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);
796 int len = strlen(obd_name) + 7;
799 ss = &ofd->ofd_seq_site;
800 lu->ld_site->ld_seq_site = ss;
801 ss->ss_lu = lu->ld_site;
802 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
804 OBD_ALLOC(name, len);
808 OBD_ALLOC_PTR(ss->ss_server_seq);
809 if (ss->ss_server_seq == NULL)
810 GOTO(out_name, rc = -ENOMEM);
812 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
813 LUSTRE_SEQ_SERVER, ss);
815 CERROR("%s: seq server init error: rc = %d\n", obd_name, rc);
816 GOTO(out_server, rc);
818 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
820 OBD_ALLOC_PTR(ss->ss_client_seq);
821 if (ss->ss_client_seq == NULL)
822 GOTO(out_server, rc = -ENOMEM);
824 snprintf(name, len, "%s-super", obd_name);
825 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
828 CERROR("%s: seq client init error: rc = %d\n", obd_name, rc);
829 GOTO(out_client, rc);
832 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
836 seq_client_fini(ss->ss_client_seq);
837 OBD_FREE_PTR(ss->ss_client_seq);
838 ss->ss_client_seq = NULL;
840 seq_server_fini(ss->ss_server_seq, env);
841 OBD_FREE_PTR(ss->ss_server_seq);
842 ss->ss_server_seq = NULL;
851 * OFD request handler for OST_SET_INFO RPC.
853 * This is OFD-specific part of request handling
855 * \param[in] tsi target session environment for this request
857 * \retval 0 if successful
858 * \retval negative value on error
860 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
862 struct ptlrpc_request *req = tgt_ses_req(tsi);
863 struct ost_body *body = NULL, *repbody;
864 void *key, *val = NULL;
865 int keylen, vallen, rc = 0;
866 bool is_grant_shrink;
870 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
872 DEBUG_REQ(D_HA, req, "no set_info key");
873 RETURN(err_serious(-EFAULT));
875 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
878 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
880 DEBUG_REQ(D_HA, req, "no set_info val");
881 RETURN(err_serious(-EFAULT));
883 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
886 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
888 /* In this case the value is actually an RMF_OST_BODY, so we
889 * transmutate the type of this PTLRPC */
890 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
892 rc = req_capsule_server_pack(tsi->tsi_pill);
896 if (is_grant_shrink) {
897 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
899 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
902 /** handle grant shrink, similar to a read request */
903 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
905 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
907 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
910 CERROR("%s: Unsupported key %s\n",
911 tgt_name(tsi->tsi_tgt), (char *)key);
914 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
921 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
923 * This function returns a list of extents which describes how a file's
924 * blocks are laid out on the disk.
926 * \param[in] env execution environment
927 * \param[in] ofd OFD device
928 * \param[in] fid FID of object
929 * \param[in] fiemap fiemap structure to fill with data
931 * \retval 0 if \a fiemap is filled with data successfully
932 * \retval negative value on error
934 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
935 struct lu_fid *fid, struct fiemap *fiemap)
937 struct ofd_object *fo;
940 fo = ofd_object_find(env, ofd, fid);
942 CERROR("%s: error finding object "DFID"\n",
943 ofd_name(ofd), PFID(fid));
947 ofd_read_lock(env, fo);
948 if (ofd_object_exists(fo))
949 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
952 ofd_read_unlock(env, fo);
953 ofd_object_put(env, fo);
958 static int ofd_lock_unlock_region(const struct lu_env *env,
959 struct ldlm_namespace *ns,
960 struct ldlm_res_id *res_id,
961 unsigned long long begin,
962 unsigned long long end)
966 struct lustre_handle lh = { 0 };
968 LASSERT(begin <= end);
970 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
974 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
975 tgt_extent_unlock(&lh, LCK_PR);
981 * Lock the sparse areas of given resource.
983 * The locking of sparse areas will cause dirty data to be flushed back from
984 * clients. This is used when getting the FIEMAP of an object to make sure
985 * there is no unaccounted cached data on clients.
987 * This function goes through \a fiemap list of extents and locks only sparse
988 * areas between extents.
990 * \param[in] ns LDLM namespace
991 * \param[in] res_id resource ID
992 * \param[in] fiemap file extents mapping on disk
993 * \param[in] locked list head of regions list
995 * \retval 0 if successful
996 * \retval negative value on error
998 static int lock_zero_regions(const struct lu_env *env,
999 struct ldlm_namespace *ns,
1000 struct ldlm_res_id *res_id,
1001 struct fiemap *fiemap)
1003 __u64 begin = fiemap->fm_start;
1006 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1010 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1011 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1012 if (fiemap_start[i].fe_logical > begin) {
1013 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1014 begin, fiemap_start[i].fe_logical);
1015 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1016 fiemap_start[i].fe_logical);
1021 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1024 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1025 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1026 begin, fiemap->fm_start + fiemap->fm_length);
1027 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1028 fiemap->fm_start + fiemap->fm_length);
1036 * OFD request handler for OST_GET_INFO RPC.
1038 * This is OFD-specific part of request handling. The OFD-specific keys are:
1039 * - KEY_LAST_ID (obsolete)
1043 * This function reads needed data from storage and fills reply with it.
1045 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1046 * and is kept for compatibility.
1048 * \param[in] tsi target session environment for this request
1050 * \retval 0 if successful
1051 * \retval negative value on error
1053 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1055 struct obd_export *exp = tsi->tsi_exp;
1056 struct ofd_device *ofd = ofd_exp(exp);
1057 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1060 int replylen, rc = 0;
1064 /* this common part for get_info rpc */
1065 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1067 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1068 RETURN(err_serious(-EPROTO));
1070 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1073 if (KEY_IS(KEY_LAST_ID)) {
1075 struct ofd_seq *oseq;
1077 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1078 rc = req_capsule_server_pack(tsi->tsi_pill);
1080 RETURN(err_serious(rc));
1082 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1084 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1085 (u64)exp->exp_filter_data.fed_group);
1089 *last_id = ofd_seq_last_oid(oseq);
1090 ofd_seq_put(tsi->tsi_env, oseq);
1091 } else if (KEY_IS(KEY_FIEMAP)) {
1092 struct ll_fiemap_info_key *fm_key;
1093 struct fiemap *fiemap;
1096 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1098 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1099 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1101 RETURN(err_serious(rc));
1103 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1105 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1107 replylen = fiemap_count_to_size(
1108 fm_key->lfik_fiemap.fm_extent_count);
1109 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1110 RCL_SERVER, replylen);
1112 rc = req_capsule_server_pack(tsi->tsi_pill);
1114 RETURN(err_serious(rc));
1116 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1120 *fiemap = fm_key->lfik_fiemap;
1121 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1123 /* LU-3219: Lock the sparse areas to make sure dirty
1124 * flushed back from client, then call fiemap again. */
1125 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1126 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1127 ost_fid_build_resid(fid, &fti->fti_resid);
1128 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1129 &fti->fti_resid, fiemap);
1131 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1134 } else if (KEY_IS(KEY_LAST_FID)) {
1135 struct ofd_device *ofd = ofd_exp(exp);
1136 struct ofd_seq *oseq;
1140 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1141 rc = req_capsule_server_pack(tsi->tsi_pill);
1143 RETURN(err_serious(rc));
1145 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1147 RETURN(err_serious(-EPROTO));
1149 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1151 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1155 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1156 ostid_seq(&fti->fti_ostid));
1158 RETURN(PTR_ERR(oseq));
1160 rc = ostid_to_fid(fid, &oseq->os_oi,
1161 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1165 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1168 ofd_seq_put(tsi->tsi_env, oseq);
1170 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1174 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1181 * OFD request handler for OST_GETATTR RPC.
1183 * This is OFD-specific part of request handling. It finds the OFD object
1184 * by its FID, gets attributes from storage and packs result to the reply.
1186 * \param[in] tsi target session environment for this request
1188 * \retval 0 if successful
1189 * \retval negative value on error
1191 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1193 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1194 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1195 struct ost_body *repbody;
1196 struct lustre_handle lh = { 0 };
1197 struct ofd_object *fo;
1199 enum ldlm_mode lock_mode = LCK_PR;
1204 LASSERT(tsi->tsi_ost_body != NULL);
1206 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1207 if (repbody == NULL)
1210 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1211 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1213 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1214 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1217 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1220 rc = tgt_extent_lock(tsi->tsi_env,
1221 tsi->tsi_tgt->lut_obd->obd_namespace,
1222 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1228 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1230 GOTO(out, rc = PTR_ERR(fo));
1232 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1236 obdo_from_la(&repbody->oa, &fti->fti_attr,
1237 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1239 /* Store object version in reply */
1240 curr_version = dt_version_get(tsi->tsi_env,
1241 ofd_object_child(fo));
1242 if ((__s64)curr_version != -EOPNOTSUPP) {
1243 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1244 repbody->oa.o_data_version = curr_version;
1247 if (fo->ofo_ff.ff_layout_version > 0) {
1248 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1249 repbody->oa.o_layout_version =
1250 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1252 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1253 PFID(&tsi->tsi_fid),
1254 repbody->oa.o_layout_version);
1258 ofd_object_put(tsi->tsi_env, fo);
1261 tgt_extent_unlock(&lh, lock_mode);
1263 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1266 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1267 repbody->oa.o_flags = OBD_FL_FLUSH;
1273 * OFD request handler for OST_SETATTR RPC.
1275 * This is OFD-specific part of request handling. It finds the OFD object
1276 * by its FID, sets attributes from request and packs result to the reply.
1278 * \param[in] tsi target session environment for this request
1280 * \retval 0 if successful
1281 * \retval negative value on error
1283 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1285 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1286 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1287 struct ost_body *body = tsi->tsi_ost_body;
1288 struct ost_body *repbody;
1289 struct ldlm_resource *res;
1290 struct ofd_object *fo;
1295 LASSERT(body != NULL);
1297 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1298 if (repbody == NULL)
1301 repbody->oa.o_oi = body->oa.o_oi;
1302 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1304 /* This would be very bad - accidentally truncating a file when
1305 * changing the time or similar - bug 12203. */
1306 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1307 body->oa.o_size != OBD_OBJECT_EOF) {
1308 static char mdsinum[48];
1310 if (body->oa.o_valid & OBD_MD_FLFID)
1311 snprintf(mdsinum, sizeof(mdsinum) - 1,
1312 "of parent "DFID, body->oa.o_parent_seq,
1313 body->oa.o_parent_oid, 0);
1317 CERROR("%s: setattr from %s is trying to truncate object "DFID
1318 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1319 PFID(&tsi->tsi_fid), mdsinum);
1323 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1325 GOTO(out, rc = PTR_ERR(fo));
1327 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1328 fti->fti_attr.la_valid &= ~LA_TYPE;
1330 /* setting objects attributes (including owner/group) */
1331 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1335 obdo_from_la(&repbody->oa, &fti->fti_attr,
1336 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1338 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1342 ofd_object_put(tsi->tsi_env, fo);
1345 /* we do not call this before to avoid lu_object_find() in
1346 * ->lvbo_update() holding another reference on the object.
1347 * otherwise concurrent destroy can make the object unavailable
1348 * for 2nd lu_object_find() waiting for the first reference
1349 * to go... deadlock! */
1350 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1351 &tsi->tsi_resid, LDLM_EXTENT, 0);
1353 ldlm_res_lvbo_update(tsi->tsi_env, res, NULL, 0);
1354 ldlm_resource_putref(res);
1361 * Destroy OST orphans.
1363 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1364 * set then we must destroy possible orphaned objects.
1366 * \param[in] env execution environment
1367 * \param[in] exp OBD export
1368 * \param[in] ofd OFD device
1369 * \param[in] oa obdo structure for reply
1371 * \retval 0 if successful
1372 * \retval negative value on error
1374 static int ofd_orphans_destroy(const struct lu_env *env,
1375 struct obd_export *exp,
1376 struct ofd_device *ofd, struct obdo *oa)
1378 struct ofd_thread_info *info = ofd_info(env);
1379 struct lu_fid *fid = &info->fti_fid;
1380 struct ost_id *oi = &oa->o_oi;
1381 struct ofd_seq *oseq;
1382 u64 seq = ostid_seq(oi);
1383 u64 end_id = ostid_id(oi);
1391 oseq = ofd_seq_get(ofd, seq);
1393 CERROR("%s: Can not find seq for "DOSTID"\n",
1394 ofd_name(ofd), POSTID(oi));
1399 last = ofd_seq_last_oid(oseq);
1402 LASSERT(exp != NULL);
1403 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1405 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1408 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1409 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1411 while (oid > end_id) {
1412 rc = fid_set_id(fid, oid);
1413 if (unlikely(rc != 0))
1416 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1417 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1418 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1419 /* this is pretty fatal... */
1420 CEMERG("%s: error destroying precreated id "
1422 ofd_name(ofd), PFID(fid), rc);
1426 ofd_seq_last_oid_set(oseq, oid);
1427 /* update last_id on disk periodically so that if we
1428 * restart * we don't need to re-scan all of the just
1429 * deleted objects. */
1430 if ((oid & 511) == 0)
1431 ofd_seq_last_oid_write(env, ofd, oseq);
1435 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1436 ofd_name(ofd), seq, oid);
1440 ofd_seq_last_oid_set(oseq, oid);
1441 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1443 /* don't reuse orphan object, return last used objid */
1444 rc = ostid_set_id(oi, last);
1450 ofd_seq_put(env, oseq);
1455 * OFD request handler for OST_CREATE RPC.
1457 * This is OFD-specific part of request handling. Its main purpose is to
1458 * create new data objects on OST, but it also used to destroy orphans.
1460 * \param[in] tsi target session environment for this request
1462 * \retval 0 if successful
1463 * \retval negative value on error
1465 static int ofd_create_hdl(struct tgt_session_info *tsi)
1467 struct ptlrpc_request *req = tgt_ses_req(tsi);
1468 struct ost_body *repbody;
1469 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1470 struct obdo *rep_oa;
1471 struct obd_export *exp = tsi->tsi_exp;
1472 struct ofd_device *ofd = ofd_exp(exp);
1473 u64 seq = ostid_seq(&oa->o_oi);
1474 u64 oid = ostid_id(&oa->o_oi);
1475 struct ofd_seq *oseq;
1483 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1486 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1487 if (repbody == NULL)
1490 down_read(&ofd->ofd_lastid_rwsem);
1491 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1492 * we may do that in the future.
1493 * Return -ENOSPC until the LAST_ID rebuilt. */
1494 if (unlikely(ofd->ofd_lastid_rebuilding))
1495 GOTO(out_sem, rc = -ENOSPC);
1497 rep_oa = &repbody->oa;
1498 rep_oa->o_oi = oa->o_oi;
1500 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1502 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1504 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1506 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1507 ofd_name(ofd), seq, PTR_ERR(oseq));
1508 GOTO(out_sem, rc = -EINVAL);
1511 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1512 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1513 if (!ofd_obd(ofd)->obd_recovering ||
1514 oid > ofd_seq_last_oid(oseq)) {
1515 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1516 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1517 ofd_seq_last_oid(oseq));
1518 GOTO(out_nolock, rc = -EINVAL);
1520 /* Do nothing here, we re-create objects during recovery
1521 * upon write replay, see ofd_preprw_write() */
1522 GOTO(out_nolock, rc = 0);
1524 /* former ofd_handle_precreate */
1525 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1526 (oa->o_flags & OBD_FL_DELORPHAN)) {
1527 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1529 /* destroy orphans */
1530 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1531 exp->exp_conn_cnt) {
1532 CERROR("%s: dropping old orphan cleanup request\n",
1534 GOTO(out_nolock, rc = 0);
1536 /* This causes inflight precreates to abort and drop lock */
1537 oseq->os_destroys_in_progress = 1;
1538 mutex_lock(&oseq->os_create_lock);
1539 if (!oseq->os_destroys_in_progress) {
1541 "%s:[%llu] destroys_in_progress already cleared\n",
1542 ofd_name(ofd), seq);
1543 rc = ostid_set_id(&rep_oa->o_oi,
1544 ofd_seq_last_oid(oseq));
1547 diff = oid - ofd_seq_last_oid(oseq);
1548 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1549 ofd_seq_last_oid(oseq), diff);
1550 if (-diff > OST_MAX_PRECREATE) {
1551 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1552 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1553 "(%llu), trust the OST\n",
1554 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1555 PFID(&oseq->os_oi.oi_fid),
1556 ofd_seq_last_oid(oseq));
1558 /* Let MDS know that we are so far ahead. */
1559 rc = ostid_set_id(&rep_oa->o_oi,
1560 ofd_seq_last_oid(oseq) + 1);
1561 } else if (diff < 0) {
1562 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1564 oseq->os_destroys_in_progress = 0;
1566 /* XXX: Used by MDS for the first time! */
1567 oseq->os_destroys_in_progress = 0;
1570 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1571 ofd->ofd_lastid_gen)) {
1572 /* Keep the export ref so we can send the reply. */
1573 ofd_obd_disconnect(class_export_get(exp));
1574 GOTO(out_nolock, rc = -ENOTCONN);
1577 mutex_lock(&oseq->os_create_lock);
1578 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1579 exp->exp_conn_cnt) {
1580 CERROR("%s: dropping old precreate request\n",
1584 /* only precreate if seq is 0, IDIF or normal and also o_id
1585 * must be specfied */
1586 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1587 !fid_seq_is_idif(seq)) || oid == 0) {
1588 diff = 1; /* shouldn't we create this right now? */
1590 diff = oid - ofd_seq_last_oid(oseq);
1591 /* Do sync create if the seq is about to used up */
1592 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1593 if (unlikely(oid >= IDIF_MAX_OID - 1))
1595 } else if (fid_seq_is_norm(seq)) {
1597 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1600 CERROR("%s : invalid o_seq "DOSTID"\n",
1601 ofd_name(ofd), POSTID(&oa->o_oi));
1602 GOTO(out, rc = -EINVAL);
1607 CERROR("%s: invalid precreate request for "
1608 DOSTID", last_id %llu. "
1609 "Likely MDS last_id corruption\n",
1610 ofd_name(ofd), POSTID(&oa->o_oi),
1611 ofd_seq_last_oid(oseq));
1612 GOTO(out, rc = -EINVAL);
1617 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1623 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1624 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1625 /* don't enforce grant during orphan recovery */
1626 granted = tgt_grant_create(tsi->tsi_env,
1627 ofd_obd(ofd)->obd_self_export,
1632 CDEBUG(D_HA, "%s: failed to acquire grant "
1633 "space for precreate (%lld): rc = %d\n",
1634 ofd_name(ofd), diff, rc);
1639 /* This can happen if a new OST is formatted and installed
1640 * in place of an old one at the same index. Instead of
1641 * precreating potentially millions of deleted old objects
1642 * (possibly filling the OST), only precreate the last batch.
1643 * LFSCK will eventually clean up any orphans. LU-14 */
1644 if (diff > 5 * OST_MAX_PRECREATE) {
1645 diff = OST_MAX_PRECREATE / 2;
1646 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1647 "OST replaced or reformatted: "
1648 "LFSCK will clean up",
1651 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1652 "%u larger than the LAST_ID "DOSTID", only "
1653 "precreating the last %lld objects.\n",
1654 ofd_name(ofd), POSTID(&oa->o_oi),
1655 5 * OST_MAX_PRECREATE,
1656 POSTID(&oseq->os_oi), diff);
1657 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1661 next_id = ofd_seq_last_oid(oseq) + 1;
1662 count = ofd_precreate_batch(ofd, (int)diff);
1664 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1665 " at %llu\n", ofd_name(ofd),
1666 count, seq, next_id);
1668 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1669 && ktime_get_seconds() > enough_time) {
1670 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1671 " created at a rate of %d/s\n",
1672 ofd_name(ofd), created, diff + created,
1673 created / DISK_TIMEOUT);
1677 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1678 oseq, count, sync_trans);
1682 } else if (rc < 0) {
1688 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1689 LCONSOLE_WARN("%s: can't create the same count of"
1690 " objects when replaying the request"
1691 " (diff is %lld). see LU-4621\n",
1692 ofd_name(ofd), diff);
1695 /* some objects got created, we can return
1696 * them, even if last creation failed */
1699 CERROR("%s: unable to precreate: rc = %d\n",
1702 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1703 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1704 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1709 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1713 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1716 mutex_unlock(&oseq->os_create_lock);
1719 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1720 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1721 struct lu_fid *fid = &info->fti_fid;
1723 /* For compatible purpose, it needs to convert back to
1724 * OST ID before put it on wire. */
1725 *fid = rep_oa->o_oi.oi_fid;
1726 fid_to_ostid(fid, &rep_oa->o_oi);
1728 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1730 ofd_seq_put(tsi->tsi_env, oseq);
1733 up_read(&ofd->ofd_lastid_rwsem);
1738 * OFD request handler for OST_DESTROY RPC.
1740 * This is OFD-specific part of request handling. It destroys data objects
1741 * related to destroyed object on MDT.
1743 * \param[in] tsi target session environment for this request
1745 * \retval 0 if successful
1746 * \retval negative value on error
1748 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1750 const struct ost_body *body = tsi->tsi_ost_body;
1751 struct ost_body *repbody;
1752 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1753 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1754 struct lu_fid *fid = &fti->fti_fid;
1761 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1764 /* This is old case for clients before Lustre 2.4 */
1765 /* If there's a DLM request, cancel the locks mentioned in it */
1766 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1768 struct ldlm_request *dlm;
1770 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1773 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1776 *fid = body->oa.o_oi.oi_fid;
1777 oid = ostid_id(&body->oa.o_oi);
1780 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1782 /* check that o_misc makes sense */
1783 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1784 count = body->oa.o_misc;
1786 count = 1; /* default case - single destroy */
1788 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1789 POSTID(&body->oa.o_oi), count);
1794 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1795 if (lrc == -ENOENT) {
1797 "%s: destroying non-existent object "DFID"\n",
1798 ofd_name(ofd), PFID(fid));
1799 /* rewrite rc with -ENOENT only if it is 0 */
1802 } else if (lrc != 0) {
1803 CERROR("%s: error destroying object "DFID": %d\n",
1804 ofd_name(ofd), PFID(fid), lrc);
1810 lrc = fid_set_id(fid, oid);
1811 if (unlikely(lrc != 0 && count > 0))
1812 GOTO(out, rc = lrc);
1815 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1821 fid_to_ostid(fid, &repbody->oa.o_oi);
1826 * OFD request handler for OST_STATFS RPC.
1828 * This function gets statfs data from storage as part of request
1831 * \param[in] tsi target session environment for this request
1833 * \retval 0 if successful
1834 * \retval negative value on error
1836 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1838 struct obd_statfs *osfs;
1843 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1845 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1847 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1848 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1850 CERROR("%s: statfs failed: rc = %d\n",
1851 tgt_name(tsi->tsi_tgt), rc);
1853 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1856 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1863 * OFD request handler for OST_SYNC RPC.
1865 * Sync object data or all filesystem data to the disk and pack the
1868 * \param[in] tsi target session environment for this request
1870 * \retval 0 if successful
1871 * \retval negative value on error
1873 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1875 struct ost_body *body = tsi->tsi_ost_body;
1876 struct ost_body *repbody;
1877 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1878 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1879 struct ofd_object *fo = NULL;
1884 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1886 /* if no objid is specified, it means "sync whole filesystem" */
1887 if (!fid_is_zero(&tsi->tsi_fid)) {
1888 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1890 RETURN(PTR_ERR(fo));
1893 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1894 fo != NULL ? ofd_object_child(fo) : NULL,
1895 repbody->oa.o_size, repbody->oa.o_blocks);
1899 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1904 repbody->oa.o_oi = body->oa.o_oi;
1905 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1907 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1909 obdo_from_la(&repbody->oa, &fti->fti_attr,
1912 /* don't return rc from getattr */
1917 ofd_object_put(tsi->tsi_env, fo);
1922 * OFD request handler for OST_PUNCH RPC.
1924 * This is part of request processing. Validate request fields,
1925 * punch (truncate) the given OFD object and pack reply.
1927 * \param[in] tsi target session environment for this request
1929 * \retval 0 if successful
1930 * \retval negative value on error
1932 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1934 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1935 struct ost_body *repbody;
1936 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1937 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1938 struct ldlm_resource *res;
1939 struct ofd_object *fo;
1941 struct lustre_handle lh = { 0, };
1948 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
1950 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1951 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
1953 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1954 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1955 RETURN(err_serious(-EPROTO));
1957 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1958 if (repbody == NULL)
1959 RETURN(err_serious(-ENOMEM));
1961 /* punch start,end are passed in o_size,o_blocks throught wire */
1965 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1968 /* standard truncate optimization: if file body is completely
1969 * destroyed, don't send data back to the server. */
1971 flags |= LDLM_FL_AST_DISCARD_DATA;
1973 repbody->oa.o_oi = oa->o_oi;
1974 repbody->oa.o_valid = OBD_MD_FLID;
1976 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1977 oa->o_flags & OBD_FL_SRVLOCK;
1980 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
1981 end, &lh, LCK_PW, &flags);
1986 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
1987 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
1988 oa->o_valid, start, end);
1990 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
1993 GOTO(out, rc = PTR_ERR(fo));
1995 la_from_obdo(&info->fti_attr, oa,
1996 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
1997 info->fti_attr.la_size = start;
1998 info->fti_attr.la_valid |= LA_SIZE;
2000 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2005 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2009 ofd_object_put(tsi->tsi_env, fo);
2012 tgt_extent_unlock(&lh, LCK_PW);
2014 /* we do not call this before to avoid lu_object_find() in
2015 * ->lvbo_update() holding another reference on the object.
2016 * otherwise concurrent destroy can make the object unavailable
2017 * for 2nd lu_object_find() waiting for the first reference
2018 * to go... deadlock! */
2019 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2022 struct ost_lvb *res_lvb;
2024 ldlm_res_lvbo_update(tsi->tsi_env, res, NULL, 0);
2025 res_lvb = res->lr_lvb_data;
2026 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2027 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2028 ldlm_resource_putref(res);
2034 static int ofd_ladvise_prefetch(const struct lu_env *env,
2035 struct ofd_object *fo,
2036 struct niobuf_local *lnb,
2037 __u64 start, __u64 end, enum dt_bufs_type dbt)
2039 struct ofd_thread_info *info = ofd_info(env);
2040 pgoff_t start_index, end_index, pages;
2041 struct niobuf_remote rnb;
2042 unsigned long nr_local;
2048 ofd_read_lock(env, fo);
2049 if (!ofd_object_exists(fo))
2050 GOTO(out_unlock, rc = -ENOENT);
2052 rc = ofd_attr_get(env, fo, &info->fti_attr);
2054 GOTO(out_unlock, rc);
2056 if (end > info->fti_attr.la_size)
2057 end = info->fti_attr.la_size;
2060 GOTO(out_unlock, rc);
2062 /* We need page aligned offset and length */
2063 start_index = start >> PAGE_SHIFT;
2064 end_index = (end - 1) >> PAGE_SHIFT;
2065 pages = end_index - start_index + 1;
2067 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2068 PTLRPC_MAX_BRW_PAGES;
2069 rnb.rnb_offset = start_index << PAGE_SHIFT;
2070 rnb.rnb_len = nr_local << PAGE_SHIFT;
2071 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, dbt);
2072 if (unlikely(rc < 0))
2075 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2076 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2079 start_index += nr_local;
2084 ofd_read_unlock(env, fo);
2089 * OFD request handler for OST_LADVISE RPC.
2091 * Tune cache or perfetch policies according to advices.
2093 * \param[in] tsi target session environment for this request
2095 * \retval 0 if successful
2096 * \retval negative errno on error
2098 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2100 struct ptlrpc_request *req = tgt_ses_req(tsi);
2101 struct obd_export *exp = tsi->tsi_exp;
2102 struct ofd_device *ofd = ofd_exp(exp);
2103 struct ost_body *body, *repbody;
2104 struct ofd_thread_info *info;
2105 struct ofd_object *fo;
2106 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2107 const struct lu_env *env = svc_thread->t_env;
2108 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2109 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2110 struct lu_ladvise *ladvise;
2112 struct ladvise_hdr *ladvise_hdr;
2113 struct obd_ioobj ioo;
2114 struct lustre_handle lockh = { 0 };
2117 struct dt_object *dob;
2123 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2124 body = tsi->tsi_ost_body;
2126 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2127 RETURN(err_serious(-EPROTO));
2129 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2130 &RMF_OST_LADVISE_HDR);
2131 if (ladvise_hdr == NULL)
2132 RETURN(err_serious(-EPROTO));
2134 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2135 ladvise_hdr->lah_count < 1)
2136 RETURN(err_serious(-EPROTO));
2138 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2139 RETURN(err_serious(-EPROTO));
2141 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2142 if (ladvise == NULL)
2143 RETURN(err_serious(-EPROTO));
2145 num_advise = req_capsule_get_size(&req->rq_pill,
2146 &RMF_OST_LADVISE, RCL_CLIENT) /
2148 if (num_advise < ladvise_hdr->lah_count)
2149 RETURN(err_serious(-EPROTO));
2151 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2152 repbody->oa = body->oa;
2154 info = ofd_info_init(env, exp);
2156 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2157 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2161 fo = ofd_object_find(env, ofd, &info->fti_fid);
2166 LASSERT(fo != NULL);
2167 dob = ofd_object_child(fo);
2169 if (ptlrpc_connection_is_local(exp->exp_connection))
2170 dbt |= DT_BUFS_TYPE_LOCAL;
2172 for (i = 0; i < num_advise; i++, ladvise++) {
2173 start = ladvise->lla_start;
2174 end = ladvise->lla_end;
2176 rc = err_serious(-EPROTO);
2180 /* Handle different advice types */
2181 switch (ladvise->lla_advice) {
2185 case LU_LADVISE_WILLREAD:
2189 ioo.ioo_oid = body->oa.o_oi;
2191 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2192 &tsi->tsi_resid, start, end - 1,
2193 &lockh, LCK_PR, &flags);
2197 req->rq_status = ofd_ladvise_prefetch(env, fo,
2200 tgt_extent_unlock(&lockh, LCK_PR);
2202 case LU_LADVISE_DONTNEED:
2203 rc = dt_ladvise(env, dob, ladvise->lla_start,
2204 ladvise->lla_end, LU_LADVISE_DONTNEED);
2211 ofd_object_put(env, fo);
2212 req->rq_status = rc;
2217 * OFD request handler for OST_QUOTACTL RPC.
2219 * This is part of request processing to validate incoming request fields,
2220 * get the requested data from OSD and pack reply.
2222 * \param[in] tsi target session environment for this request
2224 * \retval 0 if successful
2225 * \retval negative value on error
2227 static int ofd_quotactl(struct tgt_session_info *tsi)
2229 struct obd_quotactl *oqctl, *repoqc;
2230 struct lu_nodemap *nodemap;
2236 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2238 RETURN(err_serious(-EPROTO));
2240 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2242 RETURN(err_serious(-ENOMEM));
2246 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2247 if (IS_ERR(nodemap))
2248 RETURN(PTR_ERR(nodemap));
2251 if (oqctl->qc_type == USRQUOTA)
2252 id = nodemap_map_id(nodemap, NODEMAP_UID,
2253 NODEMAP_CLIENT_TO_FS,
2255 else if (oqctl->qc_type == GRPQUOTA)
2256 id = nodemap_map_id(nodemap, NODEMAP_GID,
2257 NODEMAP_CLIENT_TO_FS,
2260 nodemap_putref(nodemap);
2262 if (repoqc->qc_id != id)
2263 swap(repoqc->qc_id, id);
2265 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2267 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2270 if (repoqc->qc_id != id)
2271 swap(repoqc->qc_id, id);
2277 * Calculate the amount of time for lock prolongation.
2279 * This is helper for ofd_prolong_extent_locks() function to get
2280 * the timeout extra time.
2282 * \param[in] req current request
2284 * \retval amount of time to extend the timeout with
2286 static inline time64_t prolong_timeout(struct ptlrpc_request *req)
2288 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2289 time64_t req_timeout;
2292 return obd_timeout / 2;
2294 req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec;
2295 return max_t(time64_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
2300 * Prolong lock timeout for the given extent.
2302 * This function finds all locks related with incoming request and
2303 * prolongs their timeout.
2305 * If a client is holding a lock for a long time while it sends
2306 * read or write RPCs to the OST for the object under this lock,
2307 * then we don't want the OST to evict the client. Otherwise,
2308 * if the network or disk is very busy then the client may not
2309 * be able to make any progress to clear out dirty pages under
2310 * the lock and the application will fail.
2312 * Every time a Bulk Read/Write (BRW) request arrives for the object
2313 * covered by the lock, extend the timeout on that lock. The RPC should
2314 * contain a lock handle for the lock it is using, but this
2315 * isn't handled correctly by all client versions, and the
2316 * request may cover multiple locks.
2318 * \param[in] tsi target session environment for this request
2319 * \param[in] data struct of data to prolong locks
2322 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2323 struct ldlm_prolong_args *data)
2325 struct obdo *oa = &tsi->tsi_ost_body->oa;
2326 struct ldlm_lock *lock;
2330 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2331 data->lpa_export = tsi->tsi_exp;
2332 data->lpa_resid = tsi->tsi_resid;
2334 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2335 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2336 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2337 data->lpa_extent.end);
2339 if (oa->o_valid & OBD_MD_FLHANDLE) {
2340 /* mostly a request should be covered by only one lock, try
2342 lock = ldlm_handle2lock(&oa->o_handle);
2344 /* Fast path to check if the lock covers the whole IO
2345 * region exclusively. */
2346 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2347 &data->lpa_extent)) {
2349 LASSERT(lock->l_export == data->lpa_export);
2350 ldlm_lock_prolong_one(lock, data);
2351 LDLM_LOCK_PUT(lock);
2352 if (data->lpa_locks_cnt > 0)
2354 /* The lock was destroyed probably lets try
2357 lock->l_last_used = ktime_get();
2358 LDLM_LOCK_PUT(lock);
2363 ldlm_resource_prolong(data);
2368 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2370 * Determine if \a lock and the lock from request \a req are equivalent
2371 * by comparing their resource names, modes, and extents.
2373 * It is used to give priority to read and write RPCs being done
2374 * under this lock so that the client can drop the contended
2375 * lock more quickly and let other clients use it. This improves
2376 * overall performance in the case where the first client gets a
2377 * very large lock extent that prevents other clients from
2378 * submitting their writes.
2380 * \param[in] req ptlrpc_request being processed
2381 * \param[in] lock contended lock to match
2383 * \retval 1 if lock is matched
2384 * \retval 0 otherwise
2386 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2387 struct ldlm_lock *lock)
2389 struct niobuf_remote *rnb;
2390 struct obd_ioobj *ioo;
2391 enum ldlm_mode mode;
2392 struct ldlm_extent ext;
2393 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2397 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2398 LASSERT(ioo != NULL);
2400 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2401 LASSERT(rnb != NULL);
2403 ext.start = rnb->rnb_offset;
2404 rnb += ioo->ioo_bufcnt - 1;
2405 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2407 LASSERT(lock->l_resource != NULL);
2408 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2411 /* a bulk write can only hold a reference on a PW extent lock
2414 mode = LCK_PW | LCK_GROUP;
2415 if (opc == OST_READ)
2416 /* whereas a bulk read can be protected by either a PR or PW
2420 if (!(lock->l_granted_mode & mode))
2423 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2427 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2429 * Check for whether the given PTLRPC request (\a req) is blocking
2430 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2433 * \param[in] req the incoming request
2435 * \retval 1 if \a req is blocking an LDLM lock cancel
2436 * \retval 0 if it is not
2437 * \retval -ESTALE if lock is not found
2439 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2441 struct tgt_session_info *tsi;
2442 struct obd_ioobj *ioo;
2443 struct niobuf_remote *rnb;
2445 struct ldlm_prolong_args pa = { 0 };
2449 /* Don't use tgt_ses_info() to get session info, because lock_match()
2450 * can be called while request has no processing thread yet. */
2451 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2454 * Use LASSERT below because malformed RPCs should have
2455 * been filtered out in tgt_hpreq_handler().
2457 opc = lustre_msg_get_opc(req->rq_reqmsg);
2458 LASSERT(opc == OST_READ || opc == OST_WRITE);
2460 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2461 LASSERT(ioo != NULL);
2463 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2464 LASSERT(rnb != NULL);
2465 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2467 pa.lpa_mode = LCK_PW | LCK_GROUP;
2468 if (opc == OST_READ)
2469 pa.lpa_mode |= LCK_PR;
2471 pa.lpa_extent.start = rnb->rnb_offset;
2472 rnb += ioo->ioo_bufcnt - 1;
2473 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2475 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
2476 " (%llu->%llu)\n", tgt_name(tsi->tsi_tgt),
2477 current->comm, PFID(&tsi->tsi_fid), pa.lpa_extent.start,
2480 ofd_prolong_extent_locks(tsi, &pa);
2482 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2483 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2485 if (pa.lpa_blocks_cnt > 0)
2488 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2492 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2494 * Called after the request has been handled. It refreshes lock timeout again
2495 * so that client has more time to send lock cancel RPC.
2497 * \param[in] req request which is being processed.
2499 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2501 ofd_rw_hpreq_check(req);
2505 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2507 * This function checks if the given lock is the same by its resname, mode
2508 * and extent as one taken from the request.
2509 * It is used to give priority to punch/truncate RPCs that might lead to
2510 * the fastest release of that lock when a lock is contended.
2512 * \param[in] req ptlrpc_request being processed
2513 * \param[in] lock contended lock to match
2515 * \retval 1 if lock is matched
2516 * \retval 0 otherwise
2518 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2519 struct ldlm_lock *lock)
2521 struct tgt_session_info *tsi;
2523 struct ldlm_extent ext;
2527 /* Don't use tgt_ses_info() to get session info, because lock_match()
2528 * can be called while request has no processing thread yet. */
2529 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2532 * Use LASSERT below because malformed RPCs should have
2533 * been filtered out in tgt_hpreq_handler().
2535 LASSERT(tsi->tsi_ost_body != NULL);
2536 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2537 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2540 oa = &tsi->tsi_ost_body->oa;
2541 ext.start = oa->o_size;
2542 ext.end = oa->o_blocks;
2544 LASSERT(lock->l_resource != NULL);
2545 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2548 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2551 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2555 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2557 * High-priority queue request check for whether the given punch request
2558 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2559 * covered by an LDLM lock.
2563 * \param[in] req the incoming request
2565 * \retval 1 if \a req is blocking an LDLM lock cancel
2566 * \retval 0 if it is not
2567 * \retval -ESTALE if lock is not found
2569 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2571 struct tgt_session_info *tsi;
2573 struct ldlm_prolong_args pa = { 0 };
2577 /* Don't use tgt_ses_info() to get session info, because lock_match()
2578 * can be called while request has no processing thread yet. */
2579 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2580 LASSERT(tsi != NULL);
2581 oa = &tsi->tsi_ost_body->oa;
2583 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2584 oa->o_flags & OBD_FL_SRVLOCK));
2586 pa.lpa_mode = LCK_PW | LCK_GROUP;
2587 pa.lpa_extent.start = oa->o_size;
2588 pa.lpa_extent.end = oa->o_blocks;
2591 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2592 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2593 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2595 ofd_prolong_extent_locks(tsi, &pa);
2597 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2598 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2600 if (pa.lpa_blocks_cnt > 0)
2603 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2607 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2609 * Called after the request has been handled. It refreshes lock timeout again
2610 * so that client has more time to send lock cancel RPC.
2612 * \param[in] req request which is being processed.
2614 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2616 ofd_punch_hpreq_check(req);
2619 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2620 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2621 .hpreq_check = ofd_rw_hpreq_check,
2622 .hpreq_fini = ofd_rw_hpreq_fini
2625 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2626 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2627 .hpreq_check = ofd_punch_hpreq_check,
2628 .hpreq_fini = ofd_punch_hpreq_fini
2632 * Assign high priority operations to an IO request.
2634 * Check if the incoming request is a candidate for
2635 * high-priority processing. If it is, assign it a high
2636 * priority operations table.
2638 * \param[in] tsi target session environment for this request
2640 static void ofd_hp_brw(struct tgt_session_info *tsi)
2642 struct niobuf_remote *rnb;
2643 struct obd_ioobj *ioo;
2647 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2648 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2649 if (ioo->ioo_bufcnt > 0) {
2650 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2651 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2653 /* no high priority if server lock is needed */
2654 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2655 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2659 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2663 * Assign high priority operations to an punch request.
2665 * Check if the incoming request is a candidate for
2666 * high-priority processing. If it is, assign it a high
2667 * priority operations table.
2669 * \param[in] tsi target session environment for this request
2671 static void ofd_hp_punch(struct tgt_session_info *tsi)
2673 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2674 /* no high-priority if server lock is needed */
2675 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2676 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2677 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2678 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2680 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2683 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2684 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2685 #define OST_BRW_READ OST_READ
2686 #define OST_BRW_WRITE OST_WRITE
2689 * Table of OFD-specific request handlers
2691 * This table contains all opcodes accepted by OFD and
2692 * specifies handlers for them. The tgt_request_handler()
2693 * uses such table from each target to process incoming
2696 static struct tgt_handler ofd_tgt_handlers[] = {
2697 TGT_RPC_HANDLER(OST_FIRST_OPC,
2698 0, OST_CONNECT, tgt_connect,
2699 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2700 TGT_RPC_HANDLER(OST_FIRST_OPC,
2701 0, OST_DISCONNECT, tgt_disconnect,
2702 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2703 TGT_RPC_HANDLER(OST_FIRST_OPC,
2704 0, OST_SET_INFO, ofd_set_info_hdl,
2705 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2706 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2707 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
2708 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2709 OST_SETATTR, ofd_setattr_hdl),
2710 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2711 OST_CREATE, ofd_create_hdl),
2712 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2713 OST_DESTROY, ofd_destroy_hdl),
2714 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
2715 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2716 OST_BRW_READ, tgt_brw_read,
2718 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2719 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
2721 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2722 OST_PUNCH, ofd_punch_hdl,
2724 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
2725 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
2726 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_LADVISE, ofd_ladvise_hdl),
2729 static struct tgt_opc_slice ofd_common_slice[] = {
2731 .tos_opc_start = OST_FIRST_OPC,
2732 .tos_opc_end = OST_LAST_OPC,
2733 .tos_hs = ofd_tgt_handlers
2736 .tos_opc_start = OBD_FIRST_OPC,
2737 .tos_opc_end = OBD_LAST_OPC,
2738 .tos_hs = tgt_obd_handlers
2741 .tos_opc_start = LDLM_FIRST_OPC,
2742 .tos_opc_end = LDLM_LAST_OPC,
2743 .tos_hs = tgt_dlm_handlers
2746 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2747 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2748 .tos_hs = tgt_out_handlers
2751 .tos_opc_start = SEQ_FIRST_OPC,
2752 .tos_opc_end = SEQ_LAST_OPC,
2753 .tos_hs = seq_handlers
2756 .tos_opc_start = LFSCK_FIRST_OPC,
2757 .tos_opc_end = LFSCK_LAST_OPC,
2758 .tos_hs = tgt_lfsck_handlers
2761 .tos_opc_start = SEC_FIRST_OPC,
2762 .tos_opc_end = SEC_LAST_OPC,
2763 .tos_hs = tgt_sec_ctx_handlers
2770 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2771 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2774 * Implementation of lu_context_key::lct_key_exit.
2776 * Optional method called on lu_context_exit() for all allocated
2778 * It is used in OFD to sanitize context values which may be re-used
2779 * during another request processing by the same thread.
2781 * \param[in] ctx execution context
2782 * \param[in] key context key
2783 * \param[in] data ofd_thread_info
2785 static void ofd_key_exit(const struct lu_context *ctx,
2786 struct lu_context_key *key, void *data)
2788 struct ofd_thread_info *info = data;
2790 info->fti_env = NULL;
2791 info->fti_exp = NULL;
2794 info->fti_pre_version = 0;
2796 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2799 struct lu_context_key ofd_thread_key = {
2800 .lct_tags = LCT_DT_THREAD,
2801 .lct_init = ofd_key_init,
2802 .lct_fini = ofd_key_fini,
2803 .lct_exit = ofd_key_exit
2807 * Initialize OFD device according to parameters in the config log \a cfg.
2809 * This is the main starting point of OFD initialization. It fills all OFD
2810 * parameters with their initial values and calls other initializing functions
2811 * to set up all OFD subsystems.
2813 * \param[in] env execution environment
2814 * \param[in] m OFD device
2815 * \param[in] ldt LU device type of OFD
2816 * \param[in] cfg configuration log
2818 * \retval 0 if successful
2819 * \retval negative value on error
2821 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2822 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2824 const char *dev = lustre_cfg_string(cfg, 0);
2825 struct ofd_thread_info *info = NULL;
2826 struct obd_device *obd;
2827 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2829 struct nm_config_file *nodemap_config;
2830 struct obd_device_target *obt;
2835 obd = class_name2obd(dev);
2837 CERROR("Cannot find obd with name %s\n", dev);
2841 rc = lu_env_refill((struct lu_env *)env);
2846 obt->obt_magic = OBT_MAGIC;
2848 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2849 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2851 spin_lock_init(&m->ofd_flags_lock);
2852 m->ofd_raid_degraded = 0;
2853 m->ofd_checksum_t10pi_enforce = 0;
2854 m->ofd_sync_journal = 0;
2856 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2858 m->ofd_seq_count = 0;
2859 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2860 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2861 spin_lock_init(&m->ofd_inconsistency_lock);
2863 spin_lock_init(&m->ofd_batch_lock);
2864 init_rwsem(&m->ofd_lastid_rwsem);
2866 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2867 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2868 /* set this lu_device to obd, because error handling need it */
2869 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2871 rc = ofd_tunables_init(m);
2873 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2877 /* No connection accepted until configurations will finish */
2878 spin_lock(&obd->obd_dev_lock);
2879 obd->obd_no_conn = 1;
2880 spin_unlock(&obd->obd_dev_lock);
2881 obd->obd_replayable = 1;
2882 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2883 char *str = lustre_cfg_string(cfg, 4);
2885 if (strchr(str, 'n')) {
2886 CWARN("%s: recovery disabled\n", obd->obd_name);
2887 obd->obd_replayable = 0;
2891 info = ofd_info_init(env, NULL);
2893 GOTO(err_fini_proc, rc = -EFAULT);
2895 rc = ofd_stack_init(env, m, cfg);
2897 CERROR("Can't init device stack, rc %d\n", rc);
2898 GOTO(err_fini_proc, rc);
2901 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
2902 ofd_procfs_add_brw_stats_symlink(m);
2905 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2906 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2907 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2908 LDLM_NAMESPACE_SERVER,
2909 LDLM_NAMESPACE_GREEDY,
2911 if (m->ofd_namespace == NULL)
2912 GOTO(err_fini_stack, rc = -ENOMEM);
2913 /* set obd_namespace for compatibility with old code */
2914 obd->obd_namespace = m->ofd_namespace;
2915 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2916 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2917 m->ofd_namespace->ns_lvbp = m;
2919 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2920 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2922 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2923 OBD_FAIL_OST_ALL_REQUEST_NET,
2924 OBD_FAIL_OST_ALL_REPLY_NET);
2926 GOTO(err_free_ns, rc);
2928 tgd->tgd_reserved_pcnt = 0;
2930 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
2931 m->ofd_cksum_types_supported =
2932 obd_cksum_types_supported_server(obd->obd_name);
2933 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2934 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
2935 OFD_PRECREATE_SMALL_FS)
2936 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2938 rc = ofd_fs_setup(env, m, obd);
2940 GOTO(err_fini_lut, rc);
2942 fid.f_seq = FID_SEQ_LOCAL_NAME;
2945 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
2948 GOTO(err_fini_fs, rc);
2950 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
2952 if (IS_ERR(nodemap_config)) {
2953 rc = PTR_ERR(nodemap_config);
2955 GOTO(err_fini_los, rc);
2957 obt->obt_nodemap_config_file = nodemap_config;
2960 rc = ofd_start_inconsistency_verification_thread(m);
2962 GOTO(err_fini_nm, rc);
2964 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
2969 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
2970 obt->obt_nodemap_config_file = NULL;
2972 local_oid_storage_fini(env, m->ofd_los);
2975 ofd_fs_cleanup(env, m);
2977 tgt_fini(env, &m->ofd_lut);
2979 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
2980 obd->obd_namespace = m->ofd_namespace = NULL;
2982 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
2989 * Stop the OFD device
2991 * This function stops the OFD device and all its subsystems.
2992 * This is the end of OFD lifecycle.
2994 * \param[in] env execution environment
2995 * \param[in] m OFD device
2997 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
2999 struct obd_device *obd = ofd_obd(m);
3000 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3001 struct lfsck_stop stop;
3003 stop.ls_status = LS_PAUSED;
3005 lfsck_stop(env, m->ofd_osd, &stop);
3006 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3007 target_recovery_fini(obd);
3008 if (m->ofd_namespace != NULL)
3009 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3010 d->ld_obd->obd_force);
3012 obd_exports_barrier(obd);
3013 obd_zombie_barrier();
3015 tgt_fini(env, &m->ofd_lut);
3016 ofd_stop_inconsistency_verification_thread(m);
3017 lfsck_degister(env, m->ofd_osd);
3018 ofd_fs_cleanup(env, m);
3019 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3020 obd->u.obt.obt_nodemap_config_file = NULL;
3022 if (m->ofd_namespace != NULL) {
3023 ldlm_namespace_free_post(m->ofd_namespace);
3024 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3027 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3029 LASSERT(atomic_read(&d->ld_ref) == 0);
3030 server_put_mount(obd->obd_name, true);
3035 * Implementation of lu_device_type_operations::ldto_device_fini.
3037 * Finalize device. Dual to ofd_device_init(). It is called from
3038 * obd_precleanup() and stops the current device.
3040 * \param[in] env execution environment
3041 * \param[in] d LU device of OFD
3045 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3046 struct lu_device *d)
3049 ofd_fini(env, ofd_dev(d));
3054 * Implementation of lu_device_type_operations::ldto_device_free.
3056 * Free OFD device. Dual to ofd_device_alloc().
3058 * \param[in] env execution environment
3059 * \param[in] d LU device of OFD
3063 static struct lu_device *ofd_device_free(const struct lu_env *env,
3064 struct lu_device *d)
3066 struct ofd_device *m = ofd_dev(d);
3068 dt_device_fini(&m->ofd_dt_dev);
3074 * Implementation of lu_device_type_operations::ldto_device_alloc.
3076 * This function allocates the new OFD device. It is called from
3077 * obd_setup() if OBD device had lu_device_type defined.
3079 * \param[in] env execution environment
3080 * \param[in] t lu_device_type of OFD device
3081 * \param[in] cfg configuration log
3083 * \retval pointer to the lu_device of just allocated OFD
3084 * \retval ERR_PTR of return value on error
3086 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3087 struct lu_device_type *t,
3088 struct lustre_cfg *cfg)
3090 struct ofd_device *m;
3091 struct lu_device *l;
3096 return ERR_PTR(-ENOMEM);
3098 l = &m->ofd_dt_dev.dd_lu_dev;
3099 dt_device_init(&m->ofd_dt_dev, t);
3100 rc = ofd_init0(env, m, t, cfg);
3102 ofd_device_free(env, l);
3109 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3110 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3112 static struct lu_device_type_operations ofd_device_type_ops = {
3113 .ldto_init = ofd_type_init,
3114 .ldto_fini = ofd_type_fini,
3116 .ldto_start = ofd_type_start,
3117 .ldto_stop = ofd_type_stop,
3119 .ldto_device_alloc = ofd_device_alloc,
3120 .ldto_device_free = ofd_device_free,
3121 .ldto_device_fini = ofd_device_fini
3124 static struct lu_device_type ofd_device_type = {
3125 .ldt_tags = LU_DEVICE_DT,
3126 .ldt_name = LUSTRE_OST_NAME,
3127 .ldt_ops = &ofd_device_type_ops,
3128 .ldt_ctx_tags = LCT_DT_THREAD
3132 * Initialize OFD module.
3134 * This function is called upon module loading. It registers OFD device type
3135 * and prepares all in-memory structures used by all OFD devices.
3137 * \retval 0 if successful
3138 * \retval negative value on error
3140 static int __init ofd_init(void)
3144 rc = lu_kmem_init(ofd_caches);
3148 rc = ofd_fmd_init();
3150 lu_kmem_fini(ofd_caches);
3154 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3155 LUSTRE_OST_NAME, &ofd_device_type);
3162 * This function is called upon OFD module unloading.
3163 * It frees all related structures and unregisters OFD device type.
3165 static void __exit ofd_exit(void)
3168 lu_kmem_fini(ofd_caches);
3169 class_unregister_type(LUSTRE_OST_NAME);
3172 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3173 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3174 MODULE_VERSION(LUSTRE_VERSION_STRING);
3175 MODULE_LICENSE("GPL");
3177 module_init(ofd_init);
3178 module_exit(ofd_exit);