64 begin = fiemap->fm_start;
1009 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1013 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1014 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1015 if (fiemap_start[i].fe_logical > begin) {
1016 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1017 begin, fiemap_start[i].fe_logical);
1018 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1019 fiemap_start[i].fe_logical);
1024 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1027 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1028 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1029 begin, fiemap->fm_start + fiemap->fm_length);
1030 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1031 fiemap->fm_start + fiemap->fm_length);
1039 * OFD request handler for OST_GET_INFO RPC.
1041 * This is OFD-specific part of request handling. The OFD-specific keys are:
1042 * - KEY_LAST_ID (obsolete)
1046 * This function reads needed data from storage and fills reply with it.
1048 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1049 * and is kept for compatibility.
1051 * \param[in] tsi target session environment for this request
1053 * \retval 0 if successful
1054 * \retval negative value on error
1056 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1058 struct obd_export *exp = tsi->tsi_exp;
1059 struct ofd_device *ofd = ofd_exp(exp);
1060 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1063 int replylen, rc = 0;
1067 /* this common part for get_info rpc */
1068 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1070 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1071 RETURN(err_serious(-EPROTO));
1073 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1076 if (KEY_IS(KEY_LAST_ID)) {
1078 struct ofd_seq *oseq;
1080 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1081 rc = req_capsule_server_pack(tsi->tsi_pill);
1083 RETURN(err_serious(rc));
1085 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1087 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1088 (u64)exp->exp_filter_data.fed_group);
1092 *last_id = ofd_seq_last_oid(oseq);
1093 ofd_seq_put(tsi->tsi_env, oseq);
1094 } else if (KEY_IS(KEY_FIEMAP)) {
1095 struct ll_fiemap_info_key *fm_key;
1096 struct fiemap *fiemap;
1099 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1101 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1102 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1104 RETURN(err_serious(rc));
1106 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1108 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1110 replylen = fiemap_count_to_size(
1111 fm_key->lfik_fiemap.fm_extent_count);
1112 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1113 RCL_SERVER, replylen);
1115 rc = req_capsule_server_pack(tsi->tsi_pill);
1117 RETURN(err_serious(rc));
1119 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1123 *fiemap = fm_key->lfik_fiemap;
1124 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1126 /* LU-3219: Lock the sparse areas to make sure dirty
1127 * flushed back from client, then call fiemap again. */
1128 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1129 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1130 ost_fid_build_resid(fid, &fti->fti_resid);
1131 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1132 &fti->fti_resid, fiemap);
1134 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1137 } else if (KEY_IS(KEY_LAST_FID)) {
1138 struct ofd_device *ofd = ofd_exp(exp);
1139 struct ofd_seq *oseq;
1143 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1144 rc = req_capsule_server_pack(tsi->tsi_pill);
1146 RETURN(err_serious(rc));
1148 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1150 RETURN(err_serious(-EPROTO));
1152 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1154 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1158 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1159 ostid_seq(&fti->fti_ostid));
1161 RETURN(PTR_ERR(oseq));
1163 rc = ostid_to_fid(fid, &oseq->os_oi,
1164 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1168 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1171 ofd_seq_put(tsi->tsi_env, oseq);
1173 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1177 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1184 * OFD request handler for OST_GETATTR RPC.
1186 * This is OFD-specific part of request handling. It finds the OFD object
1187 * by its FID, gets attributes from storage and packs result to the reply.
1189 * \param[in] tsi target session environment for this request
1191 * \retval 0 if successful
1192 * \retval negative value on error
1194 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1196 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1197 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1198 struct ost_body *repbody;
1199 struct lustre_handle lh = { 0 };
1200 struct ofd_object *fo;
1202 enum ldlm_mode lock_mode = LCK_PR;
1207 LASSERT(tsi->tsi_ost_body != NULL);
1209 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1210 if (repbody == NULL)
1213 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1214 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1216 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1217 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1220 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1223 rc = tgt_extent_lock(tsi->tsi_env,
1224 tsi->tsi_tgt->lut_obd->obd_namespace,
1225 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1231 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1233 GOTO(out, rc = PTR_ERR(fo));
1235 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1239 obdo_from_la(&repbody->oa, &fti->fti_attr,
1240 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1242 /* Store object version in reply */
1243 curr_version = dt_version_get(tsi->tsi_env,
1244 ofd_object_child(fo));
1245 if ((__s64)curr_version != -EOPNOTSUPP) {
1246 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1247 repbody->oa.o_data_version = curr_version;
1250 if (fo->ofo_ff.ff_layout_version > 0) {
1251 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1252 repbody->oa.o_layout_version =
1253 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1255 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1256 PFID(&tsi->tsi_fid),
1257 repbody->oa.o_layout_version);
1261 ofd_object_put(tsi->tsi_env, fo);
1264 tgt_extent_unlock(&lh, lock_mode);
1266 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1269 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1270 repbody->oa.o_flags = OBD_FL_FLUSH;
1276 * OFD request handler for OST_SETATTR RPC.
1278 * This is OFD-specific part of request handling. It finds the OFD object
1279 * by its FID, sets attributes from request and packs result to the reply.
1281 * \param[in] tsi target session environment for this request
1283 * \retval 0 if successful
1284 * \retval negative value on error
1286 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1288 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1289 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1290 struct ost_body *body = tsi->tsi_ost_body;
1291 struct ost_body *repbody;
1292 struct ldlm_resource *res;
1293 struct ofd_object *fo;
1298 LASSERT(body != NULL);
1300 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1301 if (repbody == NULL)
1304 repbody->oa.o_oi = body->oa.o_oi;
1305 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1307 /* This would be very bad - accidentally truncating a file when
1308 * changing the time or similar - bug 12203. */
1309 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1310 body->oa.o_size != OBD_OBJECT_EOF) {
1311 static char mdsinum[48];
1313 if (body->oa.o_valid & OBD_MD_FLFID)
1314 snprintf(mdsinum, sizeof(mdsinum) - 1,
1315 "of parent "DFID, body->oa.o_parent_seq,
1316 body->oa.o_parent_oid, 0);
1320 CERROR("%s: setattr from %s is trying to truncate object "DFID
1321 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1322 PFID(&tsi->tsi_fid), mdsinum);
1326 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1328 GOTO(out, rc = PTR_ERR(fo));
1330 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1331 fti->fti_attr.la_valid &= ~LA_TYPE;
1333 /* setting objects attributes (including owner/group) */
1334 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1338 obdo_from_la(&repbody->oa, &fti->fti_attr,
1339 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1341 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1345 ofd_object_put(tsi->tsi_env, fo);
1348 /* we do not call this before to avoid lu_object_find() in
1349 * ->lvbo_update() holding another reference on the object.
1350 * otherwise concurrent destroy can make the object unavailable
1351 * for 2nd lu_object_find() waiting for the first reference
1352 * to go... deadlock! */
1353 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1354 &tsi->tsi_resid, LDLM_EXTENT, 0);
1356 ldlm_res_lvbo_update(tsi->tsi_env, res, NULL, 0);
1357 ldlm_resource_putref(res);
1364 * Destroy OST orphans.
1366 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1367 * set then we must destroy possible orphaned objects.
1369 * \param[in] env execution environment
1370 * \param[in] exp OBD export
1371 * \param[in] ofd OFD device
1372 * \param[in] oa obdo structure for reply
1374 * \retval 0 if successful
1375 * \retval negative value on error
1377 static int ofd_orphans_destroy(const struct lu_env *env,
1378 struct obd_export *exp,
1379 struct ofd_device *ofd, struct obdo *oa)
1381 struct ofd_thread_info *info = ofd_info(env);
1382 struct lu_fid *fid = &info->fti_fid;
1383 struct ost_id *oi = &oa->o_oi;
1384 struct ofd_seq *oseq;
1385 u64 seq = ostid_seq(oi);
1386 u64 end_id = ostid_id(oi);
1394 oseq = ofd_seq_get(ofd, seq);
1396 CERROR("%s: Can not find seq for "DOSTID"\n",
1397 ofd_name(ofd), POSTID(oi));
1402 last = ofd_seq_last_oid(oseq);
1405 LASSERT(exp != NULL);
1406 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1408 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1411 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1412 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1414 while (oid > end_id) {
1415 rc = fid_set_id(fid, oid);
1416 if (unlikely(rc != 0))
1419 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1420 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1421 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1422 /* this is pretty fatal... */
1423 CEMERG("%s: error destroying precreated id "
1425 ofd_name(ofd), PFID(fid), rc);
1429 ofd_seq_last_oid_set(oseq, oid);
1430 /* update last_id on disk periodically so that if we
1431 * restart * we don't need to re-scan all of the just
1432 * deleted objects. */
1433 if ((oid & 511) == 0)
1434 ofd_seq_last_oid_write(env, ofd, oseq);
1438 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1439 ofd_name(ofd), seq, oid);
1443 ofd_seq_last_oid_set(oseq, oid);
1444 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1446 /* don't reuse orphan object, return last used objid */
1447 rc = ostid_set_id(oi, last);
1453 ofd_seq_put(env, oseq);
1458 * OFD request handler for OST_CREATE RPC.
1460 * This is OFD-specific part of request handling. Its main purpose is to
1461 * create new data objects on OST, but it also used to destroy orphans.
1463 * \param[in] tsi target session environment for this request
1465 * \retval 0 if successful
1466 * \retval negative value on error
1468 static int ofd_create_hdl(struct tgt_session_info *tsi)
1470 struct ptlrpc_request *req = tgt_ses_req(tsi);
1471 struct ost_body *repbody;
1472 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1473 struct obdo *rep_oa;
1474 struct obd_export *exp = tsi->tsi_exp;
1475 struct ofd_device *ofd = ofd_exp(exp);
1476 u64 seq = ostid_seq(&oa->o_oi);
1477 u64 oid = ostid_id(&oa->o_oi);
1478 struct ofd_seq *oseq;
1486 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1489 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1490 if (repbody == NULL)
1493 down_read(&ofd->ofd_lastid_rwsem);
1494 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1495 * we may do that in the future.
1496 * Return -ENOSPC until the LAST_ID rebuilt. */
1497 if (unlikely(ofd->ofd_lastid_rebuilding))
1498 GOTO(out_sem, rc = -ENOSPC);
1500 rep_oa = &repbody->oa;
1501 rep_oa->o_oi = oa->o_oi;
1503 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1505 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1507 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1509 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1510 ofd_name(ofd), seq, PTR_ERR(oseq));
1511 GOTO(out_sem, rc = -EINVAL);
1514 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1515 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1516 if (!ofd_obd(ofd)->obd_recovering ||
1517 oid > ofd_seq_last_oid(oseq)) {
1518 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1519 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1520 ofd_seq_last_oid(oseq));
1521 GOTO(out_nolock, rc = -EINVAL);
1523 /* Do nothing here, we re-create objects during recovery
1524 * upon write replay, see ofd_preprw_write() */
1525 GOTO(out_nolock, rc = 0);
1527 /* former ofd_handle_precreate */
1528 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1529 (oa->o_flags & OBD_FL_DELORPHAN)) {
1530 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1532 /* destroy orphans */
1533 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1534 exp->exp_conn_cnt) {
1535 CERROR("%s: dropping old orphan cleanup request\n",
1537 GOTO(out_nolock, rc = 0);
1539 /* This causes inflight precreates to abort and drop lock */
1540 oseq->os_destroys_in_progress = 1;
1541 mutex_lock(&oseq->os_create_lock);
1542 if (!oseq->os_destroys_in_progress) {
1544 "%s:[%llu] destroys_in_progress already cleared\n",
1545 ofd_name(ofd), seq);
1546 rc = ostid_set_id(&rep_oa->o_oi,
1547 ofd_seq_last_oid(oseq));
1550 diff = oid - ofd_seq_last_oid(oseq);
1551 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1552 ofd_seq_last_oid(oseq), diff);
1553 if (-diff > OST_MAX_PRECREATE) {
1554 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1555 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1556 "(%llu), trust the OST\n",
1557 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1558 PFID(&oseq->os_oi.oi_fid),
1559 ofd_seq_last_oid(oseq));
1561 /* Let MDS know that we are so far ahead. */
1562 rc = ostid_set_id(&rep_oa->o_oi,
1563 ofd_seq_last_oid(oseq) + 1);
1564 } else if (diff < 0) {
1565 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1567 oseq->os_destroys_in_progress = 0;
1569 /* XXX: Used by MDS for the first time! */
1570 oseq->os_destroys_in_progress = 0;
1573 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1574 ofd->ofd_lastid_gen)) {
1575 /* Keep the export ref so we can send the reply. */
1576 ofd_obd_disconnect(class_export_get(exp));
1577 GOTO(out_nolock, rc = -ENOTCONN);
1580 mutex_lock(&oseq->os_create_lock);
1581 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1582 exp->exp_conn_cnt) {
1583 CERROR("%s: dropping old precreate request\n",
1587 /* only precreate if seq is 0, IDIF or normal and also o_id
1588 * must be specfied */
1589 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1590 !fid_seq_is_idif(seq)) || oid == 0) {
1591 diff = 1; /* shouldn't we create this right now? */
1593 diff = oid - ofd_seq_last_oid(oseq);
1594 /* Do sync create if the seq is about to used up */
1595 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1596 if (unlikely(oid >= IDIF_MAX_OID - 1))
1598 } else if (fid_seq_is_norm(seq)) {
1600 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1603 CERROR("%s : invalid o_seq "DOSTID"\n",
1604 ofd_name(ofd), POSTID(&oa->o_oi));
1605 GOTO(out, rc = -EINVAL);
1610 CERROR("%s: invalid precreate request for "
1611 DOSTID", last_id %llu. "
1612 "Likely MDS last_id corruption\n",
1613 ofd_name(ofd), POSTID(&oa->o_oi),
1614 ofd_seq_last_oid(oseq));
1615 GOTO(out, rc = -EINVAL);
1620 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1626 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1627 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1628 /* don't enforce grant during orphan recovery */
1629 granted = tgt_grant_create(tsi->tsi_env,
1630 ofd_obd(ofd)->obd_self_export,
1635 CDEBUG(D_HA, "%s: failed to acquire grant "
1636 "space for precreate (%lld): rc = %d\n",
1637 ofd_name(ofd), diff, rc);
1642 /* This can happen if a new OST is formatted and installed
1643 * in place of an old one at the same index. Instead of
1644 * precreating potentially millions of deleted old objects
1645 * (possibly filling the OST), only precreate the last batch.
1646 * LFSCK will eventually clean up any orphans. LU-14 */
1647 if (diff > 5 * OST_MAX_PRECREATE) {
1648 diff = OST_MAX_PRECREATE / 2;
1649 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1650 "OST replaced or reformatted: "
1651 "LFSCK will clean up",
1654 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1655 "%u larger than the LAST_ID "DOSTID", only "
1656 "precreating the last %lld objects.\n",
1657 ofd_name(ofd), POSTID(&oa->o_oi),
1658 5 * OST_MAX_PRECREATE,
1659 POSTID(&oseq->os_oi), diff);
1660 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1664 next_id = ofd_seq_last_oid(oseq) + 1;
1665 count = ofd_precreate_batch(ofd, (int)diff);
1667 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1668 " at %llu\n", ofd_name(ofd),
1669 count, seq, next_id);
1671 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1672 && ktime_get_seconds() > enough_time) {
1673 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1674 " created at a rate of %d/s\n",
1675 ofd_name(ofd), created, diff + created,
1676 created / DISK_TIMEOUT);
1680 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1681 oseq, count, sync_trans);
1685 } else if (rc < 0) {
1691 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1692 LCONSOLE_WARN("%s: can't create the same count of"
1693 " objects when replaying the request"
1694 " (diff is %lld). see LU-4621\n",
1695 ofd_name(ofd), diff);
1698 /* some objects got created, we can return
1699 * them, even if last creation failed */
1702 CERROR("%s: unable to precreate: rc = %d\n",
1705 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1706 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1707 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1712 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1716 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1719 mutex_unlock(&oseq->os_create_lock);
1722 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1723 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1724 struct lu_fid *fid = &info->fti_fid;
1726 /* For compatible purpose, it needs to convert back to
1727 * OST ID before put it on wire. */
1728 *fid = rep_oa->o_oi.oi_fid;
1729 fid_to_ostid(fid, &rep_oa->o_oi);
1731 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1733 ofd_seq_put(tsi->tsi_env, oseq);
1736 up_read(&ofd->ofd_lastid_rwsem);
1741 * OFD request handler for OST_DESTROY RPC.
1743 * This is OFD-specific part of request handling. It destroys data objects
1744 * related to destroyed object on MDT.
1746 * \param[in] tsi target session environment for this request
1748 * \retval 0 if successful
1749 * \retval negative value on error
1751 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1753 const struct ost_body *body = tsi->tsi_ost_body;
1754 struct ost_body *repbody;
1755 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1756 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1757 struct lu_fid *fid = &fti->fti_fid;
1764 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1767 /* This is old case for clients before Lustre 2.4 */
1768 /* If there's a DLM request, cancel the locks mentioned in it */
1769 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1771 struct ldlm_request *dlm;
1773 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1776 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1779 *fid = body->oa.o_oi.oi_fid;
1780 oid = ostid_id(&body->oa.o_oi);
1783 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1785 /* check that o_misc makes sense */
1786 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1787 count = body->oa.o_misc;
1789 count = 1; /* default case - single destroy */
1791 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1792 POSTID(&body->oa.o_oi), count);
1797 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1798 if (lrc == -ENOENT) {
1800 "%s: destroying non-existent object "DFID"\n",
1801 ofd_name(ofd), PFID(fid));
1802 /* rewrite rc with -ENOENT only if it is 0 */
1805 } else if (lrc != 0) {
1806 CERROR("%s: error destroying object "DFID": %d\n",
1807 ofd_name(ofd), PFID(fid), lrc);
1813 lrc = fid_set_id(fid, oid);
1814 if (unlikely(lrc != 0 && count > 0))
1815 GOTO(out, rc = lrc);
1818 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1824 fid_to_ostid(fid, &repbody->oa.o_oi);
1829 * OFD request handler for OST_STATFS RPC.
1831 * This function gets statfs data from storage as part of request
1834 * \param[in] tsi target session environment for this request
1836 * \retval 0 if successful
1837 * \retval negative value on error
1839 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1841 struct obd_statfs *osfs;
1846 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1848 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1850 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1851 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1853 CERROR("%s: statfs failed: rc = %d\n",
1854 tgt_name(tsi->tsi_tgt), rc);
1856 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1859 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1866 * OFD request handler for OST_SYNC RPC.
1868 * Sync object data or all filesystem data to the disk and pack the
1871 * \param[in] tsi target session environment for this request
1873 * \retval 0 if successful
1874 * \retval negative value on error
1876 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1878 struct ost_body *body = tsi->tsi_ost_body;
1879 struct ost_body *repbody;
1880 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1881 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1882 struct ofd_object *fo = NULL;
1887 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1889 /* if no objid is specified, it means "sync whole filesystem" */
1890 if (!fid_is_zero(&tsi->tsi_fid)) {
1891 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1893 RETURN(PTR_ERR(fo));
1896 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1897 fo != NULL ? ofd_object_child(fo) : NULL,
1898 repbody->oa.o_size, repbody->oa.o_blocks);
1902 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1907 repbody->oa.o_oi = body->oa.o_oi;
1908 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1910 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1912 obdo_from_la(&repbody->oa, &fti->fti_attr,
1915 /* don't return rc from getattr */
1920 ofd_object_put(tsi->tsi_env, fo);
1925 * OFD request handler for OST_PUNCH RPC.
1927 * This is part of request processing. Validate request fields,
1928 * punch (truncate) the given OFD object and pack reply.
1930 * \param[in] tsi target session environment for this request
1932 * \retval 0 if successful
1933 * \retval negative value on error
1935 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1937 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1938 struct ost_body *repbody;
1939 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1940 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1941 struct ldlm_resource *res;
1942 struct ofd_object *fo;
1944 struct lustre_handle lh = { 0, };
1951 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
1953 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1954 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
1956 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1957 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1958 RETURN(err_serious(-EPROTO));
1960 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1961 if (repbody == NULL)
1962 RETURN(err_serious(-ENOMEM));
1964 /* punch start,end are passed in o_size,o_blocks throught wire */
1968 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1971 /* standard truncate optimization: if file body is completely
1972 * destroyed, don't send data back to the server. */
1974 flags |= LDLM_FL_AST_DISCARD_DATA;
1976 repbody->oa.o_oi = oa->o_oi;
1977 repbody->oa.o_valid = OBD_MD_FLID;
1979 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1980 oa->o_flags & OBD_FL_SRVLOCK;
1983 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
1984 end, &lh, LCK_PW, &flags);
1989 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
1990 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
1991 oa->o_valid, start, end);
1993 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
1996 GOTO(out, rc = PTR_ERR(fo));
1998 la_from_obdo(&info->fti_attr, oa,
1999 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2000 info->fti_attr.la_size = start;
2001 info->fti_attr.la_valid |= LA_SIZE;
2003 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2008 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2012 ofd_object_put(tsi->tsi_env, fo);
2015 tgt_extent_unlock(&lh, LCK_PW);
2017 /* we do not call this before to avoid lu_object_find() in
2018 * ->lvbo_update() holding another reference on the object.
2019 * otherwise concurrent destroy can make the object unavailable
2020 * for 2nd lu_object_find() waiting for the first reference
2021 * to go... deadlock! */
2022 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2025 struct ost_lvb *res_lvb;
2027 ldlm_res_lvbo_update(tsi->tsi_env, res, NULL, 0);
2028 res_lvb = res->lr_lvb_data;
2029 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2030 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2031 ldlm_resource_putref(res);
2037 static int ofd_ladvise_prefetch(const struct lu_env *env,
2038 struct ofd_object *fo,
2039 struct niobuf_local *lnb,
2040 __u64 start, __u64 end, enum dt_bufs_type dbt)
2042 struct ofd_thread_info *info = ofd_info(env);
2043 pgoff_t start_index, end_index, pages;
2044 struct niobuf_remote rnb;
2045 unsigned long nr_local;
2051 ofd_read_lock(env, fo);
2052 if (!ofd_object_exists(fo))
2053 GOTO(out_unlock, rc = -ENOENT);
2055 rc = ofd_attr_get(env, fo, &info->fti_attr);
2057 GOTO(out_unlock, rc);
2059 if (end > info->fti_attr.la_size)
2060 end = info->fti_attr.la_size;
2063 GOTO(out_unlock, rc);
2065 /* We need page aligned offset and length */
2066 start_index = start >> PAGE_SHIFT;
2067 end_index = (end - 1) >> PAGE_SHIFT;
2068 pages = end_index - start_index + 1;
2070 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2071 PTLRPC_MAX_BRW_PAGES;
2072 rnb.rnb_offset = start_index << PAGE_SHIFT;
2073 rnb.rnb_len = nr_local << PAGE_SHIFT;
2074 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, dbt);
2075 if (unlikely(rc < 0))
2078 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2079 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2082 start_index += nr_local;
2087 ofd_read_unlock(env, fo);
2092 * OFD request handler for OST_LADVISE RPC.
2094 * Tune cache or perfetch policies according to advices.
2096 * \param[in] tsi target session environment for this request
2098 * \retval 0 if successful
2099 * \retval negative errno on error
2101 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2103 struct ptlrpc_request *req = tgt_ses_req(tsi);
2104 struct obd_export *exp = tsi->tsi_exp;
2105 struct ofd_device *ofd = ofd_exp(exp);
2106 struct ost_body *body, *repbody;
2107 struct ofd_thread_info *info;
2108 struct ofd_object *fo;
2109 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2110 const struct lu_env *env = svc_thread->t_env;
2111 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2112 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2113 struct lu_ladvise *ladvise;
2115 struct ladvise_hdr *ladvise_hdr;
2116 struct obd_ioobj ioo;
2117 struct lustre_handle lockh = { 0 };
2120 struct dt_object *dob;
2126 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2127 body = tsi->tsi_ost_body;
2129 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2130 RETURN(err_serious(-EPROTO));
2132 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2133 &RMF_OST_LADVISE_HDR);
2134 if (ladvise_hdr == NULL)
2135 RETURN(err_serious(-EPROTO));
2137 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2138 ladvise_hdr->lah_count < 1)
2139 RETURN(err_serious(-EPROTO));
2141 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2142 RETURN(err_serious(-EPROTO));
2144 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2145 if (ladvise == NULL)
2146 RETURN(err_serious(-EPROTO));
2148 num_advise = req_capsule_get_size(&req->rq_pill,
2149 &RMF_OST_LADVISE, RCL_CLIENT) /
2151 if (num_advise < ladvise_hdr->lah_count)
2152 RETURN(err_serious(-EPROTO));
2154 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2155 repbody->oa = body->oa;
2157 info = ofd_info_init(env, exp);
2159 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2160 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2164 fo = ofd_object_find(env, ofd, &info->fti_fid);
2169 LASSERT(fo != NULL);
2170 dob = ofd_object_child(fo);
2172 if (ptlrpc_connection_is_local(exp->exp_connection))
2173 dbt |= DT_BUFS_TYPE_LOCAL;
2175 for (i = 0; i < num_advise; i++, ladvise++) {
2176 start = ladvise->lla_start;
2177 end = ladvise->lla_end;
2179 rc = err_serious(-EPROTO);
2183 /* Handle different advice types */
2184 switch (ladvise->lla_advice) {
2188 case LU_LADVISE_WILLREAD:
2192 ioo.ioo_oid = body->oa.o_oi;
2194 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2195 &tsi->tsi_resid, start, end - 1,
2196 &lockh, LCK_PR, &flags);
2200 req->rq_status = ofd_ladvise_prefetch(env, fo,
2203 tgt_extent_unlock(&lockh, LCK_PR);
2205 case LU_LADVISE_DONTNEED:
2206 rc = dt_ladvise(env, dob, ladvise->lla_start,
2207 ladvise->lla_end, LU_LADVISE_DONTNEED);
2214 ofd_object_put(env, fo);
2215 req->rq_status = rc;
2220 * OFD request handler for OST_QUOTACTL RPC.
2222 * This is part of request processing to validate incoming request fields,
2223 * get the requested data from OSD and pack reply.
2225 * \param[in] tsi target session environment for this request
2227 * \retval 0 if successful
2228 * \retval negative value on error
2230 static int ofd_quotactl(struct tgt_session_info *tsi)
2232 struct obd_quotactl *oqctl, *repoqc;
2233 struct lu_nodemap *nodemap;
2239 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2241 RETURN(err_serious(-EPROTO));
2243 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2245 RETURN(err_serious(-ENOMEM));
2249 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2250 if (IS_ERR(nodemap))
2251 RETURN(PTR_ERR(nodemap));
2254 if (oqctl->qc_type == USRQUOTA)
2255 id = nodemap_map_id(nodemap, NODEMAP_UID,
2256 NODEMAP_CLIENT_TO_FS,
2258 else if (oqctl->qc_type == GRPQUOTA)
2259 id = nodemap_map_id(nodemap, NODEMAP_GID,
2260 NODEMAP_CLIENT_TO_FS,
2263 nodemap_putref(nodemap);
2265 if (repoqc->qc_id != id)
2266 swap(repoqc->qc_id, id);
2268 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2270 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2273 if (repoqc->qc_id != id)
2274 swap(repoqc->qc_id, id);
2280 * Calculate the amount of time for lock prolongation.
2282 * This is helper for ofd_prolong_extent_locks() function to get
2283 * the timeout extra time.
2285 * \param[in] req current request
2287 * \retval amount of time to extend the timeout with
2289 static inline time64_t prolong_timeout(struct ptlrpc_request *req)
2291 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2292 time64_t req_timeout;
2295 return obd_timeout / 2;
2297 req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec;
2298 return max_t(time64_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
2303 * Prolong lock timeout for the given extent.
2305 * This function finds all locks related with incoming request and
2306 * prolongs their timeout.
2308 * If a client is holding a lock for a long time while it sends
2309 * read or write RPCs to the OST for the object under this lock,
2310 * then we don't want the OST to evict the client. Otherwise,
2311 * if the network or disk is very busy then the client may not
2312 * be able to make any progress to clear out dirty pages under
2313 * the lock and the application will fail.
2315 * Every time a Bulk Read/Write (BRW) request arrives for the object
2316 * covered by the lock, extend the timeout on that lock. The RPC should
2317 * contain a lock handle for the lock it is using, but this
2318 * isn't handled correctly by all client versions, and the
2319 * request may cover multiple locks.
2321 * \param[in] tsi target session environment for this request
2322 * \param[in] data struct of data to prolong locks
2325 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2326 struct ldlm_prolong_args *data)
2328 struct obdo *oa = &tsi->tsi_ost_body->oa;
2329 struct ldlm_lock *lock;
2333 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2334 data->lpa_export = tsi->tsi_exp;
2335 data->lpa_resid = tsi->tsi_resid;
2337 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2338 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2339 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2340 data->lpa_extent.end);
2342 if (oa->o_valid & OBD_MD_FLHANDLE) {
2343 /* mostly a request should be covered by only one lock, try
2345 lock = ldlm_handle2lock(&oa->o_handle);
2347 /* Fast path to check if the lock covers the whole IO
2348 * region exclusively. */
2349 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2350 &data->lpa_extent)) {
2352 LASSERT(lock->l_export == data->lpa_export);
2353 ldlm_lock_prolong_one(lock, data);
2354 LDLM_LOCK_PUT(lock);
2355 if (data->lpa_locks_cnt > 0)
2357 /* The lock was destroyed probably lets try
2360 lock->l_last_used = ktime_get();
2361 LDLM_LOCK_PUT(lock);
2366 ldlm_resource_prolong(data);
2371 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2373 * Determine if \a lock and the lock from request \a req are equivalent
2374 * by comparing their resource names, modes, and extents.
2376 * It is used to give priority to read and write RPCs being done
2377 * under this lock so that the client can drop the contended
2378 * lock more quickly and let other clients use it. This improves
2379 * overall performance in the case where the first client gets a
2380 * very large lock extent that prevents other clients from
2381 * submitting their writes.
2383 * \param[in] req ptlrpc_request being processed
2384 * \param[in] lock contended lock to match
2386 * \retval 1 if lock is matched
2387 * \retval 0 otherwise
2389 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2390 struct ldlm_lock *lock)
2392 struct niobuf_remote *rnb;
2393 struct obd_ioobj *ioo;
2394 enum ldlm_mode mode;
2395 struct ldlm_extent ext;
2396 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2400 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2401 LASSERT(ioo != NULL);
2403 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2404 LASSERT(rnb != NULL);
2406 ext.start = rnb->rnb_offset;
2407 rnb += ioo->ioo_bufcnt - 1;
2408 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2410 LASSERT(lock->l_resource != NULL);
2411 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2414 /* a bulk write can only hold a reference on a PW extent lock
2417 mode = LCK_PW | LCK_GROUP;
2418 if (opc == OST_READ)
2419 /* whereas a bulk read can be protected by either a PR or PW
2423 if (!(lock->l_granted_mode & mode))
2426 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2430 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2432 * Check for whether the given PTLRPC request (\a req) is blocking
2433 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2436 * \param[in] req the incoming request
2438 * \retval 1 if \a req is blocking an LDLM lock cancel
2439 * \retval 0 if it is not
2440 * \retval -ESTALE if lock is not found
2442 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2444 struct tgt_session_info *tsi;
2445 struct obd_ioobj *ioo;
2446 struct niobuf_remote *rnb;
2448 struct ldlm_prolong_args pa = { 0 };
2452 /* Don't use tgt_ses_info() to get session info, because lock_match()
2453 * can be called while request has no processing thread yet. */
2454 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2457 * Use LASSERT below because malformed RPCs should have
2458 * been filtered out in tgt_hpreq_handler().
2460 opc = lustre_msg_get_opc(req->rq_reqmsg);
2461 LASSERT(opc == OST_READ || opc == OST_WRITE);
2463 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2464 LASSERT(ioo != NULL);
2466 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2467 LASSERT(rnb != NULL);
2468 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2470 pa.lpa_mode = LCK_PW | LCK_GROUP;
2471 if (opc == OST_READ)
2472 pa.lpa_mode |= LCK_PR;
2474 pa.lpa_extent.start = rnb->rnb_offset;
2475 rnb += ioo->ioo_bufcnt - 1;
2476 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2478 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
2479 " (%llu->%llu)\n", tgt_name(tsi->tsi_tgt),
2480 current->comm, PFID(&tsi->tsi_fid), pa.lpa_extent.start,
2483 ofd_prolong_extent_locks(tsi, &pa);
2485 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2486 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2488 if (pa.lpa_blocks_cnt > 0)
2491 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2495 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2497 * Called after the request has been handled. It refreshes lock timeout again
2498 * so that client has more time to send lock cancel RPC.
2500 * \param[in] req request which is being processed.
2502 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2504 ofd_rw_hpreq_check(req);
2508 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2510 * This function checks if the given lock is the same by its resname, mode
2511 * and extent as one taken from the request.
2512 * It is used to give priority to punch/truncate RPCs that might lead to
2513 * the fastest release of that lock when a lock is contended.
2515 * \param[in] req ptlrpc_request being processed
2516 * \param[in] lock contended lock to match
2518 * \retval 1 if lock is matched
2519 * \retval 0 otherwise
2521 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2522 struct ldlm_lock *lock)
2524 struct tgt_session_info *tsi;
2526 struct ldlm_extent ext;
2530 /* Don't use tgt_ses_info() to get session info, because lock_match()
2531 * can be called while request has no processing thread yet. */
2532 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2535 * Use LASSERT below because malformed RPCs should have
2536 * been filtered out in tgt_hpreq_handler().
2538 LASSERT(tsi->tsi_ost_body != NULL);
2539 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2540 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2543 oa = &tsi->tsi_ost_body->oa;
2544 ext.start = oa->o_size;
2545 ext.end = oa->o_blocks;
2547 LASSERT(lock->l_resource != NULL);
2548 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2551 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2554 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2558 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2560 * High-priority queue request check for whether the given punch request
2561 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2562 * covered by an LDLM lock.
2566 * \param[in] req the incoming request
2568 * \retval 1 if \a req is blocking an LDLM lock cancel
2569 * \retval 0 if it is not
2570 * \retval -ESTALE if lock is not found
2572 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2574 struct tgt_session_info *tsi;
2576 struct ldlm_prolong_args pa = { 0 };
2580 /* Don't use tgt_ses_info() to get session info, because lock_match()
2581 * can be called while request has no processing thread yet. */
2582 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2583 LASSERT(tsi != NULL);
2584 oa = &tsi->tsi_ost_body->oa;
2586 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2587 oa->o_flags & OBD_FL_SRVLOCK));
2589 pa.lpa_mode = LCK_PW | LCK_GROUP;
2590 pa.lpa_extent.start = oa->o_size;
2591 pa.lpa_extent.end = oa->o_blocks;
2594 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2595 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2596 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2598 ofd_prolong_extent_locks(tsi, &pa);
2600 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2601 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2603 if (pa.lpa_blocks_cnt > 0)
2606 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2610 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2612 * Called after the request has been handled. It refreshes lock timeout again
2613 * so that client has more time to send lock cancel RPC.
2615 * \param[in] req request which is being processed.
2617 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2619 ofd_punch_hpreq_check(req);
2622 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2623 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2624 .hpreq_check = ofd_rw_hpreq_check,
2625 .hpreq_fini = ofd_rw_hpreq_fini
2628 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2629 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2630 .hpreq_check = ofd_punch_hpreq_check,
2631 .hpreq_fini = ofd_punch_hpreq_fini
2635 * Assign high priority operations to an IO request.
2637 * Check if the incoming request is a candidate for
2638 * high-priority processing. If it is, assign it a high
2639 * priority operations table.
2641 * \param[in] tsi target session environment for this request
2643 static void ofd_hp_brw(struct tgt_session_info *tsi)
2645 struct niobuf_remote *rnb;
2646 struct obd_ioobj *ioo;
2650 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2651 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2652 if (ioo->ioo_bufcnt > 0) {
2653 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2654 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2656 /* no high priority if server lock is needed */
2657 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2658 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2662 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2666 * Assign high priority operations to an punch request.
2668 * Check if the incoming request is a candidate for
2669 * high-priority processing. If it is, assign it a high
2670 * priority operations table.
2672 * \param[in] tsi target session environment for this request
2674 static void ofd_hp_punch(struct tgt_session_info *tsi)
2676 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2677 /* no high-priority if server lock is needed */
2678 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2679 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2680 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2681 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2683 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2686 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2687 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2688 #define OST_BRW_READ OST_READ
2689 #define OST_BRW_WRITE OST_WRITE
2692 * Table of OFD-specific request handlers
2694 * This table contains all opcodes accepted by OFD and
2695 * specifies handlers for them. The tgt_request_handler()
2696 * uses such table from each target to process incoming
2699 static struct tgt_handler ofd_tgt_handlers[] = {
2700 TGT_RPC_HANDLER(OST_FIRST_OPC,
2701 0, OST_CONNECT, tgt_connect,
2702 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2703 TGT_RPC_HANDLER(OST_FIRST_OPC,
2704 0, OST_DISCONNECT, tgt_disconnect,
2705 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2706 TGT_RPC_HANDLER(OST_FIRST_OPC,
2707 0, OST_SET_INFO, ofd_set_info_hdl,
2708 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2709 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2710 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
2711 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2712 OST_SETATTR, ofd_setattr_hdl),
2713 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2714 OST_CREATE, ofd_create_hdl),
2715 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2716 OST_DESTROY, ofd_destroy_hdl),
2717 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
2718 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2719 OST_BRW_READ, tgt_brw_read,
2721 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2722 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
2724 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2725 OST_PUNCH, ofd_punch_hdl,
2727 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
2728 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
2729 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_LADVISE, ofd_ladvise_hdl),
2732 static struct tgt_opc_slice ofd_common_slice[] = {
2734 .tos_opc_start = OST_FIRST_OPC,
2735 .tos_opc_end = OST_LAST_OPC,
2736 .tos_hs = ofd_tgt_handlers
2739 .tos_opc_start = OBD_FIRST_OPC,
2740 .tos_opc_end = OBD_LAST_OPC,
2741 .tos_hs = tgt_obd_handlers
2744 .tos_opc_start = LDLM_FIRST_OPC,
2745 .tos_opc_end = LDLM_LAST_OPC,
2746 .tos_hs = tgt_dlm_handlers
2749 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2750 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2751 .tos_hs = tgt_out_handlers
2754 .tos_opc_start = SEQ_FIRST_OPC,
2755 .tos_opc_end = SEQ_LAST_OPC,
2756 .tos_hs = seq_handlers
2759 .tos_opc_start = LFSCK_FIRST_OPC,
2760 .tos_opc_end = LFSCK_LAST_OPC,
2761 .tos_hs = tgt_lfsck_handlers
2764 .tos_opc_start = SEC_FIRST_OPC,
2765 .tos_opc_end = SEC_LAST_OPC,
2766 .tos_hs = tgt_sec_ctx_handlers
2773 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2774 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2777 * Implementation of lu_context_key::lct_key_exit.
2779 * Optional method called on lu_context_exit() for all allocated
2781 * It is used in OFD to sanitize context values which may be re-used
2782 * during another request processing by the same thread.
2784 * \param[in] ctx execution context
2785 * \param[in] key context key
2786 * \param[in] data ofd_thread_info
2788 static void ofd_key_exit(const struct lu_context *ctx,
2789 struct lu_context_key *key, void *data)
2791 struct ofd_thread_info *info = data;
2793 info->fti_env = NULL;
2794 info->fti_exp = NULL;
2797 info->fti_pre_version = 0;
2799 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2802 struct lu_context_key ofd_thread_key = {
2803 .lct_tags = LCT_DT_THREAD,
2804 .lct_init = ofd_key_init,
2805 .lct_fini = ofd_key_fini,
2806 .lct_exit = ofd_key_exit
2810 * Initialize OFD device according to parameters in the config log \a cfg.
2812 * This is the main starting point of OFD initialization. It fills all OFD
2813 * parameters with their initial values and calls other initializing functions
2814 * to set up all OFD subsystems.
2816 * \param[in] env execution environment
2817 * \param[in] m OFD device
2818 * \param[in] ldt LU device type of OFD
2819 * \param[in] cfg configuration log
2821 * \retval 0 if successful
2822 * \retval negative value on error
2824 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2825 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2827 const char *dev = lustre_cfg_string(cfg, 0);
2828 struct ofd_thread_info *info = NULL;
2829 struct obd_device *obd;
2830 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2832 struct nm_config_file *nodemap_config;
2833 struct obd_device_target *obt;
2838 obd = class_name2obd(dev);
2840 CERROR("Cannot find obd with name %s\n", dev);
2844 rc = lu_env_refill((struct lu_env *)env);
2849 obt->obt_magic = OBT_MAGIC;
2851 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2852 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2854 spin_lock_init(&m->ofd_flags_lock);
2855 m->ofd_raid_degraded = 0;
2856 m->ofd_checksum_t10pi_enforce = 0;
2857 m->ofd_sync_journal = 0;
2859 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2861 m->ofd_seq_count = 0;
2862 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2863 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2864 spin_lock_init(&m->ofd_inconsistency_lock);
2866 spin_lock_init(&m->ofd_batch_lock);
2867 init_rwsem(&m->ofd_lastid_rwsem);
2869 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2870 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2871 /* set this lu_device to obd, because error handling need it */
2872 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2874 rc = ofd_tunables_init(m);
2876 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2880 /* No connection accepted until configurations will finish */
2881 spin_lock(&obd->obd_dev_lock);
2882 obd->obd_no_conn = 1;
2883 spin_unlock(&obd->obd_dev_lock);
2884 obd->obd_replayable = 1;
2885 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2886 char *str = lustre_cfg_string(cfg, 4);
2888 if (strchr(str, 'n')) {
2889 CWARN("%s: recovery disabled\n", obd->obd_name);
2890 obd->obd_replayable = 0;
2894 info = ofd_info_init(env, NULL);
2896 GOTO(err_fini_proc, rc = -EFAULT);
2898 rc = ofd_stack_init(env, m, cfg);
2900 CERROR("Can't init device stack, rc %d\n", rc);
2901 GOTO(err_fini_proc, rc);
2904 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
2905 ofd_procfs_add_brw_stats_symlink(m);
2908 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2909 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2910 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2911 LDLM_NAMESPACE_SERVER,
2912 LDLM_NAMESPACE_GREEDY,
2914 if (m->ofd_namespace == NULL)
2915 GOTO(err_fini_stack, rc = -ENOMEM);
2916 /* set obd_namespace for compatibility with old code */
2917 obd->obd_namespace = m->ofd_namespace;
2918 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2919 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2920 m->ofd_namespace->ns_lvbp = m;
2922 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2923 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2925 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2926 OBD_FAIL_OST_ALL_REQUEST_NET,
2927 OBD_FAIL_OST_ALL_REPLY_NET);
2929 GOTO(err_free_ns, rc);
2931 tgd->tgd_reserved_pcnt = 0;
2933 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
2934 m->ofd_cksum_types_supported =
2935 obd_cksum_types_supported_server(obd->obd_name);
2936 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2937 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
2938 OFD_PRECREATE_SMALL_FS)
2939 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2941 rc = ofd_fs_setup(env, m, obd);
2943 GOTO(err_fini_lut, rc);
2945 fid.f_seq = FID_SEQ_LOCAL_NAME;
2948 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
2951 GOTO(err_fini_fs, rc);
2953 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
2955 if (IS_ERR(nodemap_config)) {
2956 rc = PTR_ERR(nodemap_config);
2958 GOTO(err_fini_los, rc);
2960 obt->obt_nodemap_config_file = nodemap_config;
2963 rc = ofd_start_inconsistency_verification_thread(m);
2965 GOTO(err_fini_nm, rc);
2967 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
2972 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
2973 obt->obt_nodemap_config_file = NULL;
2975 local_oid_storage_fini(env, m->ofd_los);
2978 ofd_fs_cleanup(env, m);
2980 tgt_fini(env, &m->ofd_lut);
2982 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
2983 obd->obd_namespace = m->ofd_namespace = NULL;
2985 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
2992 * Stop the OFD device
2994 * This function stops the OFD device and all its subsystems.
2995 * This is the end of OFD lifecycle.
2997 * \param[in] env execution environment
2998 * \param[in] m OFD device
3000 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3002 struct obd_device *obd = ofd_obd(m);
3003 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3004 struct lfsck_stop stop;
3006 stop.ls_status = LS_PAUSED;
3008 lfsck_stop(env, m->ofd_osd, &stop);
3009 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3010 target_recovery_fini(obd);
3011 if (m->ofd_namespace != NULL)
3012 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3013 d->ld_obd->obd_force);
3015 obd_exports_barrier(obd);
3016 obd_zombie_barrier();
3018 tgt_fini(env, &m->ofd_lut);
3019 ofd_stop_inconsistency_verification_thread(m);
3020 lfsck_degister(env, m->ofd_osd);
3021 ofd_fs_cleanup(env, m);
3022 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3023 obd->u.obt.obt_nodemap_config_file = NULL;
3025 if (m->ofd_namespace != NULL) {
3026 ldlm_namespace_free_post(m->ofd_namespace);
3027 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3030 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3032 LASSERT(atomic_read(&d->ld_ref) == 0);
3033 server_put_mount(obd->obd_name, true);
3038 * Implementation of lu_device_type_operations::ldto_device_fini.
3040 * Finalize device. Dual to ofd_device_init(). It is called from
3041 * obd_precleanup() and stops the current device.
3043 * \param[in] env execution environment
3044 * \param[in] d LU device of OFD
3048 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3049 struct lu_device *d)
3052 ofd_fini(env, ofd_dev(d));
3057 * Implementation of lu_device_type_operations::ldto_device_free.
3059 * Free OFD device. Dual to ofd_device_alloc().
3061 * \param[in] env execution environment
3062 * \param[in] d LU device of OFD
3066 static struct lu_device *ofd_device_free(const struct lu_env *env,
3067 struct lu_device *d)
3069 struct ofd_device *m = ofd_dev(d);
3071 dt_device_fini(&m->ofd_dt_dev);
3077 * Implementation of lu_device_type_operations::ldto_device_alloc.
3079 * This function allocates the new OFD device. It is called from
3080 * obd_setup() if OBD device had lu_device_type defined.
3082 * \param[in] env execution environment
3083 * \param[in] t lu_device_type of OFD device
3084 * \param[in] cfg configuration log
3086 * \retval pointer to the lu_device of just allocated OFD
3087 * \retval ERR_PTR of return value on error
3089 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3090 struct lu_device_type *t,
3091 struct lustre_cfg *cfg)
3093 struct ofd_device *m;
3094 struct lu_device *l;
3099 return ERR_PTR(-ENOMEM);
3101 l = &m->ofd_dt_dev.dd_lu_dev;
3102 dt_device_init(&m->ofd_dt_dev, t);
3103 rc = ofd_init0(env, m, t, cfg);
3105 ofd_device_free(env, l);
3112 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3113 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3115 static struct lu_device_type_operations ofd_device_type_ops = {
3116 .ldto_init = ofd_type_init,
3117 .ldto_fini = ofd_type_fini,
3119 .ldto_start = ofd_type_start,
3120 .ldto_stop = ofd_type_stop,
3122 .ldto_device_alloc = ofd_device_alloc,
3123 .ldto_device_free = ofd_device_free,
3124 .ldto_device_fini = ofd_device_fini
3127 static struct lu_device_type ofd_device_type = {
3128 .ldt_tags = LU_DEVICE_DT,
3129 .ldt_name = LUSTRE_OST_NAME,
3130 .ldt_ops = &ofd_device_type_ops,
3131 .ldt_ctx_tags = LCT_DT_THREAD
3135 * Initialize OFD module.
3137 * This function is called upon module loading. It registers OFD device type
3138 * and prepares all in-memory structures used by all OFD devices.
3140 * \retval 0 if successful
3141 * \retval negative value on error
3143 static int __init ofd_init(void)
3147 rc = lu_kmem_init(ofd_caches);
3151 rc = ofd_fmd_init();
3153 lu_kmem_fini(ofd_caches);
3157 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3158 LUSTRE_OST_NAME, &ofd_device_type);
3165 * This function is called upon OFD module unloading.
3166 * It frees all related structures and unregisters OFD device type.
3168 static void __exit ofd_exit(void)
3171 lu_kmem_fini(ofd_caches);
3172 class_unregister_type(LUSTRE_OST_NAME);
3175 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3176 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3177 MODULE_VERSION(LUSTRE_VERSION_STRING);
3178 MODULE_LICENSE("GPL");
3180 module_init(ofd_init);
3181 module_exit(ofd_exit);