4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/ofd/ofd_dev.c
33 * This file contains OSD API methods for OBD Filter Device (OFD),
34 * request handlers and supplemental functions to set OFD up and clean it up.
36 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
37 * Author: Mike Pershin <mike.pershin@intel.com>
38 * Author: Johann Lombardi <johann.lombardi@intel.com>
41 * The OBD Filter Device (OFD) module belongs to the Object Storage
42 * Server stack and connects the RPC oriented Unified Target (TGT)
43 * layer (see lustre/include/lu_target.h) to the storage oriented OSD
44 * layer (see Documentation/osd-api.txt).
52 * OFD implements the LU and OBD device APIs and is responsible for:
54 * - Handling client requests (create, destroy, bulk IO, setattr,
55 * get_info, set_info, statfs) for the objects belonging to the OST
56 * (together with TGT).
58 * - Providing grant space management which allows clients to reserve
59 * disk space for data writeback. OFD tracks grants on global and
62 * - Handling object precreation requests from MDTs.
64 * - Operating the LDLM service that allows clients to maintain object
65 * data cache coherence.
68 #define DEBUG_SUBSYSTEM S_FILTER
70 #include <obd_class.h>
71 #include <obd_cksum.h>
72 #include <uapi/linux/lustre/lustre_param.h>
73 #include <lustre_fid.h>
74 #include <lustre_lfsck.h>
75 #include <lustre_dlm.h>
76 #include <lustre_quota.h>
77 #include <lustre_nodemap.h>
78 #include <lustre_log.h>
79 #include <linux/falloc.h>
81 #include "ofd_internal.h"
83 /* Slab for OFD object allocation */
84 static struct kmem_cache *ofd_object_kmem;
85 static struct lu_kmem_descr ofd_caches[] = {
87 .ckd_cache = &ofd_object_kmem,
88 .ckd_name = "ofd_obj",
89 .ckd_size = sizeof(struct ofd_object)
97 * Connect OFD to the next device in the stack.
99 * This function is used for device stack configuration and links OFD
100 * device with bottom OSD device.
102 * \param[in] env execution environment
103 * \param[in] m OFD device
104 * \param[in] next name of next device in the stack
105 * \param[out] exp export to return
107 * \retval 0 and export in \a exp if successful
108 * \retval negative value on error
110 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
111 const char *next, struct obd_export **exp)
113 struct obd_connect_data *data = NULL;
114 struct obd_device *obd;
120 GOTO(out, rc = -ENOMEM);
122 obd = class_name2obd(next);
124 CERROR("%s: can't locate next device: %s\n",
126 GOTO(out, rc = -ENOTCONN);
129 data->ocd_connect_flags = OBD_CONNECT_VERSION;
130 data->ocd_version = LUSTRE_VERSION_CODE;
132 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
134 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
135 ofd_name(m), next, rc);
139 m->ofd_dt_dev.dd_lu_dev.ld_site =
140 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
141 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
142 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
143 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
152 * Initialize stack of devices.
154 * This function initializes OFD-OSD device stack to serve OST requests
156 * \param[in] env execution environment
157 * \param[in] m OFD device
158 * \param[in] cfg Lustre config for this server
160 * \retval 0 if successful
161 * \retval negative value on error
163 static int ofd_stack_init(const struct lu_env *env,
164 struct ofd_device *m, struct lustre_cfg *cfg,
167 const char *dev = lustre_cfg_string(cfg, 0);
169 struct ofd_thread_info *info = ofd_info(env);
170 struct lustre_mount_info *lmi;
171 struct lustre_mount_data *lmd;
177 lmi = server_get_mount(dev);
179 CERROR("Cannot get mount info for %s!\n", dev);
183 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
185 if (lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
186 m->ofd_skip_lfsck = 1;
187 if (lmd->lmd_flags & LMD_FLG_NO_PRECREATE)
188 m->ofd_no_precreate = 1;
189 *lmd_flags = lmd->lmd_flags;
192 /* find bottom osd */
193 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
197 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
198 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
199 OBD_FREE(osdname, MTI_NAME_MAXLEN);
203 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
205 m->ofd_osd = lu2dt_dev(d);
207 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
208 "%s-osd", lustre_cfg_string(cfg, 0));
214 * Finalize the device stack OFD-OSD.
216 * This function cleans OFD-OSD device stack and
217 * disconnects OFD from the OSD.
219 * \param[in] env execution environment
220 * \param[in] m OFD device
221 * \param[in] top top device of stack
223 * \retval 0 if successful
224 * \retval negative value on error
226 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
227 struct lu_device *top)
229 struct obd_device *obd = ofd_obd(m);
230 struct lustre_cfg_bufs bufs;
231 struct lustre_cfg *lcfg;
236 lu_site_purge(env, top->ld_site, ~0);
237 /* process cleanup, pass mdt obd name to get obd umount flags */
238 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
243 lustre_cfg_bufs_set_string(&bufs, 1, flags);
244 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
247 lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs);
250 top->ld_ops->ldo_process_config(env, top, lcfg);
251 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
253 if (m->ofd_los != NULL) {
254 local_oid_storage_fini(env, m->ofd_los);
258 lu_site_purge(env, top->ld_site, ~0);
259 lu_site_print(env, top->ld_site, &top->ld_site->ls_obj_hash.nelems,
260 D_OTHER, lu_cdebug_printer);
261 LASSERT(m->ofd_osd_exp);
262 obd_disconnect(m->ofd_osd_exp);
267 static void ofd_stack_pre_fini(const struct lu_env *env, struct ofd_device *m,
268 struct lu_device *top)
270 struct lustre_cfg_bufs bufs;
271 struct lustre_cfg *lcfg;
276 lustre_cfg_bufs_reset(&bufs, ofd_name(m));
277 lustre_cfg_bufs_set_string(&bufs, 1, NULL);
278 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
280 CERROR("%s: failed to trigger LCFG_PRE_CLEANUP\n", ofd_name(m));
282 lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, &bufs);
283 top->ld_ops->ldo_process_config(env, top, lcfg);
284 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
285 lcfg->lcfg_buflens));
291 /* For interoperability, see mdt_interop_param[]. */
292 static struct cfg_interop_param ofd_interop_param[] = {
293 { "ost.quota_type", NULL },
298 * Check if parameters are symlinks to the OSD.
300 * Some parameters were moved from ofd to osd and only their
301 * symlinks were kept in ofd by LU-3106. They are:
302 * -writehthrough_cache_enable
303 * -readcache_max_filesize
307 * Since they are not included by the static lprocfs var list, a pre-check
308 * is added for them to avoid "unknown param" errors. If they are matched
309 * in this check, they will be passed to the OSD directly.
311 * \param[in] param parameters to check
313 * \retval true if param is symlink to OSD param
316 static bool match_symlink_param(char *param)
321 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
322 sval = strchr(param, '=');
324 paramlen = sval - param;
325 if (strncmp(param, "brw_stats", paramlen) == 0)
334 * Process various configuration parameters.
336 * This function is used by MGS to process specific configurations and
337 * pass them through to the next device in server stack, i.e. the OSD.
339 * \param[in] env execution environment
340 * \param[in] d LU device of OFD
341 * \param[in] cfg parameters to process
343 * \retval 0 if successful
344 * \retval negative value on error
346 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
347 struct lustre_cfg *cfg)
349 struct ofd_device *m = ofd_dev(d);
350 struct dt_device *dt_next = m->ofd_osd;
351 struct lu_device *next = &dt_next->dd_lu_dev;
356 switch (cfg->lcfg_command) {
358 /* For interoperability */
359 struct cfg_interop_param *ptr = NULL;
360 struct lustre_cfg *old_cfg = NULL;
364 param = lustre_cfg_string(cfg, 1);
366 CERROR("param is empty\n");
371 ptr = class_find_old_param(param, ofd_interop_param);
373 if (ptr->new_param == NULL) {
375 CWARN("For interoperability, skip this %s."
376 " It is obsolete.\n", ptr->old_param);
380 CWARN("Found old param %s, changed it to %s.\n",
381 ptr->old_param, ptr->new_param);
384 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
391 if (match_symlink_param(param)) {
392 rc = next->ld_ops->ldo_process_config(env, next, cfg);
396 count = class_modify_config(cfg, PARAM_OST,
397 &d->ld_obd->obd_kset.kobj);
402 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
404 /* we don't understand; pass it on */
405 rc = next->ld_ops->ldo_process_config(env, next, cfg);
408 case LCFG_SPTLRPC_CONF: {
413 /* others are passed further */
414 rc = next->ld_ops->ldo_process_config(env, next, cfg);
421 * Implementation of lu_object_operations::loo_object_init for OFD
423 * Allocate just the next object (OSD) in stack.
425 * \param[in] env execution environment
426 * \param[in] o lu_object of OFD object
427 * \param[in] conf additional configuration parameters, not used here
429 * \retval 0 if successful
430 * \retval negative value on error
432 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
433 const struct lu_object_conf *conf)
435 struct ofd_device *d = ofd_dev(o->lo_dev);
436 struct lu_device *under;
437 struct lu_object *below;
442 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
443 PFID(lu_object_fid(o)));
445 under = &d->ofd_osd->dd_lu_dev;
446 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
448 lu_object_add(o, below);
455 static void ofd_object_free_rcu(struct rcu_head *head)
457 struct ofd_object *of = container_of(head, struct ofd_object,
460 kmem_cache_free(ofd_object_kmem, of);
464 * Implementation of lu_object_operations::loo_object_free.
466 * Finish OFD object lifecycle and free its memory.
468 * \param[in] env execution environment
469 * \param[in] o LU object of OFD object
471 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
473 struct ofd_object *of = ofd_obj(o);
474 struct lu_object_header *h;
479 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
480 PFID(lu_object_fid(o)));
483 lu_object_header_fini(h);
484 OBD_FREE_PRE(of, sizeof(*of), "slab-freed");
485 call_rcu(&of->ofo_header.loh_rcu, ofd_object_free_rcu);
490 * Implementation of lu_object_operations::loo_object_print.
492 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
493 * LU_OBJECT_DEBUG() for more details about the compound object printing.
495 * \param[in] env execution environment
496 * \param[in] cookie opaque data passed to the printer function
497 * \param[in] p printer function to use
498 * \param[in] o LU object of OFD object
500 * \retval 0 if successful
501 * \retval negative value on error
503 static int ofd_object_print(const struct lu_env *env, void *cookie,
504 lu_printer_t p, const struct lu_object *o)
506 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
509 static const struct lu_object_operations ofd_obj_ops = {
510 .loo_object_init = ofd_object_init,
511 .loo_object_free = ofd_object_free,
512 .loo_object_print = ofd_object_print
516 * Implementation of lu_device_operations::lod_object_alloc.
518 * This function allocates OFD part of compound OFD-OSD object and
519 * initializes its header, because OFD is the top device in stack
521 * \param[in] env execution environment
522 * \param[in] hdr object header, NULL for OFD
523 * \param[in] d lu_device
525 * \retval allocated object if successful
526 * \retval NULL value on failed allocation
528 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
529 const struct lu_object_header *hdr,
532 struct ofd_object *of;
536 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
539 struct lu_object_header *h;
541 o = &of->ofo_obj.do_lu;
543 lu_object_header_init(h);
544 lu_object_init(o, h, d);
545 lu_object_add_top(h, o);
546 o->lo_ops = &ofd_obj_ops;
554 * Return the result of LFSCK run to the OFD.
556 * Notify OFD about result of LFSCK run. That may block the new object
557 * creation until problem is fixed by LFSCK.
559 * \param[in] env execution environment
560 * \param[in] data pointer to the OFD device
561 * \param[in] event LFSCK event type
563 * \retval 0 if successful
564 * \retval negative value on unknown event
566 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
567 enum lfsck_events event)
569 struct ofd_device *ofd = data;
570 struct obd_device *obd = ofd_obd(ofd);
573 case LE_LASTID_REBUILDING:
574 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
575 "on the device until the LAST_ID rebuilt successfully.\n",
577 down_write(&ofd->ofd_lastid_rwsem);
578 ofd->ofd_lastid_rebuilding = 1;
579 up_write(&ofd->ofd_lastid_rwsem);
581 case LE_LASTID_REBUILT: {
582 down_write(&ofd->ofd_lastid_rwsem);
583 ofd_seqs_free(env, ofd);
584 ofd->ofd_lastid_rebuilding = 0;
585 ofd->ofd_lastid_gen++;
586 up_write(&ofd->ofd_lastid_rwsem);
587 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
592 CERROR("%s: unknown lfsck event: rc = %d\n",
593 ofd_name(ofd), event);
601 * Implementation of lu_device_operations::ldo_prepare.
603 * This method is called after layer has been initialized and before it starts
604 * serving user requests. In OFD it starts lfsk check routines and initializes
607 * \param[in] env execution environment
608 * \param[in] pdev higher device in stack, NULL for OFD
609 * \param[in] dev lu_device of OFD device
611 * \retval 0 if successful
612 * \retval negative value on error
614 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
615 struct lu_device *dev)
617 struct ofd_thread_info *info;
618 struct ofd_device *ofd = ofd_dev(dev);
619 struct obd_device *obd = ofd_obd(ofd);
620 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
625 info = ofd_info_init(env, NULL);
629 /* initialize lower device */
630 rc = next->ld_ops->ldo_prepare(env, dev, next);
634 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
635 ofd_lfsck_out_notify, ofd, false);
637 CERROR("%s: failed to initialize lfsck: rc = %d\n",
642 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
643 /* The LFSCK instance is registered just now, so it must be there when
644 * register the namespace to such instance. */
645 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
647 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
648 OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_OST_PREPARE_DELAY, OBD_FAIL_ONCE,
649 (OBD_TIMEOUT_DEFAULT + 1) / 4);
650 LASSERT(obd->obd_no_conn);
651 spin_lock(&obd->obd_dev_lock);
652 obd->obd_no_conn = 0;
653 spin_unlock(&obd->obd_dev_lock);
655 if (obd->obd_recovering == 0)
656 ofd_postrecov(env, ofd);
662 * Implementation of lu_device_operations::ldo_recovery_complete.
664 * This method notifies all layers about 'recovery complete' event. That means
665 * device is in full state and consistent. An OFD calculates available grant
666 * space upon this event.
668 * \param[in] env execution environment
669 * \param[in] dev lu_device of OFD device
671 * \retval 0 if successful
672 * \retval negative value on error
674 static int ofd_recovery_complete(const struct lu_env *env,
675 struct lu_device *dev)
677 struct ofd_thread_info *oti = ofd_info(env);
678 struct ofd_device *ofd = ofd_dev(dev);
679 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
685 * Grant space for object precreation on the self export.
686 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
687 * is enough to create 10k objects. More space is then acquired for
688 * precreation in tgt_grant_create().
690 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
691 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
692 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
693 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
694 OBD_CONNECT_GRANT_PARAM;
695 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
697 rc = next->ld_ops->ldo_recovery_complete(env, next);
702 * lu_device_operations matrix for OFD device.
704 static const struct lu_device_operations ofd_lu_ops = {
705 .ldo_object_alloc = ofd_object_alloc,
706 .ldo_process_config = ofd_process_config,
707 .ldo_recovery_complete = ofd_recovery_complete,
708 .ldo_prepare = ofd_prepare,
711 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
713 * Expose OSD statistics to OFD layer.
715 * The osd interfaces to the backend file system exposes useful data
716 * such as brw_stats and read or write cache states. This same data
717 * needs to be exposed into the obdfilter (ofd) layer to maintain
718 * backwards compatibility. This function creates the symlinks in the
719 * proc layer to enable this.
721 * \param[in] ofd OFD device
723 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
725 struct obd_device *obd = ofd_obd(ofd);
726 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
727 struct kobj_type *osd_type;
730 osd_type = get_ktype(&ofd->ofd_osd->dd_kobj);
731 for (i = 0; osd_type->default_attrs[i]; i++) {
732 if (strcmp(osd_type->default_attrs[i]->name,
733 "read_cache_enable") == 0) {
734 ofd->ofd_read_cache_enable =
735 osd_type->default_attrs[i];
738 if (strcmp(osd_type->default_attrs[i]->name,
739 "readcache_max_filesize") == 0) {
740 ofd->ofd_read_cache_max_filesize =
741 osd_type->default_attrs[i];
744 if (strcmp(osd_type->default_attrs[i]->name,
745 "writethrough_cache_enable") == 0) {
746 ofd->ofd_write_cache_enable =
747 osd_type->default_attrs[i];
751 if (obd->obd_proc_entry == NULL)
754 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
755 "../../%s/%s/brw_stats",
756 osd_obd->obd_type->typ_name, obd->obd_name);
761 * Cleanup all procfs entries in OFD.
763 * \param[in] ofd OFD device
765 static void ofd_procfs_fini(struct ofd_device *ofd)
767 struct obd_device *obd = ofd_obd(ofd);
769 tgt_tunables_fini(&ofd->ofd_lut);
770 lprocfs_free_per_client_stats(obd);
771 lprocfs_obd_cleanup(obd);
772 lprocfs_free_obd_stats(obd);
773 lprocfs_job_stats_fini(obd);
777 * Stop SEQ/FID server on OFD.
779 * \param[in] env execution environment
780 * \param[in] ofd OFD device
782 * \retval 0 if successful
783 * \retval negative value on error
785 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
787 return seq_site_fini(env, &ofd->ofd_seq_site);
791 * Start SEQ/FID server on OFD.
793 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
794 * It also connects to the master server to get own FID sequence (SEQ) range
795 * to this particular OFD. Typically that happens when the OST is first
796 * formatted or in the rare case that it exhausts the local sequence range.
798 * The sequence range is allocated out to the MDTs for OST object allocations,
799 * and not directly to the clients.
801 * \param[in] env execution environment
802 * \param[in] ofd OFD device
804 * \retval 0 if successful
805 * \retval negative value on error
807 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
809 struct seq_server_site *ss = &ofd->ofd_seq_site;
810 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
811 char *obd_name = ofd_name(ofd);
813 int len = strlen(obd_name) + 7;
816 ss = &ofd->ofd_seq_site;
817 lu->ld_site->ld_seq_site = ss;
818 ss->ss_lu = lu->ld_site;
819 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
821 OBD_ALLOC(name, len);
825 OBD_ALLOC_PTR(ss->ss_server_seq);
826 if (ss->ss_server_seq == NULL)
827 GOTO(out_name, rc = -ENOMEM);
829 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
830 LUSTRE_SEQ_SERVER, ss);
832 CERROR("%s: seq server init error: rc = %d\n", obd_name, rc);
833 GOTO(out_server, rc);
835 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
837 OBD_ALLOC_PTR(ss->ss_client_seq);
838 if (ss->ss_client_seq == NULL)
839 GOTO(out_server, rc = -ENOMEM);
841 snprintf(name, len, "%s-super", obd_name);
842 seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
845 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
848 seq_client_fini(ss->ss_client_seq);
849 OBD_FREE_PTR(ss->ss_client_seq);
850 ss->ss_client_seq = NULL;
852 seq_server_fini(ss->ss_server_seq, env);
853 OBD_FREE_PTR(ss->ss_server_seq);
854 ss->ss_server_seq = NULL;
863 * OFD request handler for OST_SET_INFO RPC.
865 * This is OFD-specific part of request handling
867 * \param[in] tsi target session environment for this request
869 * \retval 0 if successful
870 * \retval negative value on error
872 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
874 struct ptlrpc_request *req = tgt_ses_req(tsi);
875 struct ost_body *body = NULL, *repbody;
876 void *key, *val = NULL;
877 int keylen, vallen, rc = 0;
878 bool is_grant_shrink;
879 ktime_t kstart = ktime_get();
883 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
885 DEBUG_REQ(D_HA, req, "no set_info key");
886 RETURN(err_serious(-EFAULT));
888 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
891 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
893 DEBUG_REQ(D_HA, req, "no set_info val");
894 RETURN(err_serious(-EFAULT));
896 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
899 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
901 /* In this case the value is actually an RMF_OST_BODY, so we
902 * transmutate the type of this PTLRPC */
903 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
905 rc = req_capsule_server_pack(tsi->tsi_pill);
909 if (is_grant_shrink) {
910 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
913 * Because we already sync grant info with client when
914 * reconnect, grant info will be cleared for resent
915 * req, otherwise, outdated grant count in the rpc
916 * would de-sync grant counters
918 if (lustre_msg_get_flags(req->rq_reqmsg) &
919 (MSG_RESENT | MSG_REPLAY)) {
920 DEBUG_REQ(D_CACHE, req,
921 "clear resent/replay req grant info");
922 body->oa.o_valid &= ~OBD_MD_FLGRANT;
925 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
928 /** handle grant shrink, similar to a read request */
929 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
931 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
933 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
936 CERROR("%s: Unsupported key %s\n",
937 tgt_name(tsi->tsi_tgt), (char *)key);
940 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
941 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
947 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
949 * This function returns a list of extents which describes how a file's
950 * blocks are laid out on the disk.
952 * \param[in] env execution environment
953 * \param[in] ofd OFD device
954 * \param[in] fid FID of object
955 * \param[in] fiemap fiemap structure to fill with data
957 * \retval 0 if \a fiemap is filled with data successfully
958 * \retval negative value on error
960 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
961 struct lu_fid *fid, struct fiemap *fiemap)
963 struct ofd_object *fo;
966 fo = ofd_object_find(env, ofd, fid);
968 CERROR("%s: error finding object "DFID"\n",
969 ofd_name(ofd), PFID(fid));
973 ofd_read_lock(env, fo);
974 if (ofd_object_exists(fo))
975 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
978 ofd_read_unlock(env, fo);
979 ofd_object_put(env, fo);
984 static int ofd_lock_unlock_region(const struct lu_env *env,
985 struct ldlm_namespace *ns,
986 struct ldlm_res_id *res_id,
987 unsigned long long begin,
988 unsigned long long end)
992 struct lustre_handle lh = { 0 };
994 LASSERT(begin <= end);
996 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
1000 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
1001 tgt_data_unlock(&lh, LCK_PR);
1007 * Lock the sparse areas of given resource.
1009 * The locking of sparse areas will cause dirty data to be flushed back from
1010 * clients. This is used when getting the FIEMAP of an object to make sure
1011 * there is no unaccounted cached data on clients.
1013 * This function goes through \a fiemap list of extents and locks only sparse
1014 * areas between extents.
1016 * \param[in] ns LDLM namespace
1017 * \param[in] res_id resource ID
1018 * \param[in] fiemap file extents mapping on disk
1019 * \param[in] locked list head of regions list
1021 * \retval 0 if successful
1022 * \retval negative value on error
1024 static int lock_zero_regions(const struct lu_env *env,
1025 struct ldlm_namespace *ns,
1026 struct ldlm_res_id *res_id,
1027 struct fiemap *fiemap)
1029 __u64 begin = fiemap->fm_start;
1032 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1036 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1037 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1038 if (fiemap_start[i].fe_logical > begin) {
1039 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1040 begin, fiemap_start[i].fe_logical);
1041 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1042 fiemap_start[i].fe_logical);
1047 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1050 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1051 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1052 begin, fiemap->fm_start + fiemap->fm_length);
1053 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1054 fiemap->fm_start + fiemap->fm_length);
1062 * OFD request handler for OST_GET_INFO RPC.
1064 * This is OFD-specific part of request handling. The OFD-specific keys are:
1065 * - KEY_LAST_ID (obsolete)
1069 * This function reads needed data from storage and fills reply with it.
1071 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1072 * and is kept for compatibility.
1074 * \param[in] tsi target session environment for this request
1076 * \retval 0 if successful
1077 * \retval negative value on error
1079 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1081 struct obd_export *exp = tsi->tsi_exp;
1082 struct ofd_device *ofd = ofd_exp(exp);
1083 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1086 int replylen, rc = 0;
1087 ktime_t kstart = ktime_get();
1091 /* this common part for get_info rpc */
1092 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1094 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1095 RETURN(err_serious(-EPROTO));
1097 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1100 if (KEY_IS(KEY_LAST_ID)) {
1102 struct ofd_seq *oseq;
1104 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1105 rc = req_capsule_server_pack(tsi->tsi_pill);
1107 RETURN(err_serious(rc));
1109 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1111 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1112 (u64)exp->exp_filter_data.fed_group);
1116 *last_id = ofd_seq_last_oid(oseq);
1117 ofd_seq_put(tsi->tsi_env, oseq);
1118 } else if (KEY_IS(KEY_FIEMAP)) {
1119 struct ll_fiemap_info_key *fm_key;
1120 struct fiemap *fiemap;
1123 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1125 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1126 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1128 RETURN(err_serious(rc));
1130 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1132 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1134 replylen = fiemap_count_to_size(
1135 fm_key->lfik_fiemap.fm_extent_count);
1136 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1137 RCL_SERVER, replylen);
1139 rc = req_capsule_server_pack(tsi->tsi_pill);
1141 RETURN(err_serious(rc));
1143 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1147 *fiemap = fm_key->lfik_fiemap;
1148 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1150 /* LU-3219: Lock the sparse areas to make sure dirty
1151 * flushed back from client, then call fiemap again. */
1152 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1153 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1154 ost_fid_build_resid(fid, &fti->fti_resid);
1155 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1156 &fti->fti_resid, fiemap);
1158 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1161 } else if (KEY_IS(KEY_LAST_FID)) {
1162 struct ofd_device *ofd = ofd_exp(exp);
1163 struct ofd_seq *oseq;
1167 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1168 rc = req_capsule_server_pack(tsi->tsi_pill);
1170 RETURN(err_serious(rc));
1172 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GET_LAST_FID))
1175 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1177 RETURN(err_serious(-EPROTO));
1179 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1181 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1185 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1186 ostid_seq(&fti->fti_ostid));
1190 rc = ostid_to_fid(fid, &oseq->os_oi,
1191 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1193 GOTO(out_put, rc = -EFAULT);
1195 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1198 ofd_seq_put(tsi->tsi_env, oseq);
1200 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1204 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1205 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1211 * OFD request handler for OST_GETATTR RPC.
1213 * This is OFD-specific part of request handling. It finds the OFD object
1214 * by its FID, gets attributes from storage and packs result to the reply.
1216 * \param[in] tsi target session environment for this request
1218 * \retval 0 if successful
1219 * \retval negative value on error
1221 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1223 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1224 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1225 struct ost_body *repbody;
1226 struct lustre_handle lh = { 0 };
1227 struct ofd_object *fo;
1229 enum ldlm_mode lock_mode = LCK_PR;
1230 ktime_t kstart = ktime_get();
1235 LASSERT(tsi->tsi_ost_body != NULL);
1237 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1238 if (repbody == NULL)
1241 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1242 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1244 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1245 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1248 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1251 rc = tgt_extent_lock(tsi->tsi_env,
1252 tsi->tsi_tgt->lut_obd->obd_namespace,
1253 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1259 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1261 GOTO(out, rc = PTR_ERR(fo));
1263 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1267 obdo_from_la(&repbody->oa, &fti->fti_attr,
1268 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1270 /* Store object version in reply */
1271 curr_version = dt_version_get(tsi->tsi_env,
1272 ofd_object_child(fo));
1273 if ((__s64)curr_version != -EOPNOTSUPP) {
1274 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1275 repbody->oa.o_data_version = curr_version;
1278 if (fo->ofo_ff.ff_layout_version > 0) {
1279 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1280 repbody->oa.o_layout_version =
1281 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1283 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1284 PFID(&tsi->tsi_fid),
1285 repbody->oa.o_layout_version);
1289 ofd_object_put(tsi->tsi_env, fo);
1292 tgt_data_unlock(&lh, lock_mode);
1294 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1295 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1297 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1298 repbody->oa.o_flags = OBD_FL_FLUSH;
1304 * OFD request handler for OST_SETATTR RPC.
1306 * This is OFD-specific part of request handling. It finds the OFD object
1307 * by its FID, sets attributes from request and packs result to the reply.
1309 * \param[in] tsi target session environment for this request
1311 * \retval 0 if successful
1312 * \retval negative value on error
1314 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1316 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1317 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1318 struct ost_body *body = tsi->tsi_ost_body;
1319 struct ost_body *repbody;
1320 struct ldlm_resource *res;
1321 struct ofd_object *fo;
1322 ktime_t kstart = ktime_get();
1327 LASSERT(body != NULL);
1329 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1330 if (repbody == NULL)
1333 repbody->oa.o_oi = body->oa.o_oi;
1334 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1336 /* This would be very bad - accidentally truncating a file when
1337 * changing the time or similar - bug 12203. */
1338 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1339 body->oa.o_size != OBD_OBJECT_EOF) {
1340 static char mdsinum[48];
1342 if (body->oa.o_valid & OBD_MD_FLFID)
1343 snprintf(mdsinum, sizeof(mdsinum) - 1,
1344 "of parent "DFID, body->oa.o_parent_seq,
1345 body->oa.o_parent_oid, 0);
1349 CERROR("%s: setattr from %s is trying to truncate object "DFID
1350 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1351 PFID(&tsi->tsi_fid), mdsinum);
1355 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1357 GOTO(out, rc = PTR_ERR(fo));
1359 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1360 fti->fti_attr.la_valid &= ~LA_TYPE;
1362 /* setting objects attributes (including owner/group) */
1363 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1367 obdo_from_la(&repbody->oa, &fti->fti_attr,
1368 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1370 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1371 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1374 ofd_object_put(tsi->tsi_env, fo);
1377 /* we do not call this before to avoid lu_object_find() in
1378 * ->lvbo_update() holding another reference on the object.
1379 * otherwise concurrent destroy can make the object unavailable
1380 * for 2nd lu_object_find() waiting for the first reference
1381 * to go... deadlock! */
1382 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1383 &tsi->tsi_resid, LDLM_EXTENT, 0);
1385 ldlm_res_lvbo_update(res, NULL, 0);
1386 ldlm_resource_putref(res);
1393 * Destroy OST orphans.
1395 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1396 * set then we must destroy possible orphaned objects.
1398 * \param[in] env execution environment
1399 * \param[in] exp OBD export
1400 * \param[in] ofd OFD device
1401 * \param[in] oa obdo structure for reply
1403 * \retval 0 if successful
1404 * \retval negative value on error
1406 static int ofd_orphans_destroy(const struct lu_env *env,
1407 struct obd_export *exp,
1408 struct ofd_device *ofd, struct obdo *oa)
1410 struct ofd_thread_info *info = ofd_info(env);
1411 struct lu_fid *fid = &info->fti_fid;
1412 struct ost_id *oi = &oa->o_oi;
1413 struct ofd_seq *oseq;
1414 u64 seq = ostid_seq(oi);
1415 u64 end_id = ostid_id(oi);
1423 oseq = ofd_seq_get(ofd, seq);
1425 CERROR("%s: Can not find seq for "DOSTID"\n",
1426 ofd_name(ofd), POSTID(oi));
1431 last = ofd_seq_last_oid(oseq);
1434 LASSERT(exp != NULL);
1435 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1437 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1440 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1441 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1443 while (oid > end_id) {
1444 rc = fid_set_id(fid, oid);
1445 if (unlikely(rc != 0))
1448 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1449 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1450 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1451 /* this is pretty fatal... */
1452 CEMERG("%s: error destroying precreated id "
1454 ofd_name(ofd), PFID(fid), rc);
1458 ofd_seq_last_oid_set(oseq, oid);
1459 /* update last_id on disk periodically so that if we
1460 * restart * we don't need to re-scan all of the just
1461 * deleted objects. */
1462 if ((oid & 511) == 0)
1463 ofd_seq_last_oid_write(env, ofd, oseq);
1467 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1468 ofd_name(ofd), seq, oid);
1472 ofd_seq_last_oid_set(oseq, oid);
1473 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1475 /* don't reuse orphan object, return last used objid */
1476 rc = ostid_set_id(oi, last);
1482 ofd_seq_put(env, oseq);
1487 * OFD request handler for OST_CREATE RPC.
1489 * This is OFD-specific part of request handling. Its main purpose is to
1490 * create new data objects on OST, but it also used to destroy orphans.
1492 * \param[in] tsi target session environment for this request
1494 * \retval 0 if successful
1495 * \retval negative value on error
1497 static int ofd_create_hdl(struct tgt_session_info *tsi)
1499 struct ptlrpc_request *req = tgt_ses_req(tsi);
1500 struct ost_body *repbody;
1501 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1502 struct obdo *rep_oa;
1503 struct obd_export *exp = tsi->tsi_exp;
1504 struct ofd_device *ofd = ofd_exp(exp);
1505 u64 seq = ostid_seq(&oa->o_oi);
1506 u64 oid = ostid_id(&oa->o_oi);
1507 struct ofd_seq *oseq;
1510 ktime_t kstart = ktime_get();
1516 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1519 if (ofd->ofd_no_precreate)
1522 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1523 if (repbody == NULL)
1526 down_read(&ofd->ofd_lastid_rwsem);
1527 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1528 * we may do that in the future.
1529 * Return -ENOSPC until the LAST_ID rebuilt. */
1530 if (unlikely(ofd->ofd_lastid_rebuilding))
1531 GOTO(out_sem, rc = -ENOSPC);
1533 rep_oa = &repbody->oa;
1534 rep_oa->o_oi = oa->o_oi;
1536 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1538 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1540 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1542 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1543 ofd_name(ofd), seq, PTR_ERR(oseq));
1544 GOTO(out_sem, rc = -EINVAL);
1547 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1548 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1549 if (!ofd_obd(ofd)->obd_recovering ||
1550 oid > ofd_seq_last_oid(oseq)) {
1551 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1552 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1553 ofd_seq_last_oid(oseq));
1554 GOTO(out_nolock, rc = -EINVAL);
1556 /* Do nothing here, we re-create objects during recovery
1557 * upon write replay, see ofd_preprw_write() */
1558 GOTO(out_nolock, rc = 0);
1560 /* former ofd_handle_precreate */
1561 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1562 (oa->o_flags & OBD_FL_DELORPHAN)) {
1563 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1565 /* destroy orphans */
1566 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1567 exp->exp_conn_cnt) {
1568 CERROR("%s: dropping old orphan cleanup request\n",
1570 GOTO(out_nolock, rc = 0);
1572 /* This causes inflight precreates to abort and drop lock */
1573 oseq->os_destroys_in_progress = 1;
1574 mutex_lock(&oseq->os_create_lock);
1575 if (!oseq->os_destroys_in_progress) {
1577 "%s:[%llu] destroys_in_progress already cleared\n",
1578 ofd_name(ofd), seq);
1579 rc = ostid_set_id(&rep_oa->o_oi,
1580 ofd_seq_last_oid(oseq));
1583 diff = oid - ofd_seq_last_oid(oseq);
1584 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1585 ofd_seq_last_oid(oseq), diff);
1586 if (-diff > OST_MAX_PRECREATE) {
1587 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1588 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1589 "(%llu), trust the OST\n",
1590 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1591 PFID(&oseq->os_oi.oi_fid),
1592 ofd_seq_last_oid(oseq));
1594 /* Let MDS know that we are so far ahead. */
1595 rc = ostid_set_id(&rep_oa->o_oi,
1596 ofd_seq_last_oid(oseq) + 1);
1597 } else if (diff < 0) {
1598 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1600 oseq->os_destroys_in_progress = 0;
1602 /* XXX: Used by MDS for the first time! */
1603 oseq->os_destroys_in_progress = 0;
1606 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1607 ofd->ofd_lastid_gen)) {
1608 /* Keep the export ref so we can send the reply. */
1609 ofd_obd_disconnect(class_export_get(exp));
1610 GOTO(out_nolock, rc = -ENOTCONN);
1613 mutex_lock(&oseq->os_create_lock);
1614 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1615 exp->exp_conn_cnt) {
1616 CERROR("%s: dropping old precreate request\n",
1620 /* only precreate if seq is 0, IDIF or normal and also o_id
1621 * must be specfied */
1622 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1623 !fid_seq_is_idif(seq)) || oid == 0) {
1624 diff = 1; /* shouldn't we create this right now? */
1626 diff = oid - ofd_seq_last_oid(oseq);
1627 /* Do sync create if the seq is about to used up */
1628 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1629 if (unlikely(oid >= IDIF_MAX_OID - 1))
1631 } else if (fid_seq_is_norm(seq)) {
1633 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1636 CERROR("%s : invalid o_seq "DOSTID"\n",
1637 ofd_name(ofd), POSTID(&oa->o_oi));
1638 GOTO(out, rc = -EINVAL);
1641 if (diff <= -OST_MAX_PRECREATE) {
1643 CERROR("%s: invalid precreate request for "
1644 DOSTID", last_id %llu. "
1645 "Likely MDS last_id corruption\n",
1646 ofd_name(ofd), POSTID(&oa->o_oi),
1647 ofd_seq_last_oid(oseq));
1648 GOTO(out, rc = -EINVAL);
1649 } else if (diff < 0) {
1651 "%s: MDS LAST_ID "DFID" (%llu) is %lld behind OST LAST_ID "DFID" (%llu), trust the OST\n",
1652 ofd_name(ofd), PFID(&oa->o_oi.oi_fid),
1653 oid, -diff, PFID(&oseq->os_oi.oi_fid),
1654 ofd_seq_last_oid(oseq));
1655 /* Let MDS know that we are so far ahead. */
1656 rc = ostid_set_id(&rep_oa->o_oi,
1657 ofd_seq_last_oid(oseq) + 1);
1662 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1668 /* This can happen if a new OST is formatted and installed
1669 * in place of an old one at the same index. Instead of
1670 * precreating potentially millions of deleted old objects
1671 * (possibly filling the OST), only precreate the last batch.
1672 * LFSCK will eventually clean up any orphans. LU-14 */
1673 if (diff > 5 * OST_MAX_PRECREATE) {
1674 /* Message below is checked in conf-sanity test_122b */
1675 LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %lld higher than LAST_ID "DOSTID", only precreating the last %u objects. OST replaced or reformatted?\n",
1676 ofd_name(ofd), POSTID(&oa->o_oi), diff,
1677 POSTID(&oseq->os_oi),
1679 /* From last created */
1680 diff = OST_MAX_PRECREATE;
1681 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1684 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1685 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1686 /* don't enforce grant during orphan recovery */
1687 granted = tgt_grant_create(tsi->tsi_env,
1688 ofd_obd(ofd)->obd_self_export,
1693 CDEBUG(D_HA, "%s: failed to acquire grant "
1694 "space for precreate (%lld): rc = %d\n",
1695 ofd_name(ofd), diff, rc);
1702 next_id = ofd_seq_last_oid(oseq) + 1;
1703 count = ofd_precreate_batch(ofd, (int)diff);
1705 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1706 " at %llu\n", ofd_name(ofd),
1707 count, seq, next_id);
1709 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1710 && ktime_get_seconds() > enough_time) {
1711 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1712 " created at a rate of %d/s\n",
1713 ofd_name(ofd), created, diff + created,
1714 created / DISK_TIMEOUT);
1718 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1719 oseq, count, sync_trans);
1723 } else if (rc < 0) {
1729 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1730 LCONSOLE_WARN("%s: can't create the same count of"
1731 " objects when replaying the request"
1732 " (diff is %lld). see LU-4621\n",
1733 ofd_name(ofd), diff);
1736 /* some objects got created, we can return
1737 * them, even if last creation failed */
1740 CERROR("%s: unable to precreate: rc = %d\n",
1743 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1744 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1745 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1750 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1754 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1755 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1756 if (unlikely(!oseq->os_last_id_synced))
1757 oseq->os_last_id_synced = 1;
1759 mutex_unlock(&oseq->os_create_lock);
1762 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1763 ofd_seq_put(tsi->tsi_env, oseq);
1766 up_read(&ofd->ofd_lastid_rwsem);
1771 * OFD request handler for OST_DESTROY RPC.
1773 * This is OFD-specific part of request handling. It destroys data objects
1774 * related to destroyed object on MDT.
1776 * \param[in] tsi target session environment for this request
1778 * \retval 0 if successful
1779 * \retval negative value on error
1781 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1783 const struct ost_body *body = tsi->tsi_ost_body;
1784 struct ost_body *repbody;
1785 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1786 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1787 struct lu_fid *fid = &fti->fti_fid;
1788 ktime_t kstart = ktime_get();
1795 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1798 /* This is old case for clients before Lustre 2.4 */
1799 /* If there's a DLM request, cancel the locks mentioned in it */
1800 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1802 struct ldlm_request *dlm;
1804 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1807 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1810 *fid = body->oa.o_oi.oi_fid;
1811 oid = ostid_id(&body->oa.o_oi);
1814 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1816 /* check that o_misc makes sense */
1817 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1818 count = body->oa.o_misc;
1820 count = 1; /* default case - single destroy */
1822 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1823 POSTID(&body->oa.o_oi), count);
1828 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1829 if (lrc == -ENOENT) {
1831 "%s: destroying non-existent object "DFID"\n",
1832 ofd_name(ofd), PFID(fid));
1833 /* rewrite rc with -ENOENT only if it is 0 */
1836 } else if (lrc != 0) {
1837 CERROR("%s: error destroying object "DFID": %d\n",
1838 ofd_name(ofd), PFID(fid), lrc);
1844 lrc = fid_set_id(fid, oid);
1845 if (unlikely(lrc != 0 && count > 0))
1846 GOTO(out, rc = lrc);
1849 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1850 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1855 fid_to_ostid(fid, &repbody->oa.o_oi);
1860 * OFD request handler for OST_STATFS RPC.
1862 * This function gets statfs data from storage as part of request
1865 * \param[in] tsi target session environment for this request
1867 * \retval 0 if successful
1868 * \retval negative value on error
1870 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1872 ktime_t kstart = ktime_get();
1873 struct obd_statfs *osfs;
1878 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1880 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1882 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1883 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1885 CERROR("%s: statfs failed: rc = %d\n",
1886 tgt_name(tsi->tsi_tgt), rc);
1888 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1891 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1892 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1898 * OFD request handler for OST_SYNC RPC.
1900 * Sync object data or all filesystem data to the disk and pack the
1903 * \param[in] tsi target session environment for this request
1905 * \retval 0 if successful
1906 * \retval negative value on error
1908 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1910 struct ost_body *body = tsi->tsi_ost_body;
1911 struct ost_body *repbody;
1912 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1913 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1914 struct ofd_object *fo = NULL;
1915 ktime_t kstart = ktime_get();
1920 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1922 /* if no objid is specified, it means "sync whole filesystem" */
1923 if (!fid_is_zero(&tsi->tsi_fid)) {
1924 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1926 RETURN(PTR_ERR(fo));
1929 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1930 fo != NULL ? ofd_object_child(fo) : NULL,
1931 repbody->oa.o_size, repbody->oa.o_blocks);
1935 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1936 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1940 repbody->oa.o_oi = body->oa.o_oi;
1941 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1943 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1945 obdo_from_la(&repbody->oa, &fti->fti_attr,
1948 /* don't return rc from getattr */
1953 ofd_object_put(tsi->tsi_env, fo);
1958 * OFD request handler for OST_FALLOCATE RPC.
1960 * This is part of request processing. Validate request fields,
1961 * preallocate the given OFD object and pack reply.
1963 * \param[in] tsi target session environment for this request
1965 * \retval 0 if successful
1966 * \retval negative value on error
1968 static int ofd_fallocate_hdl(struct tgt_session_info *tsi)
1970 struct obdo *oa = &tsi->tsi_ost_body->oa;
1971 struct ost_body *repbody;
1972 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1973 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1974 struct ldlm_resource *res;
1975 struct ofd_object *fo;
1978 struct lustre_handle lh = { 0, };
1982 ktime_t kstart = ktime_get();
1984 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1985 if (repbody == NULL)
1986 RETURN(err_serious(-ENOMEM));
1989 * fallocate() start and end are passed in o_size and o_blocks
1990 * on the wire. Clients 2.15.0 and newer should always set
1991 * the OBD_MD_FLSIZE and OBD_MD_FLBLOCKS valid flags, but some
1992 * older client versions did not. We permit older clients to
1993 * not set these flags, checking their version by proxy using
1994 * the lack of OBD_CONNECT_TRUNCLOCK to imply 2.14.0 and older.
1996 * Return -EOPNOTSUPP to also work with older clients not
1997 * supporting newer server modes.
1999 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2000 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)
2001 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 21, 53, 0)
2002 && (tgt_conn_flags(tsi) & OBD_CONNECT_OLD_FALLOC)
2005 RETURN(-EOPNOTSUPP);
2009 /* client should already limit len >= 0 */
2013 mode = oa->o_falloc_mode;
2015 * mode == 0 (which is standard prealloc) and PUNCH is supported
2016 * Rest of mode options are not supported yet.
2018 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2019 RETURN(-EOPNOTSUPP);
2021 /* PUNCH_HOLE mode should always be accompanied with KEEP_SIZE flag
2022 * Check that and add the missing flag for such invalid call with
2025 if (mode & FALLOC_FL_PUNCH_HOLE && !(mode & FALLOC_FL_KEEP_SIZE)) {
2026 CWARN("%s: PUNCH mode misses KEEP_SIZE flag, setting it\n",
2027 tsi->tsi_tgt->lut_obd->obd_name);
2028 mode |= FALLOC_FL_KEEP_SIZE;
2031 repbody->oa.o_oi = oa->o_oi;
2032 repbody->oa.o_valid = OBD_MD_FLID;
2034 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2035 oa->o_flags & OBD_FL_SRVLOCK;
2038 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid,
2039 start, end, &lh, LCK_PW, &flags);
2044 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2047 GOTO(out, rc = PTR_ERR(fo));
2049 valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID |
2050 OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME;
2051 la_from_obdo(&info->fti_attr, oa, valid);
2053 rc = ofd_object_fallocate(tsi->tsi_env, fo, start, end, mode,
2054 &info->fti_attr, oa);
2058 rc = ofd_attr_get(tsi->tsi_env, fo, &info->fti_attr);
2060 obdo_from_la(&repbody->oa, &info->fti_attr, OFD_VALID_FLAGS);
2064 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PREALLOC,
2065 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2069 ofd_object_put(tsi->tsi_env, fo);
2072 tgt_data_unlock(&lh, LCK_PW);
2074 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2077 struct ost_lvb *res_lvb;
2079 ldlm_res_lvbo_update(res, NULL, 0);
2080 res_lvb = res->lr_lvb_data;
2082 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2083 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2085 repbody->oa.o_valid |= OBD_MD_FLSIZE;
2086 repbody->oa.o_size = res_lvb->lvb_size;
2088 ldlm_resource_putref(res);
2096 * OFD request handler for OST_PUNCH RPC.
2098 * This is part of request processing. Validate request fields,
2099 * punch (truncate) the given OFD object and pack reply.
2101 * \param[in] tsi target session environment for this request
2103 * \retval 0 if successful
2104 * \retval negative value on error
2106 static int ofd_punch_hdl(struct tgt_session_info *tsi)
2108 const struct obdo *oa = &tsi->tsi_ost_body->oa;
2109 struct ost_body *repbody;
2110 struct ofd_thread_info *info = tsi2ofd_info(tsi);
2111 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
2112 struct ldlm_resource *res;
2113 struct ofd_object *fo;
2115 struct lustre_handle lh = { 0, };
2118 ktime_t kstart = ktime_get();
2123 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
2125 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2126 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2127 RETURN(err_serious(-EPROTO));
2129 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2130 if (repbody == NULL)
2131 RETURN(err_serious(-ENOMEM));
2133 /* punch start,end are passed in o_size,o_blocks throught wire */
2137 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2140 /* standard truncate optimization: if file body is completely
2141 * destroyed, don't send data back to the server. */
2143 flags |= LDLM_FL_AST_DISCARD_DATA;
2145 repbody->oa.o_oi = oa->o_oi;
2146 repbody->oa.o_valid = OBD_MD_FLID;
2148 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2149 oa->o_flags & OBD_FL_SRVLOCK;
2152 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
2153 end, &lh, LCK_PW, &flags);
2158 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2159 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2160 oa->o_valid, start, end);
2162 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2165 GOTO(out, rc = PTR_ERR(fo));
2167 la_from_obdo(&info->fti_attr, oa,
2168 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2169 info->fti_attr.la_size = start;
2170 info->fti_attr.la_valid |= LA_SIZE;
2172 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2177 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2178 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2181 ofd_object_put(tsi->tsi_env, fo);
2184 tgt_data_unlock(&lh, LCK_PW);
2186 /* we do not call this before to avoid lu_object_find() in
2187 * ->lvbo_update() holding another reference on the object.
2188 * otherwise concurrent destroy can make the object unavailable
2189 * for 2nd lu_object_find() waiting for the first reference
2190 * to go... deadlock! */
2191 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2194 struct ost_lvb *res_lvb;
2196 ldlm_res_lvbo_update(res, NULL, 0);
2197 res_lvb = res->lr_lvb_data;
2198 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2199 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2200 ldlm_resource_putref(res);
2206 static int ofd_ladvise_prefetch(const struct lu_env *env,
2207 struct ofd_object *fo,
2208 struct niobuf_local *lnb,
2209 __u64 start, __u64 end, enum dt_bufs_type dbt)
2211 struct ofd_thread_info *info = ofd_info(env);
2212 pgoff_t start_index, end_index, pages;
2213 struct niobuf_remote rnb;
2214 unsigned long nr_local;
2220 ofd_read_lock(env, fo);
2221 if (!ofd_object_exists(fo))
2222 GOTO(out_unlock, rc = -ENOENT);
2224 rc = ofd_attr_get(env, fo, &info->fti_attr);
2226 GOTO(out_unlock, rc);
2228 if (end > info->fti_attr.la_size)
2229 end = info->fti_attr.la_size;
2232 GOTO(out_unlock, rc);
2234 /* We need page aligned offset and length */
2235 start_index = start >> PAGE_SHIFT;
2236 end_index = (end - 1) >> PAGE_SHIFT;
2237 pages = end_index - start_index + 1;
2239 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2240 PTLRPC_MAX_BRW_PAGES;
2241 rnb.rnb_offset = start_index << PAGE_SHIFT;
2242 rnb.rnb_len = nr_local << PAGE_SHIFT;
2243 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb,
2244 PTLRPC_MAX_BRW_PAGES, dbt);
2245 if (unlikely(rc < 0))
2248 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2249 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2252 start_index += nr_local;
2257 ofd_read_unlock(env, fo);
2262 * OFD request handler for OST_LADVISE RPC.
2264 * Tune cache or perfetch policies according to advices.
2266 * \param[in] tsi target session environment for this request
2268 * \retval 0 if successful
2269 * \retval negative errno on error
2271 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2273 struct ptlrpc_request *req = tgt_ses_req(tsi);
2274 struct obd_export *exp = tsi->tsi_exp;
2275 struct ofd_device *ofd = ofd_exp(exp);
2276 struct ost_body *body, *repbody;
2277 struct ofd_thread_info *info;
2278 struct ofd_object *fo;
2279 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2280 const struct lu_env *env = svc_thread->t_env;
2281 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2282 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2283 struct lu_ladvise *ladvise;
2285 struct ladvise_hdr *ladvise_hdr;
2286 struct obd_ioobj ioo;
2287 struct lustre_handle lockh = { 0 };
2290 struct dt_object *dob;
2296 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2297 body = tsi->tsi_ost_body;
2299 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2300 RETURN(err_serious(-EPROTO));
2302 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2303 &RMF_OST_LADVISE_HDR);
2304 if (ladvise_hdr == NULL)
2305 RETURN(err_serious(-EPROTO));
2307 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2308 ladvise_hdr->lah_count < 1)
2309 RETURN(err_serious(-EPROTO));
2311 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2312 RETURN(err_serious(-EPROTO));
2314 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2315 if (ladvise == NULL)
2316 RETURN(err_serious(-EPROTO));
2318 num_advise = req_capsule_get_size(&req->rq_pill,
2319 &RMF_OST_LADVISE, RCL_CLIENT) /
2321 if (num_advise < ladvise_hdr->lah_count)
2322 RETURN(err_serious(-EPROTO));
2324 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2325 repbody->oa = body->oa;
2327 info = ofd_info_init(env, exp);
2329 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2330 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2334 fo = ofd_object_find(env, ofd, &info->fti_fid);
2339 LASSERT(fo != NULL);
2340 dob = ofd_object_child(fo);
2342 if (ptlrpc_connection_is_local(exp->exp_connection))
2343 dbt |= DT_BUFS_TYPE_LOCAL;
2345 for (i = 0; i < num_advise; i++, ladvise++) {
2346 start = ladvise->lla_start;
2347 end = ladvise->lla_end;
2349 rc = err_serious(-EPROTO);
2353 /* Handle different advice types */
2354 switch (ladvise->lla_advice) {
2358 case LU_LADVISE_WILLREAD:
2362 ioo.ioo_oid = body->oa.o_oi;
2364 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2365 &tsi->tsi_resid, start, end - 1,
2366 &lockh, LCK_PR, &flags);
2370 req->rq_status = ofd_ladvise_prefetch(env, fo,
2373 tgt_data_unlock(&lockh, LCK_PR);
2375 case LU_LADVISE_DONTNEED:
2376 rc = dt_ladvise(env, dob, ladvise->lla_start,
2377 ladvise->lla_end, LU_LADVISE_DONTNEED);
2384 ofd_object_put(env, fo);
2385 req->rq_status = rc;
2390 * OFD request handler for OST_QUOTACTL RPC.
2392 * This is part of request processing to validate incoming request fields,
2393 * get the requested data from OSD and pack reply.
2395 * \param[in] tsi target session environment for this request
2397 * \retval 0 if successful
2398 * \retval negative value on error
2400 static int ofd_quotactl(struct tgt_session_info *tsi)
2402 struct obd_quotactl *oqctl, *repoqc;
2403 struct lu_nodemap *nodemap;
2404 ktime_t kstart = ktime_get();
2410 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2412 RETURN(err_serious(-EPROTO));
2414 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2416 RETURN(err_serious(-ENOMEM));
2420 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2421 if (IS_ERR(nodemap))
2422 RETURN(PTR_ERR(nodemap));
2425 if (oqctl->qc_type == USRQUOTA)
2426 id = nodemap_map_id(nodemap, NODEMAP_UID,
2427 NODEMAP_CLIENT_TO_FS,
2429 else if (oqctl->qc_type == GRPQUOTA)
2430 id = nodemap_map_id(nodemap, NODEMAP_GID,
2431 NODEMAP_CLIENT_TO_FS,
2433 else if (oqctl->qc_type == PRJQUOTA)
2434 id = nodemap_map_id(nodemap, NODEMAP_PROJID,
2435 NODEMAP_CLIENT_TO_FS,
2438 nodemap_putref(nodemap);
2440 if (repoqc->qc_id != id)
2441 swap(repoqc->qc_id, id);
2443 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2445 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2446 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2448 if (repoqc->qc_id != id)
2449 swap(repoqc->qc_id, id);
2455 * Prolong lock timeout for the given extent.
2457 * This function finds all locks related with incoming request and
2458 * prolongs their timeout.
2460 * If a client is holding a lock for a long time while it sends
2461 * read or write RPCs to the OST for the object under this lock,
2462 * then we don't want the OST to evict the client. Otherwise,
2463 * if the network or disk is very busy then the client may not
2464 * be able to make any progress to clear out dirty pages under
2465 * the lock and the application will fail.
2467 * Every time a Bulk Read/Write (BRW) request arrives for the object
2468 * covered by the lock, extend the timeout on that lock. The RPC should
2469 * contain a lock handle for the lock it is using, but this
2470 * isn't handled correctly by all client versions, and the
2471 * request may cover multiple locks.
2473 * \param[in] tsi target session environment for this request
2474 * \param[in] data struct of data to prolong locks
2477 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2478 struct ldlm_prolong_args *data)
2480 struct obdo *oa = &tsi->tsi_ost_body->oa;
2481 struct ldlm_lock *lock;
2485 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2486 data->lpa_export = tsi->tsi_exp;
2487 data->lpa_resid = tsi->tsi_resid;
2489 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2490 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2491 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2492 data->lpa_extent.end);
2494 if (oa->o_valid & OBD_MD_FLHANDLE) {
2495 /* mostly a request should be covered by only one lock, try
2497 lock = ldlm_handle2lock(&oa->o_handle);
2499 /* Fast path to check if the lock covers the whole IO
2500 * region exclusively. */
2501 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2502 &data->lpa_extent)) {
2504 LASSERT(lock->l_export == data->lpa_export);
2505 ldlm_lock_prolong_one(lock, data);
2506 LDLM_LOCK_PUT(lock);
2507 if (data->lpa_locks_cnt > 0)
2509 /* The lock was destroyed probably lets try
2512 lock->l_last_used = ktime_get();
2513 LDLM_LOCK_PUT(lock);
2518 ldlm_resource_prolong(data);
2523 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2525 * Determine if \a lock and the lock from request \a req are equivalent
2526 * by comparing their resource names, modes, and extents.
2528 * It is used to give priority to read and write RPCs being done
2529 * under this lock so that the client can drop the contended
2530 * lock more quickly and let other clients use it. This improves
2531 * overall performance in the case where the first client gets a
2532 * very large lock extent that prevents other clients from
2533 * submitting their writes.
2535 * \param[in] req ptlrpc_request being processed
2536 * \param[in] lock contended lock to match
2538 * \retval 1 if lock is matched
2539 * \retval 0 otherwise
2541 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2542 struct ldlm_lock *lock)
2544 struct niobuf_remote *rnb;
2545 struct obd_ioobj *ioo;
2546 enum ldlm_mode mode;
2547 struct ldlm_extent ext;
2548 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2552 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2553 LASSERT(ioo != NULL);
2555 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2556 LASSERT(rnb != NULL);
2558 ext.start = rnb->rnb_offset;
2559 rnb += ioo->ioo_bufcnt - 1;
2560 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2562 LASSERT(lock->l_resource != NULL);
2563 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2566 /* a bulk write can only hold a reference on a PW extent lock
2569 mode = LCK_PW | LCK_GROUP;
2570 if (opc == OST_READ)
2571 /* whereas a bulk read can be protected by either a PR or PW
2575 if (!(lock->l_granted_mode & mode))
2578 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2582 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2584 * Check for whether the given PTLRPC request (\a req) is blocking
2585 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2588 * \param[in] req the incoming request
2590 * \retval 1 if \a req is blocking an LDLM lock cancel
2591 * \retval 0 if it is not
2592 * \retval -ESTALE if lock is not found
2594 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2596 struct tgt_session_info *tsi;
2597 struct obd_ioobj *ioo;
2598 struct niobuf_remote *rnb;
2600 struct ldlm_prolong_args pa = { 0 };
2604 /* Don't use tgt_ses_info() to get session info, because lock_match()
2605 * can be called while request has no processing thread yet. */
2606 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2609 * Use LASSERT below because malformed RPCs should have
2610 * been filtered out in tgt_hpreq_handler().
2612 opc = lustre_msg_get_opc(req->rq_reqmsg);
2613 LASSERT(opc == OST_READ || opc == OST_WRITE);
2615 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2616 LASSERT(ioo != NULL);
2618 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2619 LASSERT(rnb != NULL);
2620 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2622 pa.lpa_mode = LCK_PW | LCK_GROUP;
2623 if (opc == OST_READ)
2624 pa.lpa_mode |= LCK_PR;
2626 pa.lpa_extent.start = rnb->rnb_offset;
2627 rnb += ioo->ioo_bufcnt - 1;
2628 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2630 DEBUG_REQ(D_RPCTRACE, req,
2631 "%s %s: refresh rw locks for "DFID" (%llu->%llu)",
2632 tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid),
2633 pa.lpa_extent.start, pa.lpa_extent.end);
2635 ofd_prolong_extent_locks(tsi, &pa);
2637 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n",
2638 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2640 if (pa.lpa_blocks_cnt > 0)
2643 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2647 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2649 * Called after the request has been handled. It refreshes lock timeout again
2650 * so that client has more time to send lock cancel RPC.
2652 * \param[in] req request which is being processed.
2654 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2656 ofd_rw_hpreq_check(req);
2660 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2662 * This function checks if the given lock is the same by its resname, mode
2663 * and extent as one taken from the request.
2664 * It is used to give priority to punch/truncate RPCs that might lead to
2665 * the fastest release of that lock when a lock is contended.
2667 * \param[in] req ptlrpc_request being processed
2668 * \param[in] lock contended lock to match
2670 * \retval 1 if lock is matched
2671 * \retval 0 otherwise
2673 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2674 struct ldlm_lock *lock)
2676 struct tgt_session_info *tsi;
2678 struct ldlm_extent ext;
2682 /* Don't use tgt_ses_info() to get session info, because lock_match()
2683 * can be called while request has no processing thread yet. */
2684 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2687 * Use LASSERT below because malformed RPCs should have
2688 * been filtered out in tgt_hpreq_handler().
2690 LASSERT(tsi->tsi_ost_body != NULL);
2691 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2692 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2695 oa = &tsi->tsi_ost_body->oa;
2696 ext.start = oa->o_size;
2697 ext.end = oa->o_blocks;
2699 LASSERT(lock->l_resource != NULL);
2700 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2703 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2706 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2710 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2712 * High-priority queue request check for whether the given punch request
2713 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2714 * covered by an LDLM lock.
2718 * \param[in] req the incoming request
2720 * \retval 1 if \a req is blocking an LDLM lock cancel
2721 * \retval 0 if it is not
2722 * \retval -ESTALE if lock is not found
2724 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2726 struct tgt_session_info *tsi;
2728 struct ldlm_prolong_args pa = { 0 };
2732 /* Don't use tgt_ses_info() to get session info, because lock_match()
2733 * can be called while request has no processing thread yet. */
2734 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2735 LASSERT(tsi != NULL);
2736 oa = &tsi->tsi_ost_body->oa;
2738 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2739 oa->o_flags & OBD_FL_SRVLOCK));
2741 pa.lpa_mode = LCK_PW | LCK_GROUP;
2742 pa.lpa_extent.start = oa->o_size;
2743 pa.lpa_extent.end = oa->o_blocks;
2746 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2747 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2748 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2750 ofd_prolong_extent_locks(tsi, &pa);
2752 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2753 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2755 if (pa.lpa_blocks_cnt > 0)
2758 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2762 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2764 * Called after the request has been handled. It refreshes lock timeout again
2765 * so that client has more time to send lock cancel RPC.
2767 * \param[in] req request which is being processed.
2769 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2771 ofd_punch_hpreq_check(req);
2774 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2775 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2776 .hpreq_check = ofd_rw_hpreq_check,
2777 .hpreq_fini = ofd_rw_hpreq_fini
2780 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2781 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2782 .hpreq_check = ofd_punch_hpreq_check,
2783 .hpreq_fini = ofd_punch_hpreq_fini
2787 * Assign high priority operations to an IO request.
2789 * Check if the incoming request is a candidate for
2790 * high-priority processing. If it is, assign it a high
2791 * priority operations table.
2793 * \param[in] tsi target session environment for this request
2795 static void ofd_hp_brw(struct tgt_session_info *tsi)
2797 struct niobuf_remote *rnb;
2798 struct obd_ioobj *ioo;
2802 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2803 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2804 if (ioo->ioo_bufcnt > 0) {
2805 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2806 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2808 /* no high priority if server lock is needed */
2809 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2810 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2814 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2818 * Assign high priority operations to an punch request.
2820 * Check if the incoming request is a candidate for
2821 * high-priority processing. If it is, assign it a high
2822 * priority operations table.
2824 * \param[in] tsi target session environment for this request
2826 static void ofd_hp_punch(struct tgt_session_info *tsi)
2828 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2829 /* no high-priority if server lock is needed */
2830 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2831 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2832 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2833 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2835 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2838 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2839 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2840 #define OST_BRW_READ OST_READ
2841 #define OST_BRW_WRITE OST_WRITE
2844 * Table of OFD-specific request handlers
2846 * This table contains all opcodes accepted by OFD and
2847 * specifies handlers for them. The tgt_request_handler()
2848 * uses such table from each target to process incoming
2851 static struct tgt_handler ofd_tgt_handlers[] = {
2852 TGT_RPC_HANDLER(OST_FIRST_OPC,
2853 0, OST_CONNECT, tgt_connect,
2854 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2855 TGT_RPC_HANDLER(OST_FIRST_OPC,
2856 0, OST_DISCONNECT, tgt_disconnect,
2857 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2858 TGT_RPC_HANDLER(OST_FIRST_OPC,
2859 0, OST_SET_INFO, ofd_set_info_hdl,
2860 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2861 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2862 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl),
2863 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2864 OST_SETATTR, ofd_setattr_hdl),
2865 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2866 OST_CREATE, ofd_create_hdl),
2867 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2868 OST_DESTROY, ofd_destroy_hdl),
2869 TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl),
2870 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
2872 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2873 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write,
2875 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2876 OST_PUNCH, ofd_punch_hdl,
2878 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl),
2879 TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl),
2880 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl),
2881 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE, ofd_fallocate_hdl),
2882 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
2885 static struct tgt_opc_slice ofd_common_slice[] = {
2887 .tos_opc_start = OST_FIRST_OPC,
2888 .tos_opc_end = OST_LAST_OPC,
2889 .tos_hs = ofd_tgt_handlers
2892 .tos_opc_start = OBD_FIRST_OPC,
2893 .tos_opc_end = OBD_LAST_OPC,
2894 .tos_hs = tgt_obd_handlers
2897 .tos_opc_start = LDLM_FIRST_OPC,
2898 .tos_opc_end = LDLM_LAST_OPC,
2899 .tos_hs = tgt_dlm_handlers
2902 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2903 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2904 .tos_hs = tgt_out_handlers
2907 .tos_opc_start = SEQ_FIRST_OPC,
2908 .tos_opc_end = SEQ_LAST_OPC,
2909 .tos_hs = seq_handlers
2912 .tos_opc_start = LFSCK_FIRST_OPC,
2913 .tos_opc_end = LFSCK_LAST_OPC,
2914 .tos_hs = tgt_lfsck_handlers
2917 .tos_opc_start = SEC_FIRST_OPC,
2918 .tos_opc_end = SEC_LAST_OPC,
2919 .tos_hs = tgt_sec_ctx_handlers
2926 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2927 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2930 * Implementation of lu_context_key::lct_key_exit.
2932 * Optional method called on lu_context_exit() for all allocated
2934 * It is used in OFD to sanitize context values which may be re-used
2935 * during another request processing by the same thread.
2937 * \param[in] ctx execution context
2938 * \param[in] key context key
2939 * \param[in] data ofd_thread_info
2941 static void ofd_key_exit(const struct lu_context *ctx,
2942 struct lu_context_key *key, void *data)
2944 struct ofd_thread_info *info = data;
2946 info->fti_env = NULL;
2947 info->fti_exp = NULL;
2950 info->fti_pre_version = 0;
2952 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2955 struct lu_context_key ofd_thread_key = {
2956 .lct_tags = LCT_DT_THREAD,
2957 .lct_init = ofd_key_init,
2958 .lct_fini = ofd_key_fini,
2959 .lct_exit = ofd_key_exit
2963 * Initialize OFD device according to parameters in the config log \a cfg.
2965 * This is the main starting point of OFD initialization. It fills all OFD
2966 * parameters with their initial values and calls other initializing functions
2967 * to set up all OFD subsystems.
2969 * \param[in] env execution environment
2970 * \param[in] m OFD device
2971 * \param[in] ldt LU device type of OFD
2972 * \param[in] cfg configuration log
2974 * \retval 0 if successful
2975 * \retval negative value on error
2977 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2978 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2980 const char *dev = lustre_cfg_string(cfg, 0);
2981 struct ofd_thread_info *info = NULL;
2982 struct obd_device *obd;
2983 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2985 struct nm_config_file *nodemap_config;
2986 struct obd_device_target *obt;
2992 obd = class_name2obd(dev);
2994 CERROR("Cannot find obd with name %s\n", dev);
2998 rc = lu_env_refill((struct lu_env *)env);
3003 obt->obt_magic = OBT_MAGIC;
3005 spin_lock_init(&m->ofd_flags_lock);
3006 m->ofd_raid_degraded = 0;
3007 m->ofd_sync_journal = 0;
3009 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
3011 m->ofd_seq_count = 0;
3012 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
3013 spin_lock_init(&m->ofd_inconsistency_lock);
3015 m->ofd_access_log_mask = -1; /* Log all accesses if enabled. */
3017 spin_lock_init(&m->ofd_batch_lock);
3018 init_rwsem(&m->ofd_lastid_rwsem);
3020 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
3021 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
3022 /* set this lu_device to obd, because error handling need it */
3023 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
3025 /* No connection accepted until configurations will finish */
3026 spin_lock(&obd->obd_dev_lock);
3027 obd->obd_no_conn = 1;
3028 spin_unlock(&obd->obd_dev_lock);
3029 obd->obd_replayable = 1;
3030 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
3031 char *str = lustre_cfg_string(cfg, 4);
3033 if (strchr(str, 'n')) {
3034 CWARN("%s: recovery disabled\n", obd->obd_name);
3035 obd->obd_replayable = 0;
3039 info = ofd_info_init(env, NULL);
3043 rc = ofd_stack_init(env, m, cfg, &lmd_flags);
3045 CERROR("%s: can't init device stack, rc %d\n",
3050 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
3051 ofd_procfs_add_brw_stats_symlink(m);
3054 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
3055 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
3056 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
3057 LDLM_NAMESPACE_SERVER,
3058 LDLM_NAMESPACE_GREEDY,
3060 if (IS_ERR(m->ofd_namespace)) {
3061 rc = PTR_ERR(m->ofd_namespace);
3062 CERROR("%s: unable to create server namespace: rc = %d\n",
3064 m->ofd_namespace = NULL;
3065 GOTO(err_fini_stack, rc);
3067 /* set obd_namespace for compatibility with old code */
3068 obd->obd_namespace = m->ofd_namespace;
3069 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
3070 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
3071 m->ofd_namespace->ns_lvbp = m;
3073 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3074 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
3076 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
3077 OBD_FAIL_OST_ALL_REQUEST_NET,
3078 OBD_FAIL_OST_ALL_REPLY_NET);
3080 GOTO(err_free_ns, rc);
3082 if (lmd_flags & LMD_FLG_SKIP_LFSCK)
3083 m->ofd_skip_lfsck = 1;
3084 if (lmd_flags & LMD_FLG_LOCAL_RECOV)
3085 m->ofd_lut.lut_local_recovery = 1;
3087 rc = ofd_tunables_init(m);
3089 GOTO(err_fini_lut, rc);
3091 tgd->tgd_reserved_pcnt = 0;
3093 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
3094 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
3095 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
3096 OFD_PRECREATE_SMALL_FS)
3097 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
3098 m->ofd_atime_diff = OFD_DEF_ATIME_DIFF;
3100 rc = ofd_fs_setup(env, m, obd);
3102 GOTO(err_fini_proc, rc);
3104 fid.f_seq = FID_SEQ_LOCAL_NAME;
3107 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
3110 GOTO(err_fini_fs, rc);
3112 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
3114 if (IS_ERR(nodemap_config)) {
3115 rc = PTR_ERR(nodemap_config);
3117 GOTO(err_fini_los, rc);
3119 obt->obt_nodemap_config_file = nodemap_config;
3122 rc = ofd_start_inconsistency_verification_thread(m);
3124 GOTO(err_fini_nm, rc);
3126 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
3131 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
3132 obt->obt_nodemap_config_file = NULL;
3134 local_oid_storage_fini(env, m->ofd_los);
3137 ofd_fs_cleanup(env, m);
3141 tgt_fini(env, &m->ofd_lut);
3143 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3144 obd->obd_namespace = m->ofd_namespace = NULL;
3146 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3151 * Stop the OFD device
3153 * This function stops the OFD device and all its subsystems.
3154 * This is the end of OFD lifecycle.
3156 * \param[in] env execution environment
3157 * \param[in] m OFD device
3159 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3161 struct obd_device *obd = ofd_obd(m);
3162 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3163 struct lfsck_stop stop;
3165 stop.ls_status = LS_PAUSED;
3167 lfsck_stop(env, m->ofd_osd, &stop);
3168 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3169 target_recovery_fini(obd);
3170 if (m->ofd_namespace != NULL)
3171 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3172 d->ld_obd->obd_force);
3174 obd_exports_barrier(obd);
3175 obd_zombie_barrier();
3178 tgt_fini(env, &m->ofd_lut);
3179 ofd_stop_inconsistency_verification_thread(m);
3180 lfsck_degister(env, m->ofd_osd);
3181 ofd_fs_cleanup(env, m);
3182 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3183 obd->u.obt.obt_nodemap_config_file = NULL;
3185 if (m->ofd_namespace != NULL) {
3186 ldlm_namespace_free_post(m->ofd_namespace);
3187 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3190 ofd_access_log_delete(m->ofd_access_log);
3191 m->ofd_access_log = NULL;
3193 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3195 LASSERT(atomic_read(&d->ld_ref) == 0);
3196 server_put_mount(obd->obd_name, true);
3201 * Implementation of lu_device_type_operations::ldto_device_fini.
3203 * Finalize device. Dual to ofd_device_init(). It is called from
3204 * obd_precleanup() and stops the current device.
3206 * \param[in] env execution environment
3207 * \param[in] d LU device of OFD
3211 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3212 struct lu_device *d)
3215 ofd_fini(env, ofd_dev(d));
3220 * Implementation of lu_device_type_operations::ldto_device_free.
3222 * Free OFD device. Dual to ofd_device_alloc().
3224 * \param[in] env execution environment
3225 * \param[in] d LU device of OFD
3229 static struct lu_device *ofd_device_free(const struct lu_env *env,
3230 struct lu_device *d)
3232 struct ofd_device *m = ofd_dev(d);
3234 dt_device_fini(&m->ofd_dt_dev);
3240 * Implementation of lu_device_type_operations::ldto_device_alloc.
3242 * This function allocates the new OFD device. It is called from
3243 * obd_setup() if OBD device had lu_device_type defined.
3245 * \param[in] env execution environment
3246 * \param[in] t lu_device_type of OFD device
3247 * \param[in] cfg configuration log
3249 * \retval pointer to the lu_device of just allocated OFD
3250 * \retval ERR_PTR of return value on error
3252 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3253 struct lu_device_type *t,
3254 struct lustre_cfg *cfg)
3256 struct ofd_device *m;
3257 struct lu_device *l;
3262 return ERR_PTR(-ENOMEM);
3264 l = &m->ofd_dt_dev.dd_lu_dev;
3265 dt_device_init(&m->ofd_dt_dev, t);
3266 rc = ofd_init0(env, m, t, cfg);
3268 ofd_device_free(env, l);
3275 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3276 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3278 static const struct lu_device_type_operations ofd_device_type_ops = {
3279 .ldto_init = ofd_type_init,
3280 .ldto_fini = ofd_type_fini,
3282 .ldto_start = ofd_type_start,
3283 .ldto_stop = ofd_type_stop,
3285 .ldto_device_alloc = ofd_device_alloc,
3286 .ldto_device_free = ofd_device_free,
3287 .ldto_device_fini = ofd_device_fini
3290 static struct lu_device_type ofd_device_type = {
3291 .ldt_tags = LU_DEVICE_DT,
3292 .ldt_name = LUSTRE_OST_NAME,
3293 .ldt_ops = &ofd_device_type_ops,
3294 .ldt_ctx_tags = LCT_DT_THREAD
3298 * Initialize OFD module.
3300 * This function is called upon module loading. It registers OFD device type
3301 * and prepares all in-memory structures used by all OFD devices.
3303 * \retval 0 if successful
3304 * \retval negative value on error
3306 static int __init ofd_init(void)
3310 rc = lu_kmem_init(ofd_caches);
3314 rc = ofd_access_log_module_init();
3318 rc = class_register_type(&ofd_obd_ops, NULL, true,
3319 LUSTRE_OST_NAME, &ofd_device_type);
3321 goto out_ofd_access_log;
3326 ofd_access_log_module_exit();
3328 lu_kmem_fini(ofd_caches);
3336 * This function is called upon OFD module unloading.
3337 * It frees all related structures and unregisters OFD device type.
3339 static void __exit ofd_exit(void)
3341 class_unregister_type(LUSTRE_OST_NAME);
3342 ofd_access_log_module_exit();
3343 lu_kmem_fini(ofd_caches);
3346 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3347 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3348 MODULE_VERSION(LUSTRE_VERSION_STRING);
3349 MODULE_LICENSE("GPL");
3351 module_init(ofd_init);
3352 module_exit(ofd_exit);