4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/ofd/ofd_dev.c
34 * This file contains OSD API methods for OBD Filter Device (OFD),
35 * request handlers and supplemental functions to set OFD up and clean it up.
37 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
38 * Author: Mike Pershin <mike.pershin@intel.com>
39 * Author: Johann Lombardi <johann.lombardi@intel.com>
42 * The OBD Filter Device (OFD) module belongs to the Object Storage
43 * Server stack and connects the RPC oriented Unified Target (TGT)
44 * layer (see lustre/include/lu_target.h) to the storage oriented OSD
45 * layer (see Documentation/osd-api.txt).
53 * OFD implements the LU and OBD device APIs and is responsible for:
55 * - Handling client requests (create, destroy, bulk IO, setattr,
56 * get_info, set_info, statfs) for the objects belonging to the OST
57 * (together with TGT).
59 * - Providing grant space management which allows clients to reserve
60 * disk space for data writeback. OFD tracks grants on global and
63 * - Handling object precreation requests from MDTs.
65 * - Operating the LDLM service that allows clients to maintain object
66 * data cache coherence.
69 #define DEBUG_SUBSYSTEM S_FILTER
71 #include <obd_class.h>
72 #include <obd_cksum.h>
73 #include <uapi/linux/lustre/lustre_param.h>
74 #include <lustre_fid.h>
75 #include <lustre_lfsck.h>
76 #include <lustre_dlm.h>
77 #include <lustre_quota.h>
78 #include <lustre_nodemap.h>
79 #include <lustre_log.h>
80 #include <linux/falloc.h>
82 #include "ofd_internal.h"
84 /* Slab for OFD object allocation */
85 static struct kmem_cache *ofd_object_kmem;
86 static struct lu_kmem_descr ofd_caches[] = {
88 .ckd_cache = &ofd_object_kmem,
89 .ckd_name = "ofd_obj",
90 .ckd_size = sizeof(struct ofd_object)
98 * Connect OFD to the next device in the stack.
100 * This function is used for device stack configuration and links OFD
101 * device with bottom OSD device.
103 * \param[in] env execution environment
104 * \param[in] m OFD device
105 * \param[in] next name of next device in the stack
106 * \param[out] exp export to return
108 * \retval 0 and export in \a exp if successful
109 * \retval negative value on error
111 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
112 const char *next, struct obd_export **exp)
114 struct obd_connect_data *data = NULL;
115 struct obd_device *obd;
121 GOTO(out, rc = -ENOMEM);
123 obd = class_name2obd(next);
125 CERROR("%s: can't locate next device: %s\n",
127 GOTO(out, rc = -ENOTCONN);
130 data->ocd_connect_flags = OBD_CONNECT_VERSION;
131 data->ocd_version = LUSTRE_VERSION_CODE;
133 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
135 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
136 ofd_name(m), next, rc);
140 m->ofd_dt_dev.dd_lu_dev.ld_site =
141 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
142 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
143 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
144 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
153 * Initialize stack of devices.
155 * This function initializes OFD-OSD device stack to serve OST requests
157 * \param[in] env execution environment
158 * \param[in] m OFD device
159 * \param[in] cfg Lustre config for this server
161 * \retval 0 if successful
162 * \retval negative value on error
164 static int ofd_stack_init(const struct lu_env *env,
165 struct ofd_device *m, struct lustre_cfg *cfg,
168 const char *dev = lustre_cfg_string(cfg, 0);
170 struct ofd_thread_info *info = ofd_info(env);
171 struct lustre_mount_info *lmi;
172 struct lustre_mount_data *lmd;
178 lmi = server_get_mount(dev);
180 CERROR("Cannot get mount info for %s!\n", dev);
184 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
186 if (lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
187 m->ofd_skip_lfsck = 1;
188 if (lmd->lmd_flags & LMD_FLG_NO_PRECREATE)
189 m->ofd_no_precreate = 1;
190 *lmd_flags = lmd->lmd_flags;
193 /* find bottom osd */
194 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
198 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
199 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
200 OBD_FREE(osdname, MTI_NAME_MAXLEN);
204 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
206 m->ofd_osd = lu2dt_dev(d);
208 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
209 "%s-osd", lustre_cfg_string(cfg, 0));
215 * Finalize the device stack OFD-OSD.
217 * This function cleans OFD-OSD device stack and
218 * disconnects OFD from the OSD.
220 * \param[in] env execution environment
221 * \param[in] m OFD device
222 * \param[in] top top device of stack
224 * \retval 0 if successful
225 * \retval negative value on error
227 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
228 struct lu_device *top)
230 struct obd_device *obd = ofd_obd(m);
231 struct lustre_cfg_bufs bufs;
232 struct lustre_cfg *lcfg;
237 lu_site_purge(env, top->ld_site, ~0);
238 /* process cleanup, pass mdt obd name to get obd umount flags */
239 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
244 lustre_cfg_bufs_set_string(&bufs, 1, flags);
245 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
248 lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs);
251 top->ld_ops->ldo_process_config(env, top, lcfg);
252 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
254 if (m->ofd_los != NULL) {
255 local_oid_storage_fini(env, m->ofd_los);
259 lu_site_purge(env, top->ld_site, ~0);
260 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
261 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
262 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
265 LASSERT(m->ofd_osd_exp);
266 obd_disconnect(m->ofd_osd_exp);
271 static void ofd_stack_pre_fini(const struct lu_env *env, struct ofd_device *m,
272 struct lu_device *top)
274 struct lustre_cfg_bufs bufs;
275 struct lustre_cfg *lcfg;
280 lustre_cfg_bufs_reset(&bufs, ofd_name(m));
281 lustre_cfg_bufs_set_string(&bufs, 1, NULL);
282 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
284 CERROR("%s: failed to trigger LCFG_PRE_CLEANUP\n", ofd_name(m));
286 lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, &bufs);
287 top->ld_ops->ldo_process_config(env, top, lcfg);
288 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
289 lcfg->lcfg_buflens));
295 /* For interoperability, see mdt_interop_param[]. */
296 static struct cfg_interop_param ofd_interop_param[] = {
297 { "ost.quota_type", NULL },
302 * Check if parameters are symlinks to the OSD.
304 * Some parameters were moved from ofd to osd and only their
305 * symlinks were kept in ofd by LU-3106. They are:
306 * -writehthrough_cache_enable
307 * -readcache_max_filesize
311 * Since they are not included by the static lprocfs var list, a pre-check
312 * is added for them to avoid "unknown param" errors. If they are matched
313 * in this check, they will be passed to the OSD directly.
315 * \param[in] param parameters to check
317 * \retval true if param is symlink to OSD param
320 static bool match_symlink_param(char *param)
325 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
326 sval = strchr(param, '=');
328 paramlen = sval - param;
329 if (strncmp(param, "brw_stats", paramlen) == 0)
338 * Process various configuration parameters.
340 * This function is used by MGS to process specific configurations and
341 * pass them through to the next device in server stack, i.e. the OSD.
343 * \param[in] env execution environment
344 * \param[in] d LU device of OFD
345 * \param[in] cfg parameters to process
347 * \retval 0 if successful
348 * \retval negative value on error
350 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
351 struct lustre_cfg *cfg)
353 struct ofd_device *m = ofd_dev(d);
354 struct dt_device *dt_next = m->ofd_osd;
355 struct lu_device *next = &dt_next->dd_lu_dev;
360 switch (cfg->lcfg_command) {
362 /* For interoperability */
363 struct cfg_interop_param *ptr = NULL;
364 struct lustre_cfg *old_cfg = NULL;
368 param = lustre_cfg_string(cfg, 1);
370 CERROR("param is empty\n");
375 ptr = class_find_old_param(param, ofd_interop_param);
377 if (ptr->new_param == NULL) {
379 CWARN("For interoperability, skip this %s."
380 " It is obsolete.\n", ptr->old_param);
384 CWARN("Found old param %s, changed it to %s.\n",
385 ptr->old_param, ptr->new_param);
388 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
395 if (match_symlink_param(param)) {
396 rc = next->ld_ops->ldo_process_config(env, next, cfg);
400 count = class_modify_config(cfg, PARAM_OST,
401 &d->ld_obd->obd_kset.kobj);
406 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
408 /* we don't understand; pass it on */
409 rc = next->ld_ops->ldo_process_config(env, next, cfg);
412 case LCFG_SPTLRPC_CONF: {
417 /* others are passed further */
418 rc = next->ld_ops->ldo_process_config(env, next, cfg);
425 * Implementation of lu_object_operations::loo_object_init for OFD
427 * Allocate just the next object (OSD) in stack.
429 * \param[in] env execution environment
430 * \param[in] o lu_object of OFD object
431 * \param[in] conf additional configuration parameters, not used here
433 * \retval 0 if successful
434 * \retval negative value on error
436 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
437 const struct lu_object_conf *conf)
439 struct ofd_device *d = ofd_dev(o->lo_dev);
440 struct lu_device *under;
441 struct lu_object *below;
446 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
447 PFID(lu_object_fid(o)));
449 under = &d->ofd_osd->dd_lu_dev;
450 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
452 lu_object_add(o, below);
459 static void ofd_object_free_rcu(struct rcu_head *head)
461 struct ofd_object *of = container_of(head, struct ofd_object,
464 kmem_cache_free(ofd_object_kmem, of);
468 * Implementation of lu_object_operations::loo_object_free.
470 * Finish OFD object lifecycle and free its memory.
472 * \param[in] env execution environment
473 * \param[in] o LU object of OFD object
475 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
477 struct ofd_object *of = ofd_obj(o);
478 struct lu_object_header *h;
483 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
484 PFID(lu_object_fid(o)));
487 lu_object_header_fini(h);
488 OBD_FREE_PRE(of, sizeof(*of), "slab-freed");
489 call_rcu(&of->ofo_header.loh_rcu, ofd_object_free_rcu);
494 * Implementation of lu_object_operations::loo_object_print.
496 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
497 * LU_OBJECT_DEBUG() for more details about the compound object printing.
499 * \param[in] env execution environment
500 * \param[in] cookie opaque data passed to the printer function
501 * \param[in] p printer function to use
502 * \param[in] o LU object of OFD object
504 * \retval 0 if successful
505 * \retval negative value on error
507 static int ofd_object_print(const struct lu_env *env, void *cookie,
508 lu_printer_t p, const struct lu_object *o)
510 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
513 static struct lu_object_operations ofd_obj_ops = {
514 .loo_object_init = ofd_object_init,
515 .loo_object_free = ofd_object_free,
516 .loo_object_print = ofd_object_print
520 * Implementation of lu_device_operations::lod_object_alloc.
522 * This function allocates OFD part of compound OFD-OSD object and
523 * initializes its header, because OFD is the top device in stack
525 * \param[in] env execution environment
526 * \param[in] hdr object header, NULL for OFD
527 * \param[in] d lu_device
529 * \retval allocated object if successful
530 * \retval NULL value on failed allocation
532 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
533 const struct lu_object_header *hdr,
536 struct ofd_object *of;
540 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
543 struct lu_object_header *h;
545 o = &of->ofo_obj.do_lu;
547 lu_object_header_init(h);
548 lu_object_init(o, h, d);
549 lu_object_add_top(h, o);
550 o->lo_ops = &ofd_obj_ops;
558 * Return the result of LFSCK run to the OFD.
560 * Notify OFD about result of LFSCK run. That may block the new object
561 * creation until problem is fixed by LFSCK.
563 * \param[in] env execution environment
564 * \param[in] data pointer to the OFD device
565 * \param[in] event LFSCK event type
567 * \retval 0 if successful
568 * \retval negative value on unknown event
570 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
571 enum lfsck_events event)
573 struct ofd_device *ofd = data;
574 struct obd_device *obd = ofd_obd(ofd);
577 case LE_LASTID_REBUILDING:
578 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
579 "on the device until the LAST_ID rebuilt successfully.\n",
581 down_write(&ofd->ofd_lastid_rwsem);
582 ofd->ofd_lastid_rebuilding = 1;
583 up_write(&ofd->ofd_lastid_rwsem);
585 case LE_LASTID_REBUILT: {
586 down_write(&ofd->ofd_lastid_rwsem);
587 ofd_seqs_free(env, ofd);
588 ofd->ofd_lastid_rebuilding = 0;
589 ofd->ofd_lastid_gen++;
590 up_write(&ofd->ofd_lastid_rwsem);
591 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
596 CERROR("%s: unknown lfsck event: rc = %d\n",
597 ofd_name(ofd), event);
605 * Implementation of lu_device_operations::ldo_prepare.
607 * This method is called after layer has been initialized and before it starts
608 * serving user requests. In OFD it starts lfsk check routines and initializes
611 * \param[in] env execution environment
612 * \param[in] pdev higher device in stack, NULL for OFD
613 * \param[in] dev lu_device of OFD device
615 * \retval 0 if successful
616 * \retval negative value on error
618 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
619 struct lu_device *dev)
621 struct ofd_thread_info *info;
622 struct ofd_device *ofd = ofd_dev(dev);
623 struct obd_device *obd = ofd_obd(ofd);
624 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
629 info = ofd_info_init(env, NULL);
633 /* initialize lower device */
634 rc = next->ld_ops->ldo_prepare(env, dev, next);
638 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
639 ofd_lfsck_out_notify, ofd, false);
641 CERROR("%s: failed to initialize lfsck: rc = %d\n",
646 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
647 /* The LFSCK instance is registered just now, so it must be there when
648 * register the namespace to such instance. */
649 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
651 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
652 OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_OST_PREPARE_DELAY, OBD_FAIL_ONCE,
653 (OBD_TIMEOUT_DEFAULT + 1) / 4);
654 LASSERT(obd->obd_no_conn);
655 spin_lock(&obd->obd_dev_lock);
656 obd->obd_no_conn = 0;
657 spin_unlock(&obd->obd_dev_lock);
659 if (obd->obd_recovering == 0)
660 ofd_postrecov(env, ofd);
666 * Implementation of lu_device_operations::ldo_recovery_complete.
668 * This method notifies all layers about 'recovery complete' event. That means
669 * device is in full state and consistent. An OFD calculates available grant
670 * space upon this event.
672 * \param[in] env execution environment
673 * \param[in] dev lu_device of OFD device
675 * \retval 0 if successful
676 * \retval negative value on error
678 static int ofd_recovery_complete(const struct lu_env *env,
679 struct lu_device *dev)
681 struct ofd_thread_info *oti = ofd_info(env);
682 struct ofd_device *ofd = ofd_dev(dev);
683 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
689 * Grant space for object precreation on the self export.
690 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
691 * is enough to create 10k objects. More space is then acquired for
692 * precreation in tgt_grant_create().
694 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
695 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
696 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
697 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
698 OBD_CONNECT_GRANT_PARAM;
699 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
701 rc = next->ld_ops->ldo_recovery_complete(env, next);
706 * lu_device_operations matrix for OFD device.
708 static struct lu_device_operations ofd_lu_ops = {
709 .ldo_object_alloc = ofd_object_alloc,
710 .ldo_process_config = ofd_process_config,
711 .ldo_recovery_complete = ofd_recovery_complete,
712 .ldo_prepare = ofd_prepare,
715 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
717 * Expose OSD statistics to OFD layer.
719 * The osd interfaces to the backend file system exposes useful data
720 * such as brw_stats and read or write cache states. This same data
721 * needs to be exposed into the obdfilter (ofd) layer to maintain
722 * backwards compatibility. This function creates the symlinks in the
723 * proc layer to enable this.
725 * \param[in] ofd OFD device
727 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
729 struct obd_device *obd = ofd_obd(ofd);
730 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
731 struct kobj_type *osd_type;
734 osd_type = get_ktype(&ofd->ofd_osd->dd_kobj);
735 for (i = 0; osd_type->default_attrs[i]; i++) {
736 if (strcmp(osd_type->default_attrs[i]->name,
737 "read_cache_enable") == 0) {
738 ofd->ofd_read_cache_enable =
739 osd_type->default_attrs[i];
742 if (strcmp(osd_type->default_attrs[i]->name,
743 "readcache_max_filesize") == 0) {
744 ofd->ofd_read_cache_max_filesize =
745 osd_type->default_attrs[i];
748 if (strcmp(osd_type->default_attrs[i]->name,
749 "writethrough_cache_enable") == 0) {
750 ofd->ofd_write_cache_enable =
751 osd_type->default_attrs[i];
755 if (obd->obd_proc_entry == NULL)
758 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
759 "../../%s/%s/brw_stats",
760 osd_obd->obd_type->typ_name, obd->obd_name);
765 * Cleanup all procfs entries in OFD.
767 * \param[in] ofd OFD device
769 static void ofd_procfs_fini(struct ofd_device *ofd)
771 struct obd_device *obd = ofd_obd(ofd);
773 tgt_tunables_fini(&ofd->ofd_lut);
774 lprocfs_free_per_client_stats(obd);
775 lprocfs_obd_cleanup(obd);
776 lprocfs_free_obd_stats(obd);
777 lprocfs_job_stats_fini(obd);
781 * Stop SEQ/FID server on OFD.
783 * \param[in] env execution environment
784 * \param[in] ofd OFD device
786 * \retval 0 if successful
787 * \retval negative value on error
789 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
791 return seq_site_fini(env, &ofd->ofd_seq_site);
795 * Start SEQ/FID server on OFD.
797 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
798 * It also connects to the master server to get own FID sequence (SEQ) range
799 * to this particular OFD. Typically that happens when the OST is first
800 * formatted or in the rare case that it exhausts the local sequence range.
802 * The sequence range is allocated out to the MDTs for OST object allocations,
803 * and not directly to the clients.
805 * \param[in] env execution environment
806 * \param[in] ofd OFD device
808 * \retval 0 if successful
809 * \retval negative value on error
811 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
813 struct seq_server_site *ss = &ofd->ofd_seq_site;
814 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
815 char *obd_name = ofd_name(ofd);
817 int len = strlen(obd_name) + 7;
820 ss = &ofd->ofd_seq_site;
821 lu->ld_site->ld_seq_site = ss;
822 ss->ss_lu = lu->ld_site;
823 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
825 OBD_ALLOC(name, len);
829 OBD_ALLOC_PTR(ss->ss_server_seq);
830 if (ss->ss_server_seq == NULL)
831 GOTO(out_name, rc = -ENOMEM);
833 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
834 LUSTRE_SEQ_SERVER, ss);
836 CERROR("%s: seq server init error: rc = %d\n", obd_name, rc);
837 GOTO(out_server, rc);
839 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
841 OBD_ALLOC_PTR(ss->ss_client_seq);
842 if (ss->ss_client_seq == NULL)
843 GOTO(out_server, rc = -ENOMEM);
845 snprintf(name, len, "%s-super", obd_name);
846 seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
849 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
852 seq_client_fini(ss->ss_client_seq);
853 OBD_FREE_PTR(ss->ss_client_seq);
854 ss->ss_client_seq = NULL;
856 seq_server_fini(ss->ss_server_seq, env);
857 OBD_FREE_PTR(ss->ss_server_seq);
858 ss->ss_server_seq = NULL;
867 * OFD request handler for OST_SET_INFO RPC.
869 * This is OFD-specific part of request handling
871 * \param[in] tsi target session environment for this request
873 * \retval 0 if successful
874 * \retval negative value on error
876 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
878 struct ptlrpc_request *req = tgt_ses_req(tsi);
879 struct ost_body *body = NULL, *repbody;
880 void *key, *val = NULL;
881 int keylen, vallen, rc = 0;
882 bool is_grant_shrink;
886 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
888 DEBUG_REQ(D_HA, req, "no set_info key");
889 RETURN(err_serious(-EFAULT));
891 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
894 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
896 DEBUG_REQ(D_HA, req, "no set_info val");
897 RETURN(err_serious(-EFAULT));
899 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
902 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
904 /* In this case the value is actually an RMF_OST_BODY, so we
905 * transmutate the type of this PTLRPC */
906 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
908 rc = req_capsule_server_pack(tsi->tsi_pill);
912 if (is_grant_shrink) {
913 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
915 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
918 /** handle grant shrink, similar to a read request */
919 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
921 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
923 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
926 CERROR("%s: Unsupported key %s\n",
927 tgt_name(tsi->tsi_tgt), (char *)key);
930 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
937 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
939 * This function returns a list of extents which describes how a file's
940 * blocks are laid out on the disk.
942 * \param[in] env execution environment
943 * \param[in] ofd OFD device
944 * \param[in] fid FID of object
945 * \param[in] fiemap fiemap structure to fill with data
947 * \retval 0 if \a fiemap is filled with data successfully
948 * \retval negative value on error
950 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
951 struct lu_fid *fid, struct fiemap *fiemap)
953 struct ofd_object *fo;
956 fo = ofd_object_find(env, ofd, fid);
958 CERROR("%s: error finding object "DFID"\n",
959 ofd_name(ofd), PFID(fid));
963 ofd_read_lock(env, fo);
964 if (ofd_object_exists(fo))
965 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
968 ofd_read_unlock(env, fo);
969 ofd_object_put(env, fo);
974 static int ofd_lock_unlock_region(const struct lu_env *env,
975 struct ldlm_namespace *ns,
976 struct ldlm_res_id *res_id,
977 unsigned long long begin,
978 unsigned long long end)
982 struct lustre_handle lh = { 0 };
984 LASSERT(begin <= end);
986 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
990 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
991 tgt_extent_unlock(&lh, LCK_PR);
997 * Lock the sparse areas of given resource.
999 * The locking of sparse areas will cause dirty data to be flushed back from
1000 * clients. This is used when getting the FIEMAP of an object to make sure
1001 * there is no unaccounted cached data on clients.
1003 * This function goes through \a fiemap list of extents and locks only sparse
1004 * areas between extents.
1006 * \param[in] ns LDLM namespace
1007 * \param[in] res_id resource ID
1008 * \param[in] fiemap file extents mapping on disk
1009 * \param[in] locked list head of regions list
1011 * \retval 0 if successful
1012 * \retval negative value on error
1014 static int lock_zero_regions(const struct lu_env *env,
1015 struct ldlm_namespace *ns,
1016 struct ldlm_res_id *res_id,
1017 struct fiemap *fiemap)
1019 __u64 begin = fiemap->fm_start;
1022 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1026 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1027 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1028 if (fiemap_start[i].fe_logical > begin) {
1029 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1030 begin, fiemap_start[i].fe_logical);
1031 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1032 fiemap_start[i].fe_logical);
1037 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1040 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1041 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1042 begin, fiemap->fm_start + fiemap->fm_length);
1043 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1044 fiemap->fm_start + fiemap->fm_length);
1052 * OFD request handler for OST_GET_INFO RPC.
1054 * This is OFD-specific part of request handling. The OFD-specific keys are:
1055 * - KEY_LAST_ID (obsolete)
1059 * This function reads needed data from storage and fills reply with it.
1061 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1062 * and is kept for compatibility.
1064 * \param[in] tsi target session environment for this request
1066 * \retval 0 if successful
1067 * \retval negative value on error
1069 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1071 struct obd_export *exp = tsi->tsi_exp;
1072 struct ofd_device *ofd = ofd_exp(exp);
1073 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1076 int replylen, rc = 0;
1080 /* this common part for get_info rpc */
1081 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1083 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1084 RETURN(err_serious(-EPROTO));
1086 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1089 if (KEY_IS(KEY_LAST_ID)) {
1091 struct ofd_seq *oseq;
1093 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1094 rc = req_capsule_server_pack(tsi->tsi_pill);
1096 RETURN(err_serious(rc));
1098 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1100 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1101 (u64)exp->exp_filter_data.fed_group);
1105 *last_id = ofd_seq_last_oid(oseq);
1106 ofd_seq_put(tsi->tsi_env, oseq);
1107 } else if (KEY_IS(KEY_FIEMAP)) {
1108 struct ll_fiemap_info_key *fm_key;
1109 struct fiemap *fiemap;
1112 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1114 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1115 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1117 RETURN(err_serious(rc));
1119 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1121 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1123 replylen = fiemap_count_to_size(
1124 fm_key->lfik_fiemap.fm_extent_count);
1125 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1126 RCL_SERVER, replylen);
1128 rc = req_capsule_server_pack(tsi->tsi_pill);
1130 RETURN(err_serious(rc));
1132 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1136 *fiemap = fm_key->lfik_fiemap;
1137 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1139 /* LU-3219: Lock the sparse areas to make sure dirty
1140 * flushed back from client, then call fiemap again. */
1141 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1142 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1143 ost_fid_build_resid(fid, &fti->fti_resid);
1144 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1145 &fti->fti_resid, fiemap);
1147 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1150 } else if (KEY_IS(KEY_LAST_FID)) {
1151 struct ofd_device *ofd = ofd_exp(exp);
1152 struct ofd_seq *oseq;
1156 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1157 rc = req_capsule_server_pack(tsi->tsi_pill);
1159 RETURN(err_serious(rc));
1161 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1163 RETURN(err_serious(-EPROTO));
1165 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1167 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1171 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1172 ostid_seq(&fti->fti_ostid));
1174 RETURN(PTR_ERR(oseq));
1176 rc = ostid_to_fid(fid, &oseq->os_oi,
1177 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1181 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1184 ofd_seq_put(tsi->tsi_env, oseq);
1186 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1190 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1197 * OFD request handler for OST_GETATTR RPC.
1199 * This is OFD-specific part of request handling. It finds the OFD object
1200 * by its FID, gets attributes from storage and packs result to the reply.
1202 * \param[in] tsi target session environment for this request
1204 * \retval 0 if successful
1205 * \retval negative value on error
1207 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1209 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1210 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1211 struct ost_body *repbody;
1212 struct lustre_handle lh = { 0 };
1213 struct ofd_object *fo;
1215 enum ldlm_mode lock_mode = LCK_PR;
1220 LASSERT(tsi->tsi_ost_body != NULL);
1222 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1223 if (repbody == NULL)
1226 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1227 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1229 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1230 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1233 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1236 rc = tgt_extent_lock(tsi->tsi_env,
1237 tsi->tsi_tgt->lut_obd->obd_namespace,
1238 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1244 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1246 GOTO(out, rc = PTR_ERR(fo));
1248 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1252 obdo_from_la(&repbody->oa, &fti->fti_attr,
1253 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1255 /* Store object version in reply */
1256 curr_version = dt_version_get(tsi->tsi_env,
1257 ofd_object_child(fo));
1258 if ((__s64)curr_version != -EOPNOTSUPP) {
1259 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1260 repbody->oa.o_data_version = curr_version;
1263 if (fo->ofo_ff.ff_layout_version > 0) {
1264 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1265 repbody->oa.o_layout_version =
1266 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1268 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1269 PFID(&tsi->tsi_fid),
1270 repbody->oa.o_layout_version);
1274 ofd_object_put(tsi->tsi_env, fo);
1277 tgt_extent_unlock(&lh, lock_mode);
1279 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1282 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1283 repbody->oa.o_flags = OBD_FL_FLUSH;
1289 * OFD request handler for OST_SETATTR RPC.
1291 * This is OFD-specific part of request handling. It finds the OFD object
1292 * by its FID, sets attributes from request and packs result to the reply.
1294 * \param[in] tsi target session environment for this request
1296 * \retval 0 if successful
1297 * \retval negative value on error
1299 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1301 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1302 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1303 struct ost_body *body = tsi->tsi_ost_body;
1304 struct ost_body *repbody;
1305 struct ldlm_resource *res;
1306 struct ofd_object *fo;
1311 LASSERT(body != NULL);
1313 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1314 if (repbody == NULL)
1317 repbody->oa.o_oi = body->oa.o_oi;
1318 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1320 /* This would be very bad - accidentally truncating a file when
1321 * changing the time or similar - bug 12203. */
1322 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1323 body->oa.o_size != OBD_OBJECT_EOF) {
1324 static char mdsinum[48];
1326 if (body->oa.o_valid & OBD_MD_FLFID)
1327 snprintf(mdsinum, sizeof(mdsinum) - 1,
1328 "of parent "DFID, body->oa.o_parent_seq,
1329 body->oa.o_parent_oid, 0);
1333 CERROR("%s: setattr from %s is trying to truncate object "DFID
1334 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1335 PFID(&tsi->tsi_fid), mdsinum);
1339 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1341 GOTO(out, rc = PTR_ERR(fo));
1343 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1344 fti->fti_attr.la_valid &= ~LA_TYPE;
1346 /* setting objects attributes (including owner/group) */
1347 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1351 obdo_from_la(&repbody->oa, &fti->fti_attr,
1352 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1354 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1358 ofd_object_put(tsi->tsi_env, fo);
1361 /* we do not call this before to avoid lu_object_find() in
1362 * ->lvbo_update() holding another reference on the object.
1363 * otherwise concurrent destroy can make the object unavailable
1364 * for 2nd lu_object_find() waiting for the first reference
1365 * to go... deadlock! */
1366 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1367 &tsi->tsi_resid, LDLM_EXTENT, 0);
1369 ldlm_res_lvbo_update(res, NULL, 0);
1370 ldlm_resource_putref(res);
1377 * Destroy OST orphans.
1379 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1380 * set then we must destroy possible orphaned objects.
1382 * \param[in] env execution environment
1383 * \param[in] exp OBD export
1384 * \param[in] ofd OFD device
1385 * \param[in] oa obdo structure for reply
1387 * \retval 0 if successful
1388 * \retval negative value on error
1390 static int ofd_orphans_destroy(const struct lu_env *env,
1391 struct obd_export *exp,
1392 struct ofd_device *ofd, struct obdo *oa)
1394 struct ofd_thread_info *info = ofd_info(env);
1395 struct lu_fid *fid = &info->fti_fid;
1396 struct ost_id *oi = &oa->o_oi;
1397 struct ofd_seq *oseq;
1398 u64 seq = ostid_seq(oi);
1399 u64 end_id = ostid_id(oi);
1407 oseq = ofd_seq_get(ofd, seq);
1409 CERROR("%s: Can not find seq for "DOSTID"\n",
1410 ofd_name(ofd), POSTID(oi));
1415 last = ofd_seq_last_oid(oseq);
1418 LASSERT(exp != NULL);
1419 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1421 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1424 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1425 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1427 while (oid > end_id) {
1428 rc = fid_set_id(fid, oid);
1429 if (unlikely(rc != 0))
1432 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1433 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1434 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1435 /* this is pretty fatal... */
1436 CEMERG("%s: error destroying precreated id "
1438 ofd_name(ofd), PFID(fid), rc);
1442 ofd_seq_last_oid_set(oseq, oid);
1443 /* update last_id on disk periodically so that if we
1444 * restart * we don't need to re-scan all of the just
1445 * deleted objects. */
1446 if ((oid & 511) == 0)
1447 ofd_seq_last_oid_write(env, ofd, oseq);
1451 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1452 ofd_name(ofd), seq, oid);
1456 ofd_seq_last_oid_set(oseq, oid);
1457 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1459 /* don't reuse orphan object, return last used objid */
1460 rc = ostid_set_id(oi, last);
1466 ofd_seq_put(env, oseq);
1471 * OFD request handler for OST_CREATE RPC.
1473 * This is OFD-specific part of request handling. Its main purpose is to
1474 * create new data objects on OST, but it also used to destroy orphans.
1476 * \param[in] tsi target session environment for this request
1478 * \retval 0 if successful
1479 * \retval negative value on error
1481 static int ofd_create_hdl(struct tgt_session_info *tsi)
1483 struct ptlrpc_request *req = tgt_ses_req(tsi);
1484 struct ost_body *repbody;
1485 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1486 struct obdo *rep_oa;
1487 struct obd_export *exp = tsi->tsi_exp;
1488 struct ofd_device *ofd = ofd_exp(exp);
1489 u64 seq = ostid_seq(&oa->o_oi);
1490 u64 oid = ostid_id(&oa->o_oi);
1491 struct ofd_seq *oseq;
1499 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1502 if (ofd->ofd_no_precreate)
1505 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1506 if (repbody == NULL)
1509 down_read(&ofd->ofd_lastid_rwsem);
1510 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1511 * we may do that in the future.
1512 * Return -ENOSPC until the LAST_ID rebuilt. */
1513 if (unlikely(ofd->ofd_lastid_rebuilding))
1514 GOTO(out_sem, rc = -ENOSPC);
1516 rep_oa = &repbody->oa;
1517 rep_oa->o_oi = oa->o_oi;
1519 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1521 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1523 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1525 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1526 ofd_name(ofd), seq, PTR_ERR(oseq));
1527 GOTO(out_sem, rc = -EINVAL);
1530 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1531 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1532 if (!ofd_obd(ofd)->obd_recovering ||
1533 oid > ofd_seq_last_oid(oseq)) {
1534 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1535 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1536 ofd_seq_last_oid(oseq));
1537 GOTO(out_nolock, rc = -EINVAL);
1539 /* Do nothing here, we re-create objects during recovery
1540 * upon write replay, see ofd_preprw_write() */
1541 GOTO(out_nolock, rc = 0);
1543 /* former ofd_handle_precreate */
1544 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1545 (oa->o_flags & OBD_FL_DELORPHAN)) {
1546 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1548 /* destroy orphans */
1549 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1550 exp->exp_conn_cnt) {
1551 CERROR("%s: dropping old orphan cleanup request\n",
1553 GOTO(out_nolock, rc = 0);
1555 /* This causes inflight precreates to abort and drop lock */
1556 oseq->os_destroys_in_progress = 1;
1557 mutex_lock(&oseq->os_create_lock);
1558 if (!oseq->os_destroys_in_progress) {
1560 "%s:[%llu] destroys_in_progress already cleared\n",
1561 ofd_name(ofd), seq);
1562 rc = ostid_set_id(&rep_oa->o_oi,
1563 ofd_seq_last_oid(oseq));
1566 diff = oid - ofd_seq_last_oid(oseq);
1567 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1568 ofd_seq_last_oid(oseq), diff);
1569 if (-diff > OST_MAX_PRECREATE) {
1570 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1571 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1572 "(%llu), trust the OST\n",
1573 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1574 PFID(&oseq->os_oi.oi_fid),
1575 ofd_seq_last_oid(oseq));
1577 /* Let MDS know that we are so far ahead. */
1578 rc = ostid_set_id(&rep_oa->o_oi,
1579 ofd_seq_last_oid(oseq) + 1);
1580 } else if (diff < 0) {
1581 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1583 oseq->os_destroys_in_progress = 0;
1585 /* XXX: Used by MDS for the first time! */
1586 oseq->os_destroys_in_progress = 0;
1589 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1590 ofd->ofd_lastid_gen)) {
1591 /* Keep the export ref so we can send the reply. */
1592 ofd_obd_disconnect(class_export_get(exp));
1593 GOTO(out_nolock, rc = -ENOTCONN);
1596 mutex_lock(&oseq->os_create_lock);
1597 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1598 exp->exp_conn_cnt) {
1599 CERROR("%s: dropping old precreate request\n",
1603 /* only precreate if seq is 0, IDIF or normal and also o_id
1604 * must be specfied */
1605 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1606 !fid_seq_is_idif(seq)) || oid == 0) {
1607 diff = 1; /* shouldn't we create this right now? */
1609 diff = oid - ofd_seq_last_oid(oseq);
1610 /* Do sync create if the seq is about to used up */
1611 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1612 if (unlikely(oid >= IDIF_MAX_OID - 1))
1614 } else if (fid_seq_is_norm(seq)) {
1616 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1619 CERROR("%s : invalid o_seq "DOSTID"\n",
1620 ofd_name(ofd), POSTID(&oa->o_oi));
1621 GOTO(out, rc = -EINVAL);
1626 CERROR("%s: invalid precreate request for "
1627 DOSTID", last_id %llu. "
1628 "Likely MDS last_id corruption\n",
1629 ofd_name(ofd), POSTID(&oa->o_oi),
1630 ofd_seq_last_oid(oseq));
1631 GOTO(out, rc = -EINVAL);
1636 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1642 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1643 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1644 /* don't enforce grant during orphan recovery */
1645 granted = tgt_grant_create(tsi->tsi_env,
1646 ofd_obd(ofd)->obd_self_export,
1651 CDEBUG(D_HA, "%s: failed to acquire grant "
1652 "space for precreate (%lld): rc = %d\n",
1653 ofd_name(ofd), diff, rc);
1658 /* This can happen if a new OST is formatted and installed
1659 * in place of an old one at the same index. Instead of
1660 * precreating potentially millions of deleted old objects
1661 * (possibly filling the OST), only precreate the last batch.
1662 * LFSCK will eventually clean up any orphans. LU-14 */
1663 if (diff > 5 * OST_MAX_PRECREATE) {
1664 diff = OST_MAX_PRECREATE / 2;
1665 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1666 "OST replaced or reformatted: "
1667 "LFSCK will clean up",
1670 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1671 "%u larger than the LAST_ID "DOSTID", only "
1672 "precreating the last %lld objects.\n",
1673 ofd_name(ofd), POSTID(&oa->o_oi),
1674 5 * OST_MAX_PRECREATE,
1675 POSTID(&oseq->os_oi), diff);
1676 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1680 next_id = ofd_seq_last_oid(oseq) + 1;
1681 count = ofd_precreate_batch(ofd, (int)diff);
1683 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1684 " at %llu\n", ofd_name(ofd),
1685 count, seq, next_id);
1687 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1688 && ktime_get_seconds() > enough_time) {
1689 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1690 " created at a rate of %d/s\n",
1691 ofd_name(ofd), created, diff + created,
1692 created / DISK_TIMEOUT);
1696 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1697 oseq, count, sync_trans);
1701 } else if (rc < 0) {
1707 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1708 LCONSOLE_WARN("%s: can't create the same count of"
1709 " objects when replaying the request"
1710 " (diff is %lld). see LU-4621\n",
1711 ofd_name(ofd), diff);
1714 /* some objects got created, we can return
1715 * them, even if last creation failed */
1718 CERROR("%s: unable to precreate: rc = %d\n",
1721 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1722 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1723 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1728 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1732 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1734 if (unlikely(!oseq->os_last_id_synced))
1735 oseq->os_last_id_synced = 1;
1737 mutex_unlock(&oseq->os_create_lock);
1740 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1741 ofd_seq_put(tsi->tsi_env, oseq);
1744 up_read(&ofd->ofd_lastid_rwsem);
1749 * OFD request handler for OST_DESTROY RPC.
1751 * This is OFD-specific part of request handling. It destroys data objects
1752 * related to destroyed object on MDT.
1754 * \param[in] tsi target session environment for this request
1756 * \retval 0 if successful
1757 * \retval negative value on error
1759 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1761 const struct ost_body *body = tsi->tsi_ost_body;
1762 struct ost_body *repbody;
1763 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1764 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1765 struct lu_fid *fid = &fti->fti_fid;
1772 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1775 /* This is old case for clients before Lustre 2.4 */
1776 /* If there's a DLM request, cancel the locks mentioned in it */
1777 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1779 struct ldlm_request *dlm;
1781 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1784 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1787 *fid = body->oa.o_oi.oi_fid;
1788 oid = ostid_id(&body->oa.o_oi);
1791 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1793 /* check that o_misc makes sense */
1794 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1795 count = body->oa.o_misc;
1797 count = 1; /* default case - single destroy */
1799 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1800 POSTID(&body->oa.o_oi), count);
1805 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1806 if (lrc == -ENOENT) {
1808 "%s: destroying non-existent object "DFID"\n",
1809 ofd_name(ofd), PFID(fid));
1810 /* rewrite rc with -ENOENT only if it is 0 */
1813 } else if (lrc != 0) {
1814 CERROR("%s: error destroying object "DFID": %d\n",
1815 ofd_name(ofd), PFID(fid), lrc);
1821 lrc = fid_set_id(fid, oid);
1822 if (unlikely(lrc != 0 && count > 0))
1823 GOTO(out, rc = lrc);
1826 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1832 fid_to_ostid(fid, &repbody->oa.o_oi);
1837 * OFD request handler for OST_STATFS RPC.
1839 * This function gets statfs data from storage as part of request
1842 * \param[in] tsi target session environment for this request
1844 * \retval 0 if successful
1845 * \retval negative value on error
1847 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1849 struct obd_statfs *osfs;
1854 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1856 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1858 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1859 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1861 CERROR("%s: statfs failed: rc = %d\n",
1862 tgt_name(tsi->tsi_tgt), rc);
1864 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1867 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1874 * OFD request handler for OST_SYNC RPC.
1876 * Sync object data or all filesystem data to the disk and pack the
1879 * \param[in] tsi target session environment for this request
1881 * \retval 0 if successful
1882 * \retval negative value on error
1884 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1886 struct ost_body *body = tsi->tsi_ost_body;
1887 struct ost_body *repbody;
1888 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1889 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1890 struct ofd_object *fo = NULL;
1895 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1897 /* if no objid is specified, it means "sync whole filesystem" */
1898 if (!fid_is_zero(&tsi->tsi_fid)) {
1899 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1901 RETURN(PTR_ERR(fo));
1904 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1905 fo != NULL ? ofd_object_child(fo) : NULL,
1906 repbody->oa.o_size, repbody->oa.o_blocks);
1910 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1915 repbody->oa.o_oi = body->oa.o_oi;
1916 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1918 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1920 obdo_from_la(&repbody->oa, &fti->fti_attr,
1923 /* don't return rc from getattr */
1928 ofd_object_put(tsi->tsi_env, fo);
1933 * OFD request handler for OST_FALLOCATE RPC.
1935 * This is part of request processing. Validate request fields,
1936 * preallocate the given OFD object and pack reply.
1938 * \param[in] tsi target session environment for this request
1940 * \retval 0 if successful
1941 * \retval negative value on error
1943 static int ofd_fallocate_hdl(struct tgt_session_info *tsi)
1945 struct obdo *oa = &tsi->tsi_ost_body->oa;
1946 struct ost_body *repbody;
1947 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1948 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1949 struct ldlm_resource *res;
1950 struct ofd_object *fo;
1952 struct lustre_handle lh = { 0, };
1957 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1958 if (repbody == NULL)
1959 RETURN(err_serious(-ENOMEM));
1962 * fallocate start and end are passed in o_size, o_blocks
1967 mode = oa->o_falloc_mode;
1969 * Only mode == 0 (which is standard prealloc) is supported now.
1970 * Punch is not supported yet.
1972 if (mode & ~FALLOC_FL_KEEP_SIZE)
1973 RETURN(-EOPNOTSUPP);
1975 repbody->oa.o_oi = oa->o_oi;
1976 repbody->oa.o_valid = OBD_MD_FLID;
1978 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1979 oa->o_flags & OBD_FL_SRVLOCK;
1982 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid,
1983 start, end, &lh, LCK_PW, &flags);
1988 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
1991 GOTO(out, rc = PTR_ERR(fo));
1993 la_from_obdo(&info->fti_attr, oa,
1994 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
1996 rc = ofd_object_fallocate(tsi->tsi_env, fo, start, end, mode,
1997 &info->fti_attr, oa);
2001 rc = ofd_attr_get(tsi->tsi_env, fo, &info->fti_attr);
2003 obdo_from_la(&repbody->oa, &info->fti_attr,
2008 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PREALLOC,
2013 ofd_object_put(tsi->tsi_env, fo);
2016 tgt_extent_unlock(&lh, LCK_PW);
2018 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2021 struct ost_lvb *res_lvb;
2023 ldlm_res_lvbo_update(res, NULL, 0);
2024 res_lvb = res->lr_lvb_data;
2026 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2027 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2029 repbody->oa.o_valid |= OBD_MD_FLSIZE;
2030 repbody->oa.o_size = res_lvb->lvb_size;
2032 ldlm_resource_putref(res);
2041 * OFD request handler for OST_PUNCH RPC.
2043 * This is part of request processing. Validate request fields,
2044 * punch (truncate) the given OFD object and pack reply.
2046 * \param[in] tsi target session environment for this request
2048 * \retval 0 if successful
2049 * \retval negative value on error
2051 static int ofd_punch_hdl(struct tgt_session_info *tsi)
2053 const struct obdo *oa = &tsi->tsi_ost_body->oa;
2054 struct ost_body *repbody;
2055 struct ofd_thread_info *info = tsi2ofd_info(tsi);
2056 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
2057 struct ldlm_resource *res;
2058 struct ofd_object *fo;
2060 struct lustre_handle lh = { 0, };
2067 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
2069 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
2070 BUILD_BUG_ON(!(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK));
2072 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2073 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2074 RETURN(err_serious(-EPROTO));
2076 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2077 if (repbody == NULL)
2078 RETURN(err_serious(-ENOMEM));
2080 /* punch start,end are passed in o_size,o_blocks throught wire */
2084 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2087 /* standard truncate optimization: if file body is completely
2088 * destroyed, don't send data back to the server. */
2090 flags |= LDLM_FL_AST_DISCARD_DATA;
2092 repbody->oa.o_oi = oa->o_oi;
2093 repbody->oa.o_valid = OBD_MD_FLID;
2095 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2096 oa->o_flags & OBD_FL_SRVLOCK;
2099 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
2100 end, &lh, LCK_PW, &flags);
2105 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2106 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2107 oa->o_valid, start, end);
2109 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2112 GOTO(out, rc = PTR_ERR(fo));
2114 la_from_obdo(&info->fti_attr, oa,
2115 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2116 info->fti_attr.la_size = start;
2117 info->fti_attr.la_valid |= LA_SIZE;
2119 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2124 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2128 ofd_object_put(tsi->tsi_env, fo);
2131 tgt_extent_unlock(&lh, LCK_PW);
2133 /* we do not call this before to avoid lu_object_find() in
2134 * ->lvbo_update() holding another reference on the object.
2135 * otherwise concurrent destroy can make the object unavailable
2136 * for 2nd lu_object_find() waiting for the first reference
2137 * to go... deadlock! */
2138 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2141 struct ost_lvb *res_lvb;
2143 ldlm_res_lvbo_update(res, NULL, 0);
2144 res_lvb = res->lr_lvb_data;
2145 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2146 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2147 ldlm_resource_putref(res);
2153 static int ofd_ladvise_prefetch(const struct lu_env *env,
2154 struct ofd_object *fo,
2155 struct niobuf_local *lnb,
2156 __u64 start, __u64 end, enum dt_bufs_type dbt)
2158 struct ofd_thread_info *info = ofd_info(env);
2159 pgoff_t start_index, end_index, pages;
2160 struct niobuf_remote rnb;
2161 unsigned long nr_local;
2167 ofd_read_lock(env, fo);
2168 if (!ofd_object_exists(fo))
2169 GOTO(out_unlock, rc = -ENOENT);
2171 rc = ofd_attr_get(env, fo, &info->fti_attr);
2173 GOTO(out_unlock, rc);
2175 if (end > info->fti_attr.la_size)
2176 end = info->fti_attr.la_size;
2179 GOTO(out_unlock, rc);
2181 /* We need page aligned offset and length */
2182 start_index = start >> PAGE_SHIFT;
2183 end_index = (end - 1) >> PAGE_SHIFT;
2184 pages = end_index - start_index + 1;
2186 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2187 PTLRPC_MAX_BRW_PAGES;
2188 rnb.rnb_offset = start_index << PAGE_SHIFT;
2189 rnb.rnb_len = nr_local << PAGE_SHIFT;
2190 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb,
2191 PTLRPC_MAX_BRW_PAGES, dbt);
2192 if (unlikely(rc < 0))
2195 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2196 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2199 start_index += nr_local;
2204 ofd_read_unlock(env, fo);
2209 * OFD request handler for OST_LADVISE RPC.
2211 * Tune cache or perfetch policies according to advices.
2213 * \param[in] tsi target session environment for this request
2215 * \retval 0 if successful
2216 * \retval negative errno on error
2218 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2220 struct ptlrpc_request *req = tgt_ses_req(tsi);
2221 struct obd_export *exp = tsi->tsi_exp;
2222 struct ofd_device *ofd = ofd_exp(exp);
2223 struct ost_body *body, *repbody;
2224 struct ofd_thread_info *info;
2225 struct ofd_object *fo;
2226 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2227 const struct lu_env *env = svc_thread->t_env;
2228 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2229 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2230 struct lu_ladvise *ladvise;
2232 struct ladvise_hdr *ladvise_hdr;
2233 struct obd_ioobj ioo;
2234 struct lustre_handle lockh = { 0 };
2237 struct dt_object *dob;
2243 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2244 body = tsi->tsi_ost_body;
2246 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2247 RETURN(err_serious(-EPROTO));
2249 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2250 &RMF_OST_LADVISE_HDR);
2251 if (ladvise_hdr == NULL)
2252 RETURN(err_serious(-EPROTO));
2254 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2255 ladvise_hdr->lah_count < 1)
2256 RETURN(err_serious(-EPROTO));
2258 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2259 RETURN(err_serious(-EPROTO));
2261 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2262 if (ladvise == NULL)
2263 RETURN(err_serious(-EPROTO));
2265 num_advise = req_capsule_get_size(&req->rq_pill,
2266 &RMF_OST_LADVISE, RCL_CLIENT) /
2268 if (num_advise < ladvise_hdr->lah_count)
2269 RETURN(err_serious(-EPROTO));
2271 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2272 repbody->oa = body->oa;
2274 info = ofd_info_init(env, exp);
2276 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2277 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2281 fo = ofd_object_find(env, ofd, &info->fti_fid);
2286 LASSERT(fo != NULL);
2287 dob = ofd_object_child(fo);
2289 if (ptlrpc_connection_is_local(exp->exp_connection))
2290 dbt |= DT_BUFS_TYPE_LOCAL;
2292 for (i = 0; i < num_advise; i++, ladvise++) {
2293 start = ladvise->lla_start;
2294 end = ladvise->lla_end;
2296 rc = err_serious(-EPROTO);
2300 /* Handle different advice types */
2301 switch (ladvise->lla_advice) {
2305 case LU_LADVISE_WILLREAD:
2309 ioo.ioo_oid = body->oa.o_oi;
2311 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2312 &tsi->tsi_resid, start, end - 1,
2313 &lockh, LCK_PR, &flags);
2317 req->rq_status = ofd_ladvise_prefetch(env, fo,
2320 tgt_extent_unlock(&lockh, LCK_PR);
2322 case LU_LADVISE_DONTNEED:
2323 rc = dt_ladvise(env, dob, ladvise->lla_start,
2324 ladvise->lla_end, LU_LADVISE_DONTNEED);
2331 ofd_object_put(env, fo);
2332 req->rq_status = rc;
2337 * OFD request handler for OST_QUOTACTL RPC.
2339 * This is part of request processing to validate incoming request fields,
2340 * get the requested data from OSD and pack reply.
2342 * \param[in] tsi target session environment for this request
2344 * \retval 0 if successful
2345 * \retval negative value on error
2347 static int ofd_quotactl(struct tgt_session_info *tsi)
2349 struct obd_quotactl *oqctl, *repoqc;
2350 struct lu_nodemap *nodemap;
2356 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2358 RETURN(err_serious(-EPROTO));
2360 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2362 RETURN(err_serious(-ENOMEM));
2366 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2367 if (IS_ERR(nodemap))
2368 RETURN(PTR_ERR(nodemap));
2371 if (oqctl->qc_type == USRQUOTA)
2372 id = nodemap_map_id(nodemap, NODEMAP_UID,
2373 NODEMAP_CLIENT_TO_FS,
2375 else if (oqctl->qc_type == GRPQUOTA)
2376 id = nodemap_map_id(nodemap, NODEMAP_GID,
2377 NODEMAP_CLIENT_TO_FS,
2380 nodemap_putref(nodemap);
2382 if (repoqc->qc_id != id)
2383 swap(repoqc->qc_id, id);
2385 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2387 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2390 if (repoqc->qc_id != id)
2391 swap(repoqc->qc_id, id);
2397 * Prolong lock timeout for the given extent.
2399 * This function finds all locks related with incoming request and
2400 * prolongs their timeout.
2402 * If a client is holding a lock for a long time while it sends
2403 * read or write RPCs to the OST for the object under this lock,
2404 * then we don't want the OST to evict the client. Otherwise,
2405 * if the network or disk is very busy then the client may not
2406 * be able to make any progress to clear out dirty pages under
2407 * the lock and the application will fail.
2409 * Every time a Bulk Read/Write (BRW) request arrives for the object
2410 * covered by the lock, extend the timeout on that lock. The RPC should
2411 * contain a lock handle for the lock it is using, but this
2412 * isn't handled correctly by all client versions, and the
2413 * request may cover multiple locks.
2415 * \param[in] tsi target session environment for this request
2416 * \param[in] data struct of data to prolong locks
2419 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2420 struct ldlm_prolong_args *data)
2422 struct obdo *oa = &tsi->tsi_ost_body->oa;
2423 struct ldlm_lock *lock;
2427 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2428 data->lpa_export = tsi->tsi_exp;
2429 data->lpa_resid = tsi->tsi_resid;
2431 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2432 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2433 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2434 data->lpa_extent.end);
2436 if (oa->o_valid & OBD_MD_FLHANDLE) {
2437 /* mostly a request should be covered by only one lock, try
2439 lock = ldlm_handle2lock(&oa->o_handle);
2441 /* Fast path to check if the lock covers the whole IO
2442 * region exclusively. */
2443 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2444 &data->lpa_extent)) {
2446 LASSERT(lock->l_export == data->lpa_export);
2447 ldlm_lock_prolong_one(lock, data);
2448 LDLM_LOCK_PUT(lock);
2449 if (data->lpa_locks_cnt > 0)
2451 /* The lock was destroyed probably lets try
2454 lock->l_last_used = ktime_get();
2455 LDLM_LOCK_PUT(lock);
2460 ldlm_resource_prolong(data);
2465 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2467 * Determine if \a lock and the lock from request \a req are equivalent
2468 * by comparing their resource names, modes, and extents.
2470 * It is used to give priority to read and write RPCs being done
2471 * under this lock so that the client can drop the contended
2472 * lock more quickly and let other clients use it. This improves
2473 * overall performance in the case where the first client gets a
2474 * very large lock extent that prevents other clients from
2475 * submitting their writes.
2477 * \param[in] req ptlrpc_request being processed
2478 * \param[in] lock contended lock to match
2480 * \retval 1 if lock is matched
2481 * \retval 0 otherwise
2483 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2484 struct ldlm_lock *lock)
2486 struct niobuf_remote *rnb;
2487 struct obd_ioobj *ioo;
2488 enum ldlm_mode mode;
2489 struct ldlm_extent ext;
2490 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2494 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2495 LASSERT(ioo != NULL);
2497 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2498 LASSERT(rnb != NULL);
2500 ext.start = rnb->rnb_offset;
2501 rnb += ioo->ioo_bufcnt - 1;
2502 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2504 LASSERT(lock->l_resource != NULL);
2505 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2508 /* a bulk write can only hold a reference on a PW extent lock
2511 mode = LCK_PW | LCK_GROUP;
2512 if (opc == OST_READ)
2513 /* whereas a bulk read can be protected by either a PR or PW
2517 if (!(lock->l_granted_mode & mode))
2520 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2524 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2526 * Check for whether the given PTLRPC request (\a req) is blocking
2527 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2530 * \param[in] req the incoming request
2532 * \retval 1 if \a req is blocking an LDLM lock cancel
2533 * \retval 0 if it is not
2534 * \retval -ESTALE if lock is not found
2536 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2538 struct tgt_session_info *tsi;
2539 struct obd_ioobj *ioo;
2540 struct niobuf_remote *rnb;
2542 struct ldlm_prolong_args pa = { 0 };
2546 /* Don't use tgt_ses_info() to get session info, because lock_match()
2547 * can be called while request has no processing thread yet. */
2548 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2551 * Use LASSERT below because malformed RPCs should have
2552 * been filtered out in tgt_hpreq_handler().
2554 opc = lustre_msg_get_opc(req->rq_reqmsg);
2555 LASSERT(opc == OST_READ || opc == OST_WRITE);
2557 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2558 LASSERT(ioo != NULL);
2560 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2561 LASSERT(rnb != NULL);
2562 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2564 pa.lpa_mode = LCK_PW | LCK_GROUP;
2565 if (opc == OST_READ)
2566 pa.lpa_mode |= LCK_PR;
2568 pa.lpa_extent.start = rnb->rnb_offset;
2569 rnb += ioo->ioo_bufcnt - 1;
2570 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2572 DEBUG_REQ(D_RPCTRACE, req,
2573 "%s %s: refresh rw locks for "DFID" (%llu->%llu)",
2574 tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid),
2575 pa.lpa_extent.start, pa.lpa_extent.end);
2577 ofd_prolong_extent_locks(tsi, &pa);
2579 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n",
2580 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2582 if (pa.lpa_blocks_cnt > 0)
2585 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2589 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2591 * Called after the request has been handled. It refreshes lock timeout again
2592 * so that client has more time to send lock cancel RPC.
2594 * \param[in] req request which is being processed.
2596 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2598 ofd_rw_hpreq_check(req);
2602 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2604 * This function checks if the given lock is the same by its resname, mode
2605 * and extent as one taken from the request.
2606 * It is used to give priority to punch/truncate RPCs that might lead to
2607 * the fastest release of that lock when a lock is contended.
2609 * \param[in] req ptlrpc_request being processed
2610 * \param[in] lock contended lock to match
2612 * \retval 1 if lock is matched
2613 * \retval 0 otherwise
2615 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2616 struct ldlm_lock *lock)
2618 struct tgt_session_info *tsi;
2620 struct ldlm_extent ext;
2624 /* Don't use tgt_ses_info() to get session info, because lock_match()
2625 * can be called while request has no processing thread yet. */
2626 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2629 * Use LASSERT below because malformed RPCs should have
2630 * been filtered out in tgt_hpreq_handler().
2632 LASSERT(tsi->tsi_ost_body != NULL);
2633 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2634 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2637 oa = &tsi->tsi_ost_body->oa;
2638 ext.start = oa->o_size;
2639 ext.end = oa->o_blocks;
2641 LASSERT(lock->l_resource != NULL);
2642 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2645 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2648 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2652 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2654 * High-priority queue request check for whether the given punch request
2655 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2656 * covered by an LDLM lock.
2660 * \param[in] req the incoming request
2662 * \retval 1 if \a req is blocking an LDLM lock cancel
2663 * \retval 0 if it is not
2664 * \retval -ESTALE if lock is not found
2666 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2668 struct tgt_session_info *tsi;
2670 struct ldlm_prolong_args pa = { 0 };
2674 /* Don't use tgt_ses_info() to get session info, because lock_match()
2675 * can be called while request has no processing thread yet. */
2676 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2677 LASSERT(tsi != NULL);
2678 oa = &tsi->tsi_ost_body->oa;
2680 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2681 oa->o_flags & OBD_FL_SRVLOCK));
2683 pa.lpa_mode = LCK_PW | LCK_GROUP;
2684 pa.lpa_extent.start = oa->o_size;
2685 pa.lpa_extent.end = oa->o_blocks;
2688 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2689 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2690 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2692 ofd_prolong_extent_locks(tsi, &pa);
2694 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2695 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2697 if (pa.lpa_blocks_cnt > 0)
2700 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2704 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2706 * Called after the request has been handled. It refreshes lock timeout again
2707 * so that client has more time to send lock cancel RPC.
2709 * \param[in] req request which is being processed.
2711 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2713 ofd_punch_hpreq_check(req);
2716 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2717 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2718 .hpreq_check = ofd_rw_hpreq_check,
2719 .hpreq_fini = ofd_rw_hpreq_fini
2722 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2723 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2724 .hpreq_check = ofd_punch_hpreq_check,
2725 .hpreq_fini = ofd_punch_hpreq_fini
2729 * Assign high priority operations to an IO request.
2731 * Check if the incoming request is a candidate for
2732 * high-priority processing. If it is, assign it a high
2733 * priority operations table.
2735 * \param[in] tsi target session environment for this request
2737 static void ofd_hp_brw(struct tgt_session_info *tsi)
2739 struct niobuf_remote *rnb;
2740 struct obd_ioobj *ioo;
2744 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2745 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2746 if (ioo->ioo_bufcnt > 0) {
2747 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2748 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2750 /* no high priority if server lock is needed */
2751 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2752 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2756 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2760 * Assign high priority operations to an punch request.
2762 * Check if the incoming request is a candidate for
2763 * high-priority processing. If it is, assign it a high
2764 * priority operations table.
2766 * \param[in] tsi target session environment for this request
2768 static void ofd_hp_punch(struct tgt_session_info *tsi)
2770 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2771 /* no high-priority if server lock is needed */
2772 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2773 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2774 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2775 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2777 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2780 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2781 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2782 #define OST_BRW_READ OST_READ
2783 #define OST_BRW_WRITE OST_WRITE
2786 * Table of OFD-specific request handlers
2788 * This table contains all opcodes accepted by OFD and
2789 * specifies handlers for them. The tgt_request_handler()
2790 * uses such table from each target to process incoming
2793 static struct tgt_handler ofd_tgt_handlers[] = {
2794 TGT_RPC_HANDLER(OST_FIRST_OPC,
2795 0, OST_CONNECT, tgt_connect,
2796 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2797 TGT_RPC_HANDLER(OST_FIRST_OPC,
2798 0, OST_DISCONNECT, tgt_disconnect,
2799 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2800 TGT_RPC_HANDLER(OST_FIRST_OPC,
2801 0, OST_SET_INFO, ofd_set_info_hdl,
2802 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2803 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2804 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl),
2805 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2806 OST_SETATTR, ofd_setattr_hdl),
2807 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2808 OST_CREATE, ofd_create_hdl),
2809 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2810 OST_DESTROY, ofd_destroy_hdl),
2811 TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl),
2812 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
2814 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2815 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write,
2817 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2818 OST_PUNCH, ofd_punch_hdl,
2820 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl),
2821 TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl),
2822 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl),
2823 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE, ofd_fallocate_hdl)
2826 static struct tgt_opc_slice ofd_common_slice[] = {
2828 .tos_opc_start = OST_FIRST_OPC,
2829 .tos_opc_end = OST_LAST_OPC,
2830 .tos_hs = ofd_tgt_handlers
2833 .tos_opc_start = OBD_FIRST_OPC,
2834 .tos_opc_end = OBD_LAST_OPC,
2835 .tos_hs = tgt_obd_handlers
2838 .tos_opc_start = LDLM_FIRST_OPC,
2839 .tos_opc_end = LDLM_LAST_OPC,
2840 .tos_hs = tgt_dlm_handlers
2843 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2844 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2845 .tos_hs = tgt_out_handlers
2848 .tos_opc_start = SEQ_FIRST_OPC,
2849 .tos_opc_end = SEQ_LAST_OPC,
2850 .tos_hs = seq_handlers
2853 .tos_opc_start = LFSCK_FIRST_OPC,
2854 .tos_opc_end = LFSCK_LAST_OPC,
2855 .tos_hs = tgt_lfsck_handlers
2858 .tos_opc_start = SEC_FIRST_OPC,
2859 .tos_opc_end = SEC_LAST_OPC,
2860 .tos_hs = tgt_sec_ctx_handlers
2867 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2868 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2871 * Implementation of lu_context_key::lct_key_exit.
2873 * Optional method called on lu_context_exit() for all allocated
2875 * It is used in OFD to sanitize context values which may be re-used
2876 * during another request processing by the same thread.
2878 * \param[in] ctx execution context
2879 * \param[in] key context key
2880 * \param[in] data ofd_thread_info
2882 static void ofd_key_exit(const struct lu_context *ctx,
2883 struct lu_context_key *key, void *data)
2885 struct ofd_thread_info *info = data;
2887 info->fti_env = NULL;
2888 info->fti_exp = NULL;
2891 info->fti_pre_version = 0;
2893 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2896 struct lu_context_key ofd_thread_key = {
2897 .lct_tags = LCT_DT_THREAD,
2898 .lct_init = ofd_key_init,
2899 .lct_fini = ofd_key_fini,
2900 .lct_exit = ofd_key_exit
2904 * Initialize OFD device according to parameters in the config log \a cfg.
2906 * This is the main starting point of OFD initialization. It fills all OFD
2907 * parameters with their initial values and calls other initializing functions
2908 * to set up all OFD subsystems.
2910 * \param[in] env execution environment
2911 * \param[in] m OFD device
2912 * \param[in] ldt LU device type of OFD
2913 * \param[in] cfg configuration log
2915 * \retval 0 if successful
2916 * \retval negative value on error
2918 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2919 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2921 const char *dev = lustre_cfg_string(cfg, 0);
2922 struct ofd_thread_info *info = NULL;
2923 struct obd_device *obd;
2924 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2926 struct nm_config_file *nodemap_config;
2927 struct obd_device_target *obt;
2933 obd = class_name2obd(dev);
2935 CERROR("Cannot find obd with name %s\n", dev);
2939 rc = lu_env_refill((struct lu_env *)env);
2944 obt->obt_magic = OBT_MAGIC;
2946 spin_lock_init(&m->ofd_flags_lock);
2947 m->ofd_raid_degraded = 0;
2948 m->ofd_checksum_t10pi_enforce = 0;
2949 m->ofd_sync_journal = 0;
2951 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2953 m->ofd_seq_count = 0;
2954 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2955 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2956 spin_lock_init(&m->ofd_inconsistency_lock);
2958 m->ofd_access_log_mask = -1; /* Log all accesses if enabled. */
2960 spin_lock_init(&m->ofd_batch_lock);
2961 init_rwsem(&m->ofd_lastid_rwsem);
2963 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2964 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2965 /* set this lu_device to obd, because error handling need it */
2966 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2968 /* No connection accepted until configurations will finish */
2969 spin_lock(&obd->obd_dev_lock);
2970 obd->obd_no_conn = 1;
2971 spin_unlock(&obd->obd_dev_lock);
2972 obd->obd_replayable = 1;
2973 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2974 char *str = lustre_cfg_string(cfg, 4);
2976 if (strchr(str, 'n')) {
2977 CWARN("%s: recovery disabled\n", obd->obd_name);
2978 obd->obd_replayable = 0;
2982 info = ofd_info_init(env, NULL);
2986 rc = ofd_stack_init(env, m, cfg, &lmd_flags);
2988 CERROR("%s: can't init device stack, rc %d\n",
2993 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
2994 ofd_procfs_add_brw_stats_symlink(m);
2997 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2998 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2999 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
3000 LDLM_NAMESPACE_SERVER,
3001 LDLM_NAMESPACE_GREEDY,
3003 if (m->ofd_namespace == NULL)
3004 GOTO(err_fini_stack, rc = -ENOMEM);
3005 /* set obd_namespace for compatibility with old code */
3006 obd->obd_namespace = m->ofd_namespace;
3007 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
3008 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
3009 m->ofd_namespace->ns_lvbp = m;
3011 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3012 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
3014 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
3015 OBD_FAIL_OST_ALL_REQUEST_NET,
3016 OBD_FAIL_OST_ALL_REPLY_NET);
3018 GOTO(err_free_ns, rc);
3020 if (lmd_flags & LMD_FLG_SKIP_LFSCK)
3021 m->ofd_skip_lfsck = 1;
3022 if (lmd_flags & LMD_FLG_LOCAL_RECOV)
3023 m->ofd_lut.lut_local_recovery = 1;
3025 rc = ofd_tunables_init(m);
3027 GOTO(err_fini_lut, rc);
3029 tgd->tgd_reserved_pcnt = 0;
3031 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
3032 m->ofd_cksum_types_supported =
3033 obd_cksum_types_supported_server(obd->obd_name);
3034 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
3035 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
3036 OFD_PRECREATE_SMALL_FS)
3037 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
3038 m->ofd_atime_diff = OFD_DEF_ATIME_DIFF;
3040 rc = ofd_fs_setup(env, m, obd);
3042 GOTO(err_fini_proc, rc);
3044 fid.f_seq = FID_SEQ_LOCAL_NAME;
3047 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
3050 GOTO(err_fini_fs, rc);
3052 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
3054 if (IS_ERR(nodemap_config)) {
3055 rc = PTR_ERR(nodemap_config);
3057 GOTO(err_fini_los, rc);
3059 obt->obt_nodemap_config_file = nodemap_config;
3062 rc = ofd_start_inconsistency_verification_thread(m);
3064 GOTO(err_fini_nm, rc);
3066 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
3071 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
3072 obt->obt_nodemap_config_file = NULL;
3074 local_oid_storage_fini(env, m->ofd_los);
3077 ofd_fs_cleanup(env, m);
3081 tgt_fini(env, &m->ofd_lut);
3083 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3084 obd->obd_namespace = m->ofd_namespace = NULL;
3086 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3091 * Stop the OFD device
3093 * This function stops the OFD device and all its subsystems.
3094 * This is the end of OFD lifecycle.
3096 * \param[in] env execution environment
3097 * \param[in] m OFD device
3099 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3101 struct obd_device *obd = ofd_obd(m);
3102 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3103 struct lfsck_stop stop;
3105 stop.ls_status = LS_PAUSED;
3107 lfsck_stop(env, m->ofd_osd, &stop);
3108 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3109 target_recovery_fini(obd);
3110 if (m->ofd_namespace != NULL)
3111 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3112 d->ld_obd->obd_force);
3114 obd_exports_barrier(obd);
3115 obd_zombie_barrier();
3118 tgt_fini(env, &m->ofd_lut);
3119 ofd_stop_inconsistency_verification_thread(m);
3120 lfsck_degister(env, m->ofd_osd);
3121 ofd_fs_cleanup(env, m);
3122 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3123 obd->u.obt.obt_nodemap_config_file = NULL;
3125 if (m->ofd_namespace != NULL) {
3126 ldlm_namespace_free_post(m->ofd_namespace);
3127 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3130 ofd_access_log_delete(m->ofd_access_log);
3131 m->ofd_access_log = NULL;
3133 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3135 LASSERT(atomic_read(&d->ld_ref) == 0);
3136 server_put_mount(obd->obd_name, true);
3141 * Implementation of lu_device_type_operations::ldto_device_fini.
3143 * Finalize device. Dual to ofd_device_init(). It is called from
3144 * obd_precleanup() and stops the current device.
3146 * \param[in] env execution environment
3147 * \param[in] d LU device of OFD
3151 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3152 struct lu_device *d)
3155 ofd_fini(env, ofd_dev(d));
3160 * Implementation of lu_device_type_operations::ldto_device_free.
3162 * Free OFD device. Dual to ofd_device_alloc().
3164 * \param[in] env execution environment
3165 * \param[in] d LU device of OFD
3169 static struct lu_device *ofd_device_free(const struct lu_env *env,
3170 struct lu_device *d)
3172 struct ofd_device *m = ofd_dev(d);
3174 dt_device_fini(&m->ofd_dt_dev);
3180 * Implementation of lu_device_type_operations::ldto_device_alloc.
3182 * This function allocates the new OFD device. It is called from
3183 * obd_setup() if OBD device had lu_device_type defined.
3185 * \param[in] env execution environment
3186 * \param[in] t lu_device_type of OFD device
3187 * \param[in] cfg configuration log
3189 * \retval pointer to the lu_device of just allocated OFD
3190 * \retval ERR_PTR of return value on error
3192 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3193 struct lu_device_type *t,
3194 struct lustre_cfg *cfg)
3196 struct ofd_device *m;
3197 struct lu_device *l;
3202 return ERR_PTR(-ENOMEM);
3204 l = &m->ofd_dt_dev.dd_lu_dev;
3205 dt_device_init(&m->ofd_dt_dev, t);
3206 rc = ofd_init0(env, m, t, cfg);
3208 ofd_device_free(env, l);
3215 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3216 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3218 static struct lu_device_type_operations ofd_device_type_ops = {
3219 .ldto_init = ofd_type_init,
3220 .ldto_fini = ofd_type_fini,
3222 .ldto_start = ofd_type_start,
3223 .ldto_stop = ofd_type_stop,
3225 .ldto_device_alloc = ofd_device_alloc,
3226 .ldto_device_free = ofd_device_free,
3227 .ldto_device_fini = ofd_device_fini
3230 static struct lu_device_type ofd_device_type = {
3231 .ldt_tags = LU_DEVICE_DT,
3232 .ldt_name = LUSTRE_OST_NAME,
3233 .ldt_ops = &ofd_device_type_ops,
3234 .ldt_ctx_tags = LCT_DT_THREAD
3238 * Initialize OFD module.
3240 * This function is called upon module loading. It registers OFD device type
3241 * and prepares all in-memory structures used by all OFD devices.
3243 * \retval 0 if successful
3244 * \retval negative value on error
3246 static int __init ofd_init(void)
3250 rc = lu_kmem_init(ofd_caches);
3254 rc = ofd_access_log_module_init();
3258 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3259 LUSTRE_OST_NAME, &ofd_device_type);
3261 goto out_ofd_access_log;
3266 ofd_access_log_module_exit();
3268 lu_kmem_fini(ofd_caches);
3276 * This function is called upon OFD module unloading.
3277 * It frees all related structures and unregisters OFD device type.
3279 static void __exit ofd_exit(void)
3281 class_unregister_type(LUSTRE_OST_NAME);
3282 ofd_access_log_module_exit();
3283 lu_kmem_fini(ofd_caches);
3286 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3287 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3288 MODULE_VERSION(LUSTRE_VERSION_STRING);
3289 MODULE_LICENSE("GPL");
3291 module_init(ofd_init);
3292 module_exit(ofd_exit);