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;
547 range_lock_tree_init(&of->ofo_write_tree);
555 * Return the result of LFSCK run to the OFD.
557 * Notify OFD about result of LFSCK run. That may block the new object
558 * creation until problem is fixed by LFSCK.
560 * \param[in] env execution environment
561 * \param[in] data pointer to the OFD device
562 * \param[in] event LFSCK event type
564 * \retval 0 if successful
565 * \retval negative value on unknown event
567 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
568 enum lfsck_events event)
570 struct ofd_device *ofd = data;
571 struct obd_device *obd = ofd_obd(ofd);
574 case LE_LASTID_REBUILDING:
575 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
576 "on the device until the LAST_ID rebuilt successfully.\n",
578 down_write(&ofd->ofd_lastid_rwsem);
579 ofd->ofd_lastid_rebuilding = 1;
580 up_write(&ofd->ofd_lastid_rwsem);
582 case LE_LASTID_REBUILT: {
583 down_write(&ofd->ofd_lastid_rwsem);
584 ofd_seqs_free(env, ofd);
585 ofd->ofd_lastid_rebuilding = 0;
586 ofd->ofd_lastid_gen++;
587 up_write(&ofd->ofd_lastid_rwsem);
588 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
593 CERROR("%s: unknown lfsck event: rc = %d\n",
594 ofd_name(ofd), event);
602 * Implementation of lu_device_operations::ldo_prepare.
604 * This method is called after layer has been initialized and before it starts
605 * serving user requests. In OFD it starts lfsk check routines and initializes
608 * \param[in] env execution environment
609 * \param[in] pdev higher device in stack, NULL for OFD
610 * \param[in] dev lu_device of OFD device
612 * \retval 0 if successful
613 * \retval negative value on error
615 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
616 struct lu_device *dev)
618 struct ofd_thread_info *info;
619 struct ofd_device *ofd = ofd_dev(dev);
620 struct obd_device *obd = ofd_obd(ofd);
621 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
626 info = ofd_info_init(env, NULL);
630 /* initialize lower device */
631 rc = next->ld_ops->ldo_prepare(env, dev, next);
635 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
636 ofd_lfsck_out_notify, ofd, false);
638 CERROR("%s: failed to initialize lfsck: rc = %d\n",
643 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
644 /* The LFSCK instance is registered just now, so it must be there when
645 * register the namespace to such instance. */
646 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
648 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
649 OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_OST_PREPARE_DELAY, OBD_FAIL_ONCE,
650 (OBD_TIMEOUT_DEFAULT + 1) / 4);
651 LASSERT(obd->obd_no_conn);
652 spin_lock(&obd->obd_dev_lock);
653 obd->obd_no_conn = 0;
654 spin_unlock(&obd->obd_dev_lock);
656 if (obd->obd_recovering == 0)
657 ofd_postrecov(env, ofd);
663 * Implementation of lu_device_operations::ldo_recovery_complete.
665 * This method notifies all layers about 'recovery complete' event. That means
666 * device is in full state and consistent. An OFD calculates available grant
667 * space upon this event.
669 * \param[in] env execution environment
670 * \param[in] dev lu_device of OFD device
672 * \retval 0 if successful
673 * \retval negative value on error
675 static int ofd_recovery_complete(const struct lu_env *env,
676 struct lu_device *dev)
678 struct ofd_thread_info *oti = ofd_info(env);
679 struct ofd_device *ofd = ofd_dev(dev);
680 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
686 * Grant space for object precreation on the self export.
687 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
688 * is enough to create 10k objects. More space is then acquired for
689 * precreation in tgt_grant_create().
691 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
692 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
693 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
694 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
695 OBD_CONNECT_GRANT_PARAM;
696 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
698 rc = next->ld_ops->ldo_recovery_complete(env, next);
703 * lu_device_operations matrix for OFD device.
705 static const struct lu_device_operations ofd_lu_ops = {
706 .ldo_object_alloc = ofd_object_alloc,
707 .ldo_process_config = ofd_process_config,
708 .ldo_recovery_complete = ofd_recovery_complete,
709 .ldo_prepare = ofd_prepare,
712 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
714 * Expose OSD statistics to OFD layer.
716 * The osd interfaces to the backend file system exposes useful data
717 * such as brw_stats and read or write cache states. This same data
718 * needs to be exposed into the obdfilter (ofd) layer to maintain
719 * backwards compatibility. This function creates the symlinks in the
720 * proc layer to enable this.
722 * \param[in] ofd OFD device
724 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
726 struct obd_device *obd = ofd_obd(ofd);
727 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
728 struct kobj_type *osd_type;
731 osd_type = get_ktype(&ofd->ofd_osd->dd_kobj);
732 for (i = 0; osd_type->default_attrs[i]; i++) {
733 if (strcmp(osd_type->default_attrs[i]->name,
734 "read_cache_enable") == 0) {
735 ofd->ofd_read_cache_enable =
736 osd_type->default_attrs[i];
739 if (strcmp(osd_type->default_attrs[i]->name,
740 "readcache_max_filesize") == 0) {
741 ofd->ofd_read_cache_max_filesize =
742 osd_type->default_attrs[i];
745 if (strcmp(osd_type->default_attrs[i]->name,
746 "writethrough_cache_enable") == 0) {
747 ofd->ofd_write_cache_enable =
748 osd_type->default_attrs[i];
752 if (obd->obd_proc_entry == NULL)
755 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
756 "../../%s/%s/brw_stats",
757 osd_obd->obd_type->typ_name, obd->obd_name);
762 * Cleanup all procfs entries in OFD.
764 * \param[in] ofd OFD device
766 static void ofd_procfs_fini(struct ofd_device *ofd)
768 struct obd_device *obd = ofd_obd(ofd);
770 tgt_tunables_fini(&ofd->ofd_lut);
771 lprocfs_free_per_client_stats(obd);
772 lprocfs_obd_cleanup(obd);
773 lprocfs_free_obd_stats(obd);
774 lprocfs_job_stats_fini(obd);
778 * Stop SEQ/FID server on OFD.
780 * \param[in] env execution environment
781 * \param[in] ofd OFD device
783 * \retval 0 if successful
784 * \retval negative value on error
786 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
788 return seq_site_fini(env, &ofd->ofd_seq_site);
792 * Start SEQ/FID server on OFD.
794 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
795 * It also connects to the master server to get own FID sequence (SEQ) range
796 * to this particular OFD. Typically that happens when the OST is first
797 * formatted or in the rare case that it exhausts the local sequence range.
799 * The sequence range is allocated out to the MDTs for OST object allocations,
800 * and not directly to the clients.
802 * \param[in] env execution environment
803 * \param[in] ofd OFD device
805 * \retval 0 if successful
806 * \retval negative value on error
808 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
810 struct seq_server_site *ss = &ofd->ofd_seq_site;
811 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
812 char *obd_name = ofd_name(ofd);
814 int len = strlen(obd_name) + 7;
817 ss = &ofd->ofd_seq_site;
818 lu->ld_site->ld_seq_site = ss;
819 ss->ss_lu = lu->ld_site;
820 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
822 OBD_ALLOC(name, len);
826 OBD_ALLOC_PTR(ss->ss_server_seq);
827 if (ss->ss_server_seq == NULL)
828 GOTO(out_name, rc = -ENOMEM);
830 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
831 LUSTRE_SEQ_SERVER, ss);
833 CERROR("%s: seq server init error: rc = %d\n", obd_name, rc);
834 GOTO(out_server, rc);
836 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
838 OBD_ALLOC_PTR(ss->ss_client_seq);
839 if (ss->ss_client_seq == NULL)
840 GOTO(out_server, rc = -ENOMEM);
842 snprintf(name, len, "%s-super", obd_name);
843 seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
846 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
849 seq_client_fini(ss->ss_client_seq);
850 OBD_FREE_PTR(ss->ss_client_seq);
851 ss->ss_client_seq = NULL;
853 seq_server_fini(ss->ss_server_seq, env);
854 OBD_FREE_PTR(ss->ss_server_seq);
855 ss->ss_server_seq = NULL;
864 * OFD request handler for OST_SET_INFO RPC.
866 * This is OFD-specific part of request handling
868 * \param[in] tsi target session environment for this request
870 * \retval 0 if successful
871 * \retval negative value on error
873 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
875 struct ptlrpc_request *req = tgt_ses_req(tsi);
876 struct ost_body *body = NULL, *repbody;
877 void *key, *val = NULL;
878 int keylen, vallen, rc = 0;
879 bool is_grant_shrink;
880 ktime_t kstart = ktime_get();
884 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
886 DEBUG_REQ(D_HA, req, "no set_info key");
887 RETURN(err_serious(-EFAULT));
889 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
892 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
894 DEBUG_REQ(D_HA, req, "no set_info val");
895 RETURN(err_serious(-EFAULT));
897 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
900 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
902 /* In this case the value is actually an RMF_OST_BODY, so we
903 * transmutate the type of this PTLRPC */
904 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
906 rc = req_capsule_server_pack(tsi->tsi_pill);
910 if (is_grant_shrink) {
911 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
914 * Because we already sync grant info with client when
915 * reconnect, grant info will be cleared for resent
916 * req, otherwise, outdated grant count in the rpc
917 * would de-sync grant counters
919 if (lustre_msg_get_flags(req->rq_reqmsg) &
920 (MSG_RESENT | MSG_REPLAY)) {
921 DEBUG_REQ(D_CACHE, req,
922 "clear resent/replay req grant info");
923 body->oa.o_valid &= ~OBD_MD_FLGRANT;
926 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
929 /** handle grant shrink, similar to a read request */
930 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
932 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
934 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
937 CERROR("%s: Unsupported key %s\n",
938 tgt_name(tsi->tsi_tgt), (char *)key);
941 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
942 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
948 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
950 * This function returns a list of extents which describes how a file's
951 * blocks are laid out on the disk.
953 * \param[in] env execution environment
954 * \param[in] ofd OFD device
955 * \param[in] fid FID of object
956 * \param[in] fiemap fiemap structure to fill with data
958 * \retval 0 if \a fiemap is filled with data successfully
959 * \retval negative value on error
961 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
962 struct lu_fid *fid, struct fiemap *fiemap)
964 struct ofd_object *fo;
967 fo = ofd_object_find(env, ofd, fid);
969 CERROR("%s: error finding object "DFID"\n",
970 ofd_name(ofd), PFID(fid));
974 ofd_read_lock(env, fo);
975 if (ofd_object_exists(fo))
976 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
979 ofd_read_unlock(env, fo);
980 ofd_object_put(env, fo);
985 static int ofd_lock_unlock_region(const struct lu_env *env,
986 struct ldlm_namespace *ns,
987 struct ldlm_res_id *res_id,
988 unsigned long long begin,
989 unsigned long long end)
993 struct lustre_handle lh = { 0 };
995 LASSERT(begin <= end);
997 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
1001 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
1002 tgt_data_unlock(&lh, LCK_PR);
1008 * Lock the sparse areas of given resource.
1010 * The locking of sparse areas will cause dirty data to be flushed back from
1011 * clients. This is used when getting the FIEMAP of an object to make sure
1012 * there is no unaccounted cached data on clients.
1014 * This function goes through \a fiemap list of extents and locks only sparse
1015 * areas between extents.
1017 * \param[in] ns LDLM namespace
1018 * \param[in] res_id resource ID
1019 * \param[in] fiemap file extents mapping on disk
1020 * \param[in] locked list head of regions list
1022 * \retval 0 if successful
1023 * \retval negative value on error
1025 static int lock_zero_regions(const struct lu_env *env,
1026 struct ldlm_namespace *ns,
1027 struct ldlm_res_id *res_id,
1028 struct fiemap *fiemap)
1030 __u64 begin = fiemap->fm_start;
1033 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1037 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1038 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1039 if (fiemap_start[i].fe_logical > begin) {
1040 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1041 begin, fiemap_start[i].fe_logical);
1042 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1043 fiemap_start[i].fe_logical);
1048 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1051 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1052 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1053 begin, fiemap->fm_start + fiemap->fm_length);
1054 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1055 fiemap->fm_start + fiemap->fm_length);
1063 * OFD request handler for OST_GET_INFO RPC.
1065 * This is OFD-specific part of request handling. The OFD-specific keys are:
1066 * - KEY_LAST_ID (obsolete)
1070 * This function reads needed data from storage and fills reply with it.
1072 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1073 * and is kept for compatibility.
1075 * \param[in] tsi target session environment for this request
1077 * \retval 0 if successful
1078 * \retval negative value on error
1080 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1082 struct obd_export *exp = tsi->tsi_exp;
1083 struct ofd_device *ofd = ofd_exp(exp);
1084 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1087 int replylen, rc = 0;
1088 ktime_t kstart = ktime_get();
1092 /* this common part for get_info rpc */
1093 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1095 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1096 RETURN(err_serious(-EPROTO));
1098 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1101 if (KEY_IS(KEY_LAST_ID)) {
1103 struct ofd_seq *oseq;
1105 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1106 rc = req_capsule_server_pack(tsi->tsi_pill);
1108 RETURN(err_serious(rc));
1110 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1112 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1113 (u64)exp->exp_filter_data.fed_group);
1117 *last_id = ofd_seq_last_oid(oseq);
1118 ofd_seq_put(tsi->tsi_env, oseq);
1119 } else if (KEY_IS(KEY_FIEMAP)) {
1120 struct ll_fiemap_info_key *fm_key;
1121 struct fiemap *fiemap;
1124 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1126 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1127 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1129 RETURN(err_serious(rc));
1131 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1133 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1135 replylen = fiemap_count_to_size(
1136 fm_key->lfik_fiemap.fm_extent_count);
1137 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1138 RCL_SERVER, replylen);
1140 rc = req_capsule_server_pack(tsi->tsi_pill);
1142 RETURN(err_serious(rc));
1144 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1148 *fiemap = fm_key->lfik_fiemap;
1149 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1151 /* LU-3219: Lock the sparse areas to make sure dirty
1152 * flushed back from client, then call fiemap again. */
1153 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1154 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1155 ost_fid_build_resid(fid, &fti->fti_resid);
1156 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1157 &fti->fti_resid, fiemap);
1159 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1162 } else if (KEY_IS(KEY_LAST_FID)) {
1163 struct ofd_device *ofd = ofd_exp(exp);
1164 struct ofd_seq *oseq;
1168 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1169 rc = req_capsule_server_pack(tsi->tsi_pill);
1171 RETURN(err_serious(rc));
1173 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GET_LAST_FID))
1176 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1178 RETURN(err_serious(-EPROTO));
1180 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1182 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1186 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1187 ostid_seq(&fti->fti_ostid));
1191 rc = ostid_to_fid(fid, &oseq->os_oi,
1192 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1194 GOTO(out_put, rc = -EFAULT);
1196 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1199 ofd_seq_put(tsi->tsi_env, oseq);
1201 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1205 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1206 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1212 * OFD request handler for OST_GETATTR RPC.
1214 * This is OFD-specific part of request handling. It finds the OFD object
1215 * by its FID, gets attributes from storage and packs result to the reply.
1217 * \param[in] tsi target session environment for this request
1219 * \retval 0 if successful
1220 * \retval negative value on error
1222 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1224 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1225 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1226 struct ost_body *repbody;
1227 struct lustre_handle lh = { 0 };
1228 struct ofd_object *fo;
1230 enum ldlm_mode lock_mode = LCK_PR;
1231 ktime_t kstart = ktime_get();
1236 LASSERT(tsi->tsi_ost_body != NULL);
1238 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1239 if (repbody == NULL)
1242 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1243 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1245 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1246 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1249 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1252 rc = tgt_extent_lock(tsi->tsi_env,
1253 tsi->tsi_tgt->lut_obd->obd_namespace,
1254 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1260 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1262 GOTO(out, rc = PTR_ERR(fo));
1264 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1268 obdo_from_la(&repbody->oa, &fti->fti_attr,
1269 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1271 /* Store object version in reply */
1272 curr_version = dt_version_get(tsi->tsi_env,
1273 ofd_object_child(fo));
1274 if ((__s64)curr_version != -EOPNOTSUPP) {
1275 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1276 repbody->oa.o_data_version = curr_version;
1279 if (fo->ofo_ff.ff_layout_version > 0) {
1280 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1281 repbody->oa.o_layout_version =
1282 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1284 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1285 PFID(&tsi->tsi_fid),
1286 repbody->oa.o_layout_version);
1290 ofd_object_put(tsi->tsi_env, fo);
1293 tgt_data_unlock(&lh, lock_mode);
1295 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1296 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1298 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1299 repbody->oa.o_flags = OBD_FL_FLUSH;
1305 * OFD request handler for OST_SETATTR RPC.
1307 * This is OFD-specific part of request handling. It finds the OFD object
1308 * by its FID, sets attributes from request and packs result to the reply.
1310 * \param[in] tsi target session environment for this request
1312 * \retval 0 if successful
1313 * \retval negative value on error
1315 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1317 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1318 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1319 struct ost_body *body = tsi->tsi_ost_body;
1320 struct ost_body *repbody;
1321 struct ldlm_resource *res;
1322 struct ofd_object *fo;
1323 ktime_t kstart = ktime_get();
1328 LASSERT(body != NULL);
1330 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1331 if (repbody == NULL)
1334 repbody->oa.o_oi = body->oa.o_oi;
1335 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1337 /* This would be very bad - accidentally truncating a file when
1338 * changing the time or similar - bug 12203. */
1339 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1340 body->oa.o_size != OBD_OBJECT_EOF) {
1341 static char mdsinum[48];
1343 if (body->oa.o_valid & OBD_MD_FLFID)
1344 snprintf(mdsinum, sizeof(mdsinum) - 1,
1345 "of parent "DFID, body->oa.o_parent_seq,
1346 body->oa.o_parent_oid, 0);
1350 CERROR("%s: setattr from %s is trying to truncate object "DFID
1351 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1352 PFID(&tsi->tsi_fid), mdsinum);
1356 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1358 GOTO(out, rc = PTR_ERR(fo));
1360 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1361 fti->fti_attr.la_valid &= ~LA_TYPE;
1363 /* setting objects attributes (including owner/group) */
1364 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1368 obdo_from_la(&repbody->oa, &fti->fti_attr,
1369 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1371 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1372 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1375 ofd_object_put(tsi->tsi_env, fo);
1378 /* we do not call this before to avoid lu_object_find() in
1379 * ->lvbo_update() holding another reference on the object.
1380 * otherwise concurrent destroy can make the object unavailable
1381 * for 2nd lu_object_find() waiting for the first reference
1382 * to go... deadlock! */
1383 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1384 &tsi->tsi_resid, LDLM_EXTENT, 0);
1386 ldlm_res_lvbo_update(res, NULL, 0);
1387 ldlm_resource_putref(res);
1394 * Destroy OST orphans.
1396 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1397 * set then we must destroy possible orphaned objects.
1399 * \param[in] env execution environment
1400 * \param[in] exp OBD export
1401 * \param[in] ofd OFD device
1402 * \param[in] oa obdo structure for reply
1404 * \retval 0 if successful
1405 * \retval negative value on error
1407 static int ofd_orphans_destroy(const struct lu_env *env,
1408 struct obd_export *exp,
1409 struct ofd_device *ofd, struct obdo *oa)
1411 struct ofd_thread_info *info = ofd_info(env);
1412 struct lu_fid *fid = &info->fti_fid;
1413 struct ost_id *oi = &oa->o_oi;
1414 struct ofd_seq *oseq;
1415 u64 seq = ostid_seq(oi);
1416 u64 end_id = ostid_id(oi);
1424 oseq = ofd_seq_get(ofd, seq);
1426 CERROR("%s: Can not find seq for "DOSTID"\n",
1427 ofd_name(ofd), POSTID(oi));
1432 last = ofd_seq_last_oid(oseq);
1435 LASSERT(exp != NULL);
1436 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1438 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1441 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1442 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1444 while (oid > end_id) {
1445 rc = fid_set_id(fid, oid);
1446 if (unlikely(rc != 0))
1449 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1450 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1451 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1452 /* this is pretty fatal... */
1453 CEMERG("%s: error destroying precreated id "
1455 ofd_name(ofd), PFID(fid), rc);
1459 ofd_seq_last_oid_set(oseq, oid);
1460 /* update last_id on disk periodically so that if we
1461 * restart * we don't need to re-scan all of the just
1462 * deleted objects. */
1463 if ((oid & 511) == 0)
1464 ofd_seq_last_oid_write(env, ofd, oseq);
1468 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1469 ofd_name(ofd), seq, oid);
1473 ofd_seq_last_oid_set(oseq, oid);
1474 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1476 /* don't reuse orphan object, return last used objid */
1477 rc = ostid_set_id(oi, last);
1483 ofd_seq_put(env, oseq);
1488 * OFD request handler for OST_CREATE RPC.
1490 * This is OFD-specific part of request handling. Its main purpose is to
1491 * create new data objects on OST, but it also used to destroy orphans.
1493 * \param[in] tsi target session environment for this request
1495 * \retval 0 if successful
1496 * \retval negative value on error
1498 static int ofd_create_hdl(struct tgt_session_info *tsi)
1500 struct ptlrpc_request *req = tgt_ses_req(tsi);
1501 struct ost_body *repbody;
1502 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1503 struct obdo *rep_oa;
1504 struct obd_export *exp = tsi->tsi_exp;
1505 struct ofd_device *ofd = ofd_exp(exp);
1506 u64 seq = ostid_seq(&oa->o_oi);
1507 u64 oid = ostid_id(&oa->o_oi);
1508 struct ofd_seq *oseq;
1511 ktime_t kstart = ktime_get();
1517 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1520 if (ofd->ofd_no_precreate)
1523 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1524 if (repbody == NULL)
1527 down_read(&ofd->ofd_lastid_rwsem);
1528 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1529 * we may do that in the future.
1530 * Return -ENOSPC until the LAST_ID rebuilt. */
1531 if (unlikely(ofd->ofd_lastid_rebuilding))
1532 GOTO(out_sem, rc = -ENOSPC);
1534 rep_oa = &repbody->oa;
1535 rep_oa->o_oi = oa->o_oi;
1537 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1539 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1541 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1543 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1544 ofd_name(ofd), seq, PTR_ERR(oseq));
1545 GOTO(out_sem, rc = -EINVAL);
1548 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1549 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1550 if (!ofd_obd(ofd)->obd_recovering ||
1551 oid > ofd_seq_last_oid(oseq)) {
1552 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1553 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1554 ofd_seq_last_oid(oseq));
1555 GOTO(out_nolock, rc = -EINVAL);
1557 /* Do nothing here, we re-create objects during recovery
1558 * upon write replay, see ofd_preprw_write() */
1559 GOTO(out_nolock, rc = 0);
1561 /* former ofd_handle_precreate */
1562 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1563 (oa->o_flags & OBD_FL_DELORPHAN)) {
1564 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1566 /* destroy orphans */
1567 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1568 exp->exp_conn_cnt) {
1569 CERROR("%s: dropping old orphan cleanup request\n",
1571 GOTO(out_nolock, rc = 0);
1573 /* This causes inflight precreates to abort and drop lock */
1574 oseq->os_destroys_in_progress = 1;
1575 mutex_lock(&oseq->os_create_lock);
1576 if (!oseq->os_destroys_in_progress) {
1578 "%s:[%llu] destroys_in_progress already cleared\n",
1579 ofd_name(ofd), seq);
1580 rc = ostid_set_id(&rep_oa->o_oi,
1581 ofd_seq_last_oid(oseq));
1584 diff = oid - ofd_seq_last_oid(oseq);
1585 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1586 ofd_seq_last_oid(oseq), diff);
1587 if (-diff > OST_MAX_PRECREATE) {
1588 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1589 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1590 "(%llu), trust the OST\n",
1591 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1592 PFID(&oseq->os_oi.oi_fid),
1593 ofd_seq_last_oid(oseq));
1595 /* Let MDS know that we are so far ahead. */
1596 rc = ostid_set_id(&rep_oa->o_oi,
1597 ofd_seq_last_oid(oseq) + 1);
1598 } else if (diff < 0) {
1599 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1601 oseq->os_destroys_in_progress = 0;
1603 /* XXX: Used by MDS for the first time! */
1604 oseq->os_destroys_in_progress = 0;
1607 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1608 ofd->ofd_lastid_gen)) {
1609 /* Keep the export ref so we can send the reply. */
1610 ofd_obd_disconnect(class_export_get(exp));
1611 GOTO(out_nolock, rc = -ENOTCONN);
1614 mutex_lock(&oseq->os_create_lock);
1615 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1616 exp->exp_conn_cnt) {
1617 CERROR("%s: dropping old precreate request\n",
1621 /* only precreate if seq is 0, IDIF or normal and also o_id
1622 * must be specfied */
1623 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1624 !fid_seq_is_idif(seq)) || oid == 0) {
1625 diff = 1; /* shouldn't we create this right now? */
1627 diff = oid - ofd_seq_last_oid(oseq);
1628 /* Do sync create if the seq is about to used up */
1629 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1630 if (unlikely(oid >= IDIF_MAX_OID - 1))
1632 } else if (fid_seq_is_norm(seq)) {
1634 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1637 CERROR("%s : invalid o_seq "DOSTID"\n",
1638 ofd_name(ofd), POSTID(&oa->o_oi));
1639 GOTO(out, rc = -EINVAL);
1642 if (diff <= -OST_MAX_PRECREATE) {
1644 CERROR("%s: invalid precreate request for "
1645 DOSTID", last_id %llu. "
1646 "Likely MDS last_id corruption\n",
1647 ofd_name(ofd), POSTID(&oa->o_oi),
1648 ofd_seq_last_oid(oseq));
1649 GOTO(out, rc = -EINVAL);
1650 } else if (diff < 0) {
1652 "%s: MDS LAST_ID "DFID" (%llu) is %lld behind OST LAST_ID "DFID" (%llu), trust the OST\n",
1653 ofd_name(ofd), PFID(&oa->o_oi.oi_fid),
1654 oid, -diff, PFID(&oseq->os_oi.oi_fid),
1655 ofd_seq_last_oid(oseq));
1656 /* Let MDS know that we are so far ahead. */
1657 rc = ostid_set_id(&rep_oa->o_oi,
1658 ofd_seq_last_oid(oseq) + 1);
1663 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1669 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1670 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1671 /* don't enforce grant during orphan recovery */
1672 granted = tgt_grant_create(tsi->tsi_env,
1673 ofd_obd(ofd)->obd_self_export,
1678 CDEBUG(D_HA, "%s: failed to acquire grant "
1679 "space for precreate (%lld): rc = %d\n",
1680 ofd_name(ofd), diff, rc);
1685 /* This can happen if a new OST is formatted and installed
1686 * in place of an old one at the same index. Instead of
1687 * precreating potentially millions of deleted old objects
1688 * (possibly filling the OST), only precreate the last batch.
1689 * LFSCK will eventually clean up any orphans. LU-14 */
1690 if (diff > 5 * OST_MAX_PRECREATE) {
1691 /* Message below is checked in conf-sanity test_122b */
1692 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",
1693 ofd_name(ofd), POSTID(&oa->o_oi), diff,
1694 POSTID(&oseq->os_oi),
1695 OST_MAX_PRECREATE / 2);
1696 diff = OST_MAX_PRECREATE / 2;
1697 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1701 next_id = ofd_seq_last_oid(oseq) + 1;
1702 count = ofd_precreate_batch(ofd, (int)diff);
1704 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1705 " at %llu\n", ofd_name(ofd),
1706 count, seq, next_id);
1708 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1709 && ktime_get_seconds() > enough_time) {
1710 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1711 " created at a rate of %d/s\n",
1712 ofd_name(ofd), created, diff + created,
1713 created / DISK_TIMEOUT);
1717 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1718 oseq, count, sync_trans);
1722 } else if (rc < 0) {
1728 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1729 LCONSOLE_WARN("%s: can't create the same count of"
1730 " objects when replaying the request"
1731 " (diff is %lld). see LU-4621\n",
1732 ofd_name(ofd), diff);
1735 /* some objects got created, we can return
1736 * them, even if last creation failed */
1739 CERROR("%s: unable to precreate: rc = %d\n",
1742 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1743 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1744 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1749 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1753 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1754 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1755 if (unlikely(!oseq->os_last_id_synced))
1756 oseq->os_last_id_synced = 1;
1758 mutex_unlock(&oseq->os_create_lock);
1761 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1762 ofd_seq_put(tsi->tsi_env, oseq);
1765 up_read(&ofd->ofd_lastid_rwsem);
1770 * OFD request handler for OST_DESTROY RPC.
1772 * This is OFD-specific part of request handling. It destroys data objects
1773 * related to destroyed object on MDT.
1775 * \param[in] tsi target session environment for this request
1777 * \retval 0 if successful
1778 * \retval negative value on error
1780 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1782 const struct ost_body *body = tsi->tsi_ost_body;
1783 struct ost_body *repbody;
1784 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1785 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1786 struct lu_fid *fid = &fti->fti_fid;
1787 ktime_t kstart = ktime_get();
1794 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1797 /* This is old case for clients before Lustre 2.4 */
1798 /* If there's a DLM request, cancel the locks mentioned in it */
1799 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1801 struct ldlm_request *dlm;
1803 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1806 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1809 *fid = body->oa.o_oi.oi_fid;
1810 oid = ostid_id(&body->oa.o_oi);
1813 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1815 /* check that o_misc makes sense */
1816 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1817 count = body->oa.o_misc;
1819 count = 1; /* default case - single destroy */
1821 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1822 POSTID(&body->oa.o_oi), count);
1827 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1828 if (lrc == -ENOENT) {
1830 "%s: destroying non-existent object "DFID"\n",
1831 ofd_name(ofd), PFID(fid));
1832 /* rewrite rc with -ENOENT only if it is 0 */
1835 } else if (lrc != 0) {
1836 CERROR("%s: error destroying object "DFID": %d\n",
1837 ofd_name(ofd), PFID(fid), lrc);
1843 lrc = fid_set_id(fid, oid);
1844 if (unlikely(lrc != 0 && count > 0))
1845 GOTO(out, rc = lrc);
1848 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1849 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1854 fid_to_ostid(fid, &repbody->oa.o_oi);
1859 * OFD request handler for OST_STATFS RPC.
1861 * This function gets statfs data from storage as part of request
1864 * \param[in] tsi target session environment for this request
1866 * \retval 0 if successful
1867 * \retval negative value on error
1869 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1871 ktime_t kstart = ktime_get();
1872 struct obd_statfs *osfs;
1877 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1879 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1881 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1882 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1884 CERROR("%s: statfs failed: rc = %d\n",
1885 tgt_name(tsi->tsi_tgt), rc);
1887 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1890 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1891 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1897 * OFD request handler for OST_SYNC RPC.
1899 * Sync object data or all filesystem data to the disk and pack the
1902 * \param[in] tsi target session environment for this request
1904 * \retval 0 if successful
1905 * \retval negative value on error
1907 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1909 struct ost_body *body = tsi->tsi_ost_body;
1910 struct ost_body *repbody;
1911 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1912 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1913 struct ofd_object *fo = NULL;
1914 ktime_t kstart = ktime_get();
1919 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1921 /* if no objid is specified, it means "sync whole filesystem" */
1922 if (!fid_is_zero(&tsi->tsi_fid)) {
1923 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1925 RETURN(PTR_ERR(fo));
1928 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1929 fo != NULL ? ofd_object_child(fo) : NULL,
1930 repbody->oa.o_size, repbody->oa.o_blocks);
1934 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1935 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1939 repbody->oa.o_oi = body->oa.o_oi;
1940 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1942 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1944 obdo_from_la(&repbody->oa, &fti->fti_attr,
1947 /* don't return rc from getattr */
1952 ofd_object_put(tsi->tsi_env, fo);
1957 * OFD request handler for OST_FALLOCATE RPC.
1959 * This is part of request processing. Validate request fields,
1960 * preallocate the given OFD object and pack reply.
1962 * \param[in] tsi target session environment for this request
1964 * \retval 0 if successful
1965 * \retval negative value on error
1967 static int ofd_fallocate_hdl(struct tgt_session_info *tsi)
1969 struct obdo *oa = &tsi->tsi_ost_body->oa;
1970 struct ost_body *repbody;
1971 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1972 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1973 struct ldlm_resource *res;
1974 struct ofd_object *fo;
1977 struct lustre_handle lh = { 0, };
1981 ktime_t kstart = ktime_get();
1983 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1984 if (repbody == NULL)
1985 RETURN(err_serious(-ENOMEM));
1988 * fallocate() start and end are passed in o_size and o_blocks
1989 * on the wire. Clients 2.15.0 and newer should always set
1990 * the OBD_MD_FLSIZE and OBD_MD_FLBLOCKS valid flags, but some
1991 * older client versions did not. We permit older clients to
1992 * not set these flags, checking their version by proxy using
1993 * the lack of OBD_CONNECT_TRUNCLOCK to imply 2.14.0 and older.
1995 * Return -EOPNOTSUPP to also work with older clients not
1996 * supporting newer server modes.
1998 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1999 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)
2000 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 21, 53, 0)
2001 && (tgt_conn_flags(tsi) & OBD_CONNECT_OLD_FALLOC)
2004 RETURN(-EOPNOTSUPP);
2008 /* client should already limit len >= 0 */
2012 mode = oa->o_falloc_mode;
2014 * mode == 0 (which is standard prealloc) and PUNCH is supported
2015 * Rest of mode options are not supported yet.
2017 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2018 RETURN(-EOPNOTSUPP);
2020 /* PUNCH_HOLE mode should always be accompanied with KEEP_SIZE flag
2021 * Check that and add the missing flag for such invalid call with
2024 if (mode & FALLOC_FL_PUNCH_HOLE && !(mode & FALLOC_FL_KEEP_SIZE)) {
2025 CWARN("%s: PUNCH mode misses KEEP_SIZE flag, setting it\n",
2026 tsi->tsi_tgt->lut_obd->obd_name);
2027 mode |= FALLOC_FL_KEEP_SIZE;
2030 repbody->oa.o_oi = oa->o_oi;
2031 repbody->oa.o_valid = OBD_MD_FLID;
2033 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2034 oa->o_flags & OBD_FL_SRVLOCK;
2037 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid,
2038 start, end, &lh, LCK_PW, &flags);
2043 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2046 GOTO(out, rc = PTR_ERR(fo));
2048 valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID |
2049 OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME;
2050 la_from_obdo(&info->fti_attr, oa, valid);
2052 rc = ofd_object_fallocate(tsi->tsi_env, fo, start, end, mode,
2053 &info->fti_attr, oa);
2057 rc = ofd_attr_get(tsi->tsi_env, fo, &info->fti_attr);
2059 obdo_from_la(&repbody->oa, &info->fti_attr, OFD_VALID_FLAGS);
2063 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PREALLOC,
2064 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2068 ofd_object_put(tsi->tsi_env, fo);
2071 tgt_data_unlock(&lh, LCK_PW);
2073 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2076 struct ost_lvb *res_lvb;
2078 ldlm_res_lvbo_update(res, NULL, 0);
2079 res_lvb = res->lr_lvb_data;
2081 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2082 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2084 repbody->oa.o_valid |= OBD_MD_FLSIZE;
2085 repbody->oa.o_size = res_lvb->lvb_size;
2087 ldlm_resource_putref(res);
2095 * OFD request handler for OST_PUNCH RPC.
2097 * This is part of request processing. Validate request fields,
2098 * punch (truncate) the given OFD object and pack reply.
2100 * \param[in] tsi target session environment for this request
2102 * \retval 0 if successful
2103 * \retval negative value on error
2105 static int ofd_punch_hdl(struct tgt_session_info *tsi)
2107 const struct obdo *oa = &tsi->tsi_ost_body->oa;
2108 struct ost_body *repbody;
2109 struct ofd_thread_info *info = tsi2ofd_info(tsi);
2110 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
2111 struct ldlm_resource *res;
2112 struct ofd_object *fo;
2114 struct lustre_handle lh = { 0, };
2117 ktime_t kstart = ktime_get();
2122 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
2124 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2125 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2126 RETURN(err_serious(-EPROTO));
2128 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2129 if (repbody == NULL)
2130 RETURN(err_serious(-ENOMEM));
2132 /* punch start,end are passed in o_size,o_blocks throught wire */
2136 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2139 /* standard truncate optimization: if file body is completely
2140 * destroyed, don't send data back to the server. */
2142 flags |= LDLM_FL_AST_DISCARD_DATA;
2144 repbody->oa.o_oi = oa->o_oi;
2145 repbody->oa.o_valid = OBD_MD_FLID;
2147 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2148 oa->o_flags & OBD_FL_SRVLOCK;
2151 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
2152 end, &lh, LCK_PW, &flags);
2157 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2158 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2159 oa->o_valid, start, end);
2161 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2164 GOTO(out, rc = PTR_ERR(fo));
2166 la_from_obdo(&info->fti_attr, oa,
2167 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2168 info->fti_attr.la_size = start;
2169 info->fti_attr.la_valid |= LA_SIZE;
2171 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2176 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2177 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2180 ofd_object_put(tsi->tsi_env, fo);
2183 tgt_data_unlock(&lh, LCK_PW);
2185 /* we do not call this before to avoid lu_object_find() in
2186 * ->lvbo_update() holding another reference on the object.
2187 * otherwise concurrent destroy can make the object unavailable
2188 * for 2nd lu_object_find() waiting for the first reference
2189 * to go... deadlock! */
2190 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2193 struct ost_lvb *res_lvb;
2195 ldlm_res_lvbo_update(res, NULL, 0);
2196 res_lvb = res->lr_lvb_data;
2197 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2198 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2199 ldlm_resource_putref(res);
2205 static int ofd_ladvise_prefetch(const struct lu_env *env,
2206 struct ofd_object *fo,
2207 struct niobuf_local *lnb,
2208 __u64 start, __u64 end, enum dt_bufs_type dbt)
2210 struct ofd_thread_info *info = ofd_info(env);
2211 pgoff_t start_index, end_index, pages;
2212 struct niobuf_remote rnb;
2213 unsigned long nr_local;
2219 ofd_read_lock(env, fo);
2220 if (!ofd_object_exists(fo))
2221 GOTO(out_unlock, rc = -ENOENT);
2223 rc = ofd_attr_get(env, fo, &info->fti_attr);
2225 GOTO(out_unlock, rc);
2227 if (end > info->fti_attr.la_size)
2228 end = info->fti_attr.la_size;
2231 GOTO(out_unlock, rc);
2233 /* We need page aligned offset and length */
2234 start_index = start >> PAGE_SHIFT;
2235 end_index = (end - 1) >> PAGE_SHIFT;
2236 pages = end_index - start_index + 1;
2238 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2239 PTLRPC_MAX_BRW_PAGES;
2240 rnb.rnb_offset = start_index << PAGE_SHIFT;
2241 rnb.rnb_len = nr_local << PAGE_SHIFT;
2242 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb,
2243 PTLRPC_MAX_BRW_PAGES, dbt);
2244 if (unlikely(rc < 0))
2247 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2248 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2251 start_index += nr_local;
2256 ofd_read_unlock(env, fo);
2261 * OFD request handler for OST_LADVISE RPC.
2263 * Tune cache or perfetch policies according to advices.
2265 * \param[in] tsi target session environment for this request
2267 * \retval 0 if successful
2268 * \retval negative errno on error
2270 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2272 struct ptlrpc_request *req = tgt_ses_req(tsi);
2273 struct obd_export *exp = tsi->tsi_exp;
2274 struct ofd_device *ofd = ofd_exp(exp);
2275 struct ost_body *body, *repbody;
2276 struct ofd_thread_info *info;
2277 struct ofd_object *fo;
2278 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2279 const struct lu_env *env = svc_thread->t_env;
2280 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2281 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2282 struct lu_ladvise *ladvise;
2284 struct ladvise_hdr *ladvise_hdr;
2285 struct obd_ioobj ioo;
2286 struct lustre_handle lockh = { 0 };
2289 struct dt_object *dob;
2295 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2296 body = tsi->tsi_ost_body;
2298 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2299 RETURN(err_serious(-EPROTO));
2301 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2302 &RMF_OST_LADVISE_HDR);
2303 if (ladvise_hdr == NULL)
2304 RETURN(err_serious(-EPROTO));
2306 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2307 ladvise_hdr->lah_count < 1)
2308 RETURN(err_serious(-EPROTO));
2310 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2311 RETURN(err_serious(-EPROTO));
2313 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2314 if (ladvise == NULL)
2315 RETURN(err_serious(-EPROTO));
2317 num_advise = req_capsule_get_size(&req->rq_pill,
2318 &RMF_OST_LADVISE, RCL_CLIENT) /
2320 if (num_advise < ladvise_hdr->lah_count)
2321 RETURN(err_serious(-EPROTO));
2323 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2324 repbody->oa = body->oa;
2326 info = ofd_info_init(env, exp);
2328 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2329 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2333 fo = ofd_object_find(env, ofd, &info->fti_fid);
2338 LASSERT(fo != NULL);
2339 dob = ofd_object_child(fo);
2341 if (ptlrpc_connection_is_local(exp->exp_connection))
2342 dbt |= DT_BUFS_TYPE_LOCAL;
2344 for (i = 0; i < num_advise; i++, ladvise++) {
2345 start = ladvise->lla_start;
2346 end = ladvise->lla_end;
2348 rc = err_serious(-EPROTO);
2352 /* Handle different advice types */
2353 switch (ladvise->lla_advice) {
2357 case LU_LADVISE_WILLREAD:
2361 ioo.ioo_oid = body->oa.o_oi;
2363 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2364 &tsi->tsi_resid, start, end - 1,
2365 &lockh, LCK_PR, &flags);
2369 req->rq_status = ofd_ladvise_prefetch(env, fo,
2372 tgt_data_unlock(&lockh, LCK_PR);
2374 case LU_LADVISE_DONTNEED:
2375 rc = dt_ladvise(env, dob, ladvise->lla_start,
2376 ladvise->lla_end, LU_LADVISE_DONTNEED);
2383 ofd_object_put(env, fo);
2384 req->rq_status = rc;
2389 * OFD request handler for OST_QUOTACTL RPC.
2391 * This is part of request processing to validate incoming request fields,
2392 * get the requested data from OSD and pack reply.
2394 * \param[in] tsi target session environment for this request
2396 * \retval 0 if successful
2397 * \retval negative value on error
2399 static int ofd_quotactl(struct tgt_session_info *tsi)
2401 struct obd_quotactl *oqctl, *repoqc;
2402 struct lu_nodemap *nodemap;
2403 ktime_t kstart = ktime_get();
2409 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2411 RETURN(err_serious(-EPROTO));
2413 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2415 RETURN(err_serious(-ENOMEM));
2419 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2420 if (IS_ERR(nodemap))
2421 RETURN(PTR_ERR(nodemap));
2424 if (oqctl->qc_type == USRQUOTA)
2425 id = nodemap_map_id(nodemap, NODEMAP_UID,
2426 NODEMAP_CLIENT_TO_FS,
2428 else if (oqctl->qc_type == GRPQUOTA)
2429 id = nodemap_map_id(nodemap, NODEMAP_GID,
2430 NODEMAP_CLIENT_TO_FS,
2432 else if (oqctl->qc_type == PRJQUOTA)
2433 id = nodemap_map_id(nodemap, NODEMAP_PROJID,
2434 NODEMAP_CLIENT_TO_FS,
2437 nodemap_putref(nodemap);
2439 if (repoqc->qc_id != id)
2440 swap(repoqc->qc_id, id);
2442 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2444 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2445 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2447 if (repoqc->qc_id != id)
2448 swap(repoqc->qc_id, id);
2454 * Prolong lock timeout for the given extent.
2456 * This function finds all locks related with incoming request and
2457 * prolongs their timeout.
2459 * If a client is holding a lock for a long time while it sends
2460 * read or write RPCs to the OST for the object under this lock,
2461 * then we don't want the OST to evict the client. Otherwise,
2462 * if the network or disk is very busy then the client may not
2463 * be able to make any progress to clear out dirty pages under
2464 * the lock and the application will fail.
2466 * Every time a Bulk Read/Write (BRW) request arrives for the object
2467 * covered by the lock, extend the timeout on that lock. The RPC should
2468 * contain a lock handle for the lock it is using, but this
2469 * isn't handled correctly by all client versions, and the
2470 * request may cover multiple locks.
2472 * \param[in] tsi target session environment for this request
2473 * \param[in] data struct of data to prolong locks
2476 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2477 struct ldlm_prolong_args *data)
2479 struct obdo *oa = &tsi->tsi_ost_body->oa;
2480 struct ldlm_lock *lock;
2484 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2485 data->lpa_export = tsi->tsi_exp;
2486 data->lpa_resid = tsi->tsi_resid;
2488 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2489 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2490 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2491 data->lpa_extent.end);
2493 if (oa->o_valid & OBD_MD_FLHANDLE) {
2494 /* mostly a request should be covered by only one lock, try
2496 lock = ldlm_handle2lock(&oa->o_handle);
2498 /* Fast path to check if the lock covers the whole IO
2499 * region exclusively. */
2500 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2501 &data->lpa_extent)) {
2503 LASSERT(lock->l_export == data->lpa_export);
2504 ldlm_lock_prolong_one(lock, data);
2505 LDLM_LOCK_PUT(lock);
2506 if (data->lpa_locks_cnt > 0)
2508 /* The lock was destroyed probably lets try
2511 lock->l_last_used = ktime_get();
2512 LDLM_LOCK_PUT(lock);
2517 ldlm_resource_prolong(data);
2522 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2524 * Determine if \a lock and the lock from request \a req are equivalent
2525 * by comparing their resource names, modes, and extents.
2527 * It is used to give priority to read and write RPCs being done
2528 * under this lock so that the client can drop the contended
2529 * lock more quickly and let other clients use it. This improves
2530 * overall performance in the case where the first client gets a
2531 * very large lock extent that prevents other clients from
2532 * submitting their writes.
2534 * \param[in] req ptlrpc_request being processed
2535 * \param[in] lock contended lock to match
2537 * \retval 1 if lock is matched
2538 * \retval 0 otherwise
2540 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2541 struct ldlm_lock *lock)
2543 struct niobuf_remote *rnb;
2544 struct obd_ioobj *ioo;
2545 enum ldlm_mode mode;
2546 struct ldlm_extent ext;
2547 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2551 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2552 LASSERT(ioo != NULL);
2554 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2555 LASSERT(rnb != NULL);
2557 ext.start = rnb->rnb_offset;
2558 rnb += ioo->ioo_bufcnt - 1;
2559 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2561 LASSERT(lock->l_resource != NULL);
2562 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2565 /* a bulk write can only hold a reference on a PW extent lock
2568 mode = LCK_PW | LCK_GROUP;
2569 if (opc == OST_READ)
2570 /* whereas a bulk read can be protected by either a PR or PW
2574 if (!(lock->l_granted_mode & mode))
2577 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2581 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2583 * Check for whether the given PTLRPC request (\a req) is blocking
2584 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2587 * \param[in] req the incoming request
2589 * \retval 1 if \a req is blocking an LDLM lock cancel
2590 * \retval 0 if it is not
2591 * \retval -ESTALE if lock is not found
2593 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2595 struct tgt_session_info *tsi;
2596 struct obd_ioobj *ioo;
2597 struct niobuf_remote *rnb;
2599 struct ldlm_prolong_args pa = { 0 };
2603 /* Don't use tgt_ses_info() to get session info, because lock_match()
2604 * can be called while request has no processing thread yet. */
2605 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2608 * Use LASSERT below because malformed RPCs should have
2609 * been filtered out in tgt_hpreq_handler().
2611 opc = lustre_msg_get_opc(req->rq_reqmsg);
2612 LASSERT(opc == OST_READ || opc == OST_WRITE);
2614 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2615 LASSERT(ioo != NULL);
2617 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2618 LASSERT(rnb != NULL);
2619 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2621 pa.lpa_mode = LCK_PW | LCK_GROUP;
2622 if (opc == OST_READ)
2623 pa.lpa_mode |= LCK_PR;
2625 pa.lpa_extent.start = rnb->rnb_offset;
2626 rnb += ioo->ioo_bufcnt - 1;
2627 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2629 DEBUG_REQ(D_RPCTRACE, req,
2630 "%s %s: refresh rw locks for "DFID" (%llu->%llu)",
2631 tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid),
2632 pa.lpa_extent.start, pa.lpa_extent.end);
2634 ofd_prolong_extent_locks(tsi, &pa);
2636 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n",
2637 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2639 if (pa.lpa_blocks_cnt > 0)
2642 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2646 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2648 * Called after the request has been handled. It refreshes lock timeout again
2649 * so that client has more time to send lock cancel RPC.
2651 * \param[in] req request which is being processed.
2653 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2655 ofd_rw_hpreq_check(req);
2659 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2661 * This function checks if the given lock is the same by its resname, mode
2662 * and extent as one taken from the request.
2663 * It is used to give priority to punch/truncate RPCs that might lead to
2664 * the fastest release of that lock when a lock is contended.
2666 * \param[in] req ptlrpc_request being processed
2667 * \param[in] lock contended lock to match
2669 * \retval 1 if lock is matched
2670 * \retval 0 otherwise
2672 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2673 struct ldlm_lock *lock)
2675 struct tgt_session_info *tsi;
2677 struct ldlm_extent ext;
2681 /* Don't use tgt_ses_info() to get session info, because lock_match()
2682 * can be called while request has no processing thread yet. */
2683 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2686 * Use LASSERT below because malformed RPCs should have
2687 * been filtered out in tgt_hpreq_handler().
2689 LASSERT(tsi->tsi_ost_body != NULL);
2690 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2691 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2694 oa = &tsi->tsi_ost_body->oa;
2695 ext.start = oa->o_size;
2696 ext.end = oa->o_blocks;
2698 LASSERT(lock->l_resource != NULL);
2699 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2702 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2705 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2709 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2711 * High-priority queue request check for whether the given punch request
2712 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2713 * covered by an LDLM lock.
2717 * \param[in] req the incoming request
2719 * \retval 1 if \a req is blocking an LDLM lock cancel
2720 * \retval 0 if it is not
2721 * \retval -ESTALE if lock is not found
2723 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2725 struct tgt_session_info *tsi;
2727 struct ldlm_prolong_args pa = { 0 };
2731 /* Don't use tgt_ses_info() to get session info, because lock_match()
2732 * can be called while request has no processing thread yet. */
2733 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2734 LASSERT(tsi != NULL);
2735 oa = &tsi->tsi_ost_body->oa;
2737 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2738 oa->o_flags & OBD_FL_SRVLOCK));
2740 pa.lpa_mode = LCK_PW | LCK_GROUP;
2741 pa.lpa_extent.start = oa->o_size;
2742 pa.lpa_extent.end = oa->o_blocks;
2745 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2746 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2747 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2749 ofd_prolong_extent_locks(tsi, &pa);
2751 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2752 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2754 if (pa.lpa_blocks_cnt > 0)
2757 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2761 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2763 * Called after the request has been handled. It refreshes lock timeout again
2764 * so that client has more time to send lock cancel RPC.
2766 * \param[in] req request which is being processed.
2768 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2770 ofd_punch_hpreq_check(req);
2773 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2774 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2775 .hpreq_check = ofd_rw_hpreq_check,
2776 .hpreq_fini = ofd_rw_hpreq_fini
2779 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2780 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2781 .hpreq_check = ofd_punch_hpreq_check,
2782 .hpreq_fini = ofd_punch_hpreq_fini
2786 * Assign high priority operations to an IO request.
2788 * Check if the incoming request is a candidate for
2789 * high-priority processing. If it is, assign it a high
2790 * priority operations table.
2792 * \param[in] tsi target session environment for this request
2794 static void ofd_hp_brw(struct tgt_session_info *tsi)
2796 struct niobuf_remote *rnb;
2797 struct obd_ioobj *ioo;
2801 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2802 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2803 if (ioo->ioo_bufcnt > 0) {
2804 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2805 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2807 /* no high priority if server lock is needed */
2808 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2809 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2813 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2817 * Assign high priority operations to an punch request.
2819 * Check if the incoming request is a candidate for
2820 * high-priority processing. If it is, assign it a high
2821 * priority operations table.
2823 * \param[in] tsi target session environment for this request
2825 static void ofd_hp_punch(struct tgt_session_info *tsi)
2827 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2828 /* no high-priority if server lock is needed */
2829 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2830 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2831 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2832 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2834 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2837 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2838 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2839 #define OST_BRW_READ OST_READ
2840 #define OST_BRW_WRITE OST_WRITE
2843 * Table of OFD-specific request handlers
2845 * This table contains all opcodes accepted by OFD and
2846 * specifies handlers for them. The tgt_request_handler()
2847 * uses such table from each target to process incoming
2850 static struct tgt_handler ofd_tgt_handlers[] = {
2851 TGT_RPC_HANDLER(OST_FIRST_OPC,
2852 0, OST_CONNECT, tgt_connect,
2853 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2854 TGT_RPC_HANDLER(OST_FIRST_OPC,
2855 0, OST_DISCONNECT, tgt_disconnect,
2856 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2857 TGT_RPC_HANDLER(OST_FIRST_OPC,
2858 0, OST_SET_INFO, ofd_set_info_hdl,
2859 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2860 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2861 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl),
2862 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2863 OST_SETATTR, ofd_setattr_hdl),
2864 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2865 OST_CREATE, ofd_create_hdl),
2866 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2867 OST_DESTROY, ofd_destroy_hdl),
2868 TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl),
2869 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
2871 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2872 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write,
2874 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2875 OST_PUNCH, ofd_punch_hdl,
2877 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl),
2878 TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl),
2879 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl),
2880 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE, ofd_fallocate_hdl),
2881 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
2884 static struct tgt_opc_slice ofd_common_slice[] = {
2886 .tos_opc_start = OST_FIRST_OPC,
2887 .tos_opc_end = OST_LAST_OPC,
2888 .tos_hs = ofd_tgt_handlers
2891 .tos_opc_start = OBD_FIRST_OPC,
2892 .tos_opc_end = OBD_LAST_OPC,
2893 .tos_hs = tgt_obd_handlers
2896 .tos_opc_start = LDLM_FIRST_OPC,
2897 .tos_opc_end = LDLM_LAST_OPC,
2898 .tos_hs = tgt_dlm_handlers
2901 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2902 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2903 .tos_hs = tgt_out_handlers
2906 .tos_opc_start = SEQ_FIRST_OPC,
2907 .tos_opc_end = SEQ_LAST_OPC,
2908 .tos_hs = seq_handlers
2911 .tos_opc_start = LFSCK_FIRST_OPC,
2912 .tos_opc_end = LFSCK_LAST_OPC,
2913 .tos_hs = tgt_lfsck_handlers
2916 .tos_opc_start = SEC_FIRST_OPC,
2917 .tos_opc_end = SEC_LAST_OPC,
2918 .tos_hs = tgt_sec_ctx_handlers
2925 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2926 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2929 * Implementation of lu_context_key::lct_key_exit.
2931 * Optional method called on lu_context_exit() for all allocated
2933 * It is used in OFD to sanitize context values which may be re-used
2934 * during another request processing by the same thread.
2936 * \param[in] ctx execution context
2937 * \param[in] key context key
2938 * \param[in] data ofd_thread_info
2940 static void ofd_key_exit(const struct lu_context *ctx,
2941 struct lu_context_key *key, void *data)
2943 struct ofd_thread_info *info = data;
2945 info->fti_env = NULL;
2946 info->fti_exp = NULL;
2949 info->fti_pre_version = 0;
2951 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2954 struct lu_context_key ofd_thread_key = {
2955 .lct_tags = LCT_DT_THREAD,
2956 .lct_init = ofd_key_init,
2957 .lct_fini = ofd_key_fini,
2958 .lct_exit = ofd_key_exit
2962 * Initialize OFD device according to parameters in the config log \a cfg.
2964 * This is the main starting point of OFD initialization. It fills all OFD
2965 * parameters with their initial values and calls other initializing functions
2966 * to set up all OFD subsystems.
2968 * \param[in] env execution environment
2969 * \param[in] m OFD device
2970 * \param[in] ldt LU device type of OFD
2971 * \param[in] cfg configuration log
2973 * \retval 0 if successful
2974 * \retval negative value on error
2976 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2977 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2979 const char *dev = lustre_cfg_string(cfg, 0);
2980 struct ofd_thread_info *info = NULL;
2981 struct obd_device *obd;
2982 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2984 struct nm_config_file *nodemap_config;
2985 struct obd_device_target *obt;
2991 obd = class_name2obd(dev);
2993 CERROR("Cannot find obd with name %s\n", dev);
2997 rc = lu_env_refill((struct lu_env *)env);
3002 obt->obt_magic = OBT_MAGIC;
3004 spin_lock_init(&m->ofd_flags_lock);
3005 m->ofd_raid_degraded = 0;
3006 m->ofd_sync_journal = 0;
3008 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
3010 m->ofd_seq_count = 0;
3011 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
3012 spin_lock_init(&m->ofd_inconsistency_lock);
3014 m->ofd_access_log_mask = -1; /* Log all accesses if enabled. */
3016 spin_lock_init(&m->ofd_batch_lock);
3017 init_rwsem(&m->ofd_lastid_rwsem);
3019 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
3020 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
3021 /* set this lu_device to obd, because error handling need it */
3022 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
3024 /* No connection accepted until configurations will finish */
3025 spin_lock(&obd->obd_dev_lock);
3026 obd->obd_no_conn = 1;
3027 spin_unlock(&obd->obd_dev_lock);
3028 obd->obd_replayable = 1;
3029 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
3030 char *str = lustre_cfg_string(cfg, 4);
3032 if (strchr(str, 'n')) {
3033 CWARN("%s: recovery disabled\n", obd->obd_name);
3034 obd->obd_replayable = 0;
3038 info = ofd_info_init(env, NULL);
3042 rc = ofd_stack_init(env, m, cfg, &lmd_flags);
3044 CERROR("%s: can't init device stack, rc %d\n",
3049 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
3050 ofd_procfs_add_brw_stats_symlink(m);
3053 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
3054 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
3055 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
3056 LDLM_NAMESPACE_SERVER,
3057 LDLM_NAMESPACE_GREEDY,
3059 if (IS_ERR(m->ofd_namespace)) {
3060 rc = PTR_ERR(m->ofd_namespace);
3061 CERROR("%s: unable to create server namespace: rc = %d\n",
3063 m->ofd_namespace = NULL;
3064 GOTO(err_fini_stack, rc);
3066 /* set obd_namespace for compatibility with old code */
3067 obd->obd_namespace = m->ofd_namespace;
3068 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
3069 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
3070 m->ofd_namespace->ns_lvbp = m;
3072 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3073 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
3075 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
3076 OBD_FAIL_OST_ALL_REQUEST_NET,
3077 OBD_FAIL_OST_ALL_REPLY_NET);
3079 GOTO(err_free_ns, rc);
3081 if (lmd_flags & LMD_FLG_SKIP_LFSCK)
3082 m->ofd_skip_lfsck = 1;
3083 if (lmd_flags & LMD_FLG_LOCAL_RECOV)
3084 m->ofd_lut.lut_local_recovery = 1;
3086 rc = ofd_tunables_init(m);
3088 GOTO(err_fini_lut, rc);
3090 tgd->tgd_reserved_pcnt = 0;
3092 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
3093 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
3094 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
3095 OFD_PRECREATE_SMALL_FS)
3096 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
3097 m->ofd_atime_diff = OFD_DEF_ATIME_DIFF;
3099 rc = ofd_fs_setup(env, m, obd);
3101 GOTO(err_fini_proc, rc);
3103 fid.f_seq = FID_SEQ_LOCAL_NAME;
3106 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
3109 GOTO(err_fini_fs, rc);
3111 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
3113 if (IS_ERR(nodemap_config)) {
3114 rc = PTR_ERR(nodemap_config);
3116 GOTO(err_fini_los, rc);
3118 obt->obt_nodemap_config_file = nodemap_config;
3121 rc = ofd_start_inconsistency_verification_thread(m);
3123 GOTO(err_fini_nm, rc);
3125 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
3130 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
3131 obt->obt_nodemap_config_file = NULL;
3133 local_oid_storage_fini(env, m->ofd_los);
3136 ofd_fs_cleanup(env, m);
3140 tgt_fini(env, &m->ofd_lut);
3142 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3143 obd->obd_namespace = m->ofd_namespace = NULL;
3145 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3150 * Stop the OFD device
3152 * This function stops the OFD device and all its subsystems.
3153 * This is the end of OFD lifecycle.
3155 * \param[in] env execution environment
3156 * \param[in] m OFD device
3158 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3160 struct obd_device *obd = ofd_obd(m);
3161 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3162 struct lfsck_stop stop;
3164 stop.ls_status = LS_PAUSED;
3166 lfsck_stop(env, m->ofd_osd, &stop);
3167 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3168 target_recovery_fini(obd);
3169 if (m->ofd_namespace != NULL)
3170 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3171 d->ld_obd->obd_force);
3173 obd_exports_barrier(obd);
3174 obd_zombie_barrier();
3177 tgt_fini(env, &m->ofd_lut);
3178 ofd_stop_inconsistency_verification_thread(m);
3179 lfsck_degister(env, m->ofd_osd);
3180 ofd_fs_cleanup(env, m);
3181 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3182 obd->u.obt.obt_nodemap_config_file = NULL;
3184 if (m->ofd_namespace != NULL) {
3185 ldlm_namespace_free_post(m->ofd_namespace);
3186 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3189 ofd_access_log_delete(m->ofd_access_log);
3190 m->ofd_access_log = NULL;
3192 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3194 LASSERT(atomic_read(&d->ld_ref) == 0);
3195 server_put_mount(obd->obd_name, true);
3200 * Implementation of lu_device_type_operations::ldto_device_fini.
3202 * Finalize device. Dual to ofd_device_init(). It is called from
3203 * obd_precleanup() and stops the current device.
3205 * \param[in] env execution environment
3206 * \param[in] d LU device of OFD
3210 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3211 struct lu_device *d)
3214 ofd_fini(env, ofd_dev(d));
3219 * Implementation of lu_device_type_operations::ldto_device_free.
3221 * Free OFD device. Dual to ofd_device_alloc().
3223 * \param[in] env execution environment
3224 * \param[in] d LU device of OFD
3228 static struct lu_device *ofd_device_free(const struct lu_env *env,
3229 struct lu_device *d)
3231 struct ofd_device *m = ofd_dev(d);
3233 dt_device_fini(&m->ofd_dt_dev);
3239 * Implementation of lu_device_type_operations::ldto_device_alloc.
3241 * This function allocates the new OFD device. It is called from
3242 * obd_setup() if OBD device had lu_device_type defined.
3244 * \param[in] env execution environment
3245 * \param[in] t lu_device_type of OFD device
3246 * \param[in] cfg configuration log
3248 * \retval pointer to the lu_device of just allocated OFD
3249 * \retval ERR_PTR of return value on error
3251 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3252 struct lu_device_type *t,
3253 struct lustre_cfg *cfg)
3255 struct ofd_device *m;
3256 struct lu_device *l;
3261 return ERR_PTR(-ENOMEM);
3263 l = &m->ofd_dt_dev.dd_lu_dev;
3264 dt_device_init(&m->ofd_dt_dev, t);
3265 rc = ofd_init0(env, m, t, cfg);
3267 ofd_device_free(env, l);
3274 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3275 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3277 static const struct lu_device_type_operations ofd_device_type_ops = {
3278 .ldto_init = ofd_type_init,
3279 .ldto_fini = ofd_type_fini,
3281 .ldto_start = ofd_type_start,
3282 .ldto_stop = ofd_type_stop,
3284 .ldto_device_alloc = ofd_device_alloc,
3285 .ldto_device_free = ofd_device_free,
3286 .ldto_device_fini = ofd_device_fini
3289 static struct lu_device_type ofd_device_type = {
3290 .ldt_tags = LU_DEVICE_DT,
3291 .ldt_name = LUSTRE_OST_NAME,
3292 .ldt_ops = &ofd_device_type_ops,
3293 .ldt_ctx_tags = LCT_DT_THREAD
3297 * Initialize OFD module.
3299 * This function is called upon module loading. It registers OFD device type
3300 * and prepares all in-memory structures used by all OFD devices.
3302 * \retval 0 if successful
3303 * \retval negative value on error
3305 static int __init ofd_init(void)
3309 rc = lu_kmem_init(ofd_caches);
3313 rc = ofd_access_log_module_init();
3317 rc = class_register_type(&ofd_obd_ops, NULL, true,
3318 LUSTRE_OST_NAME, &ofd_device_type);
3320 goto out_ofd_access_log;
3325 ofd_access_log_module_exit();
3327 lu_kmem_fini(ofd_caches);
3335 * This function is called upon OFD module unloading.
3336 * It frees all related structures and unregisters OFD device type.
3338 static void __exit ofd_exit(void)
3340 class_unregister_type(LUSTRE_OST_NAME);
3341 ofd_access_log_module_exit();
3342 lu_kmem_fini(ofd_caches);
3345 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3346 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3347 MODULE_VERSION(LUSTRE_VERSION_STRING);
3348 MODULE_LICENSE("GPL");
3350 module_init(ofd_init);
3351 module_exit(ofd_exit);