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;
883 ktime_t kstart = ktime_get();
887 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
889 DEBUG_REQ(D_HA, req, "no set_info key");
890 RETURN(err_serious(-EFAULT));
892 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
895 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
897 DEBUG_REQ(D_HA, req, "no set_info val");
898 RETURN(err_serious(-EFAULT));
900 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
903 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
905 /* In this case the value is actually an RMF_OST_BODY, so we
906 * transmutate the type of this PTLRPC */
907 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
909 rc = req_capsule_server_pack(tsi->tsi_pill);
913 if (is_grant_shrink) {
914 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
916 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
919 /** handle grant shrink, similar to a read request */
920 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
922 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
924 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
927 CERROR("%s: Unsupported key %s\n",
928 tgt_name(tsi->tsi_tgt), (char *)key);
931 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
932 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
938 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
940 * This function returns a list of extents which describes how a file's
941 * blocks are laid out on the disk.
943 * \param[in] env execution environment
944 * \param[in] ofd OFD device
945 * \param[in] fid FID of object
946 * \param[in] fiemap fiemap structure to fill with data
948 * \retval 0 if \a fiemap is filled with data successfully
949 * \retval negative value on error
951 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
952 struct lu_fid *fid, struct fiemap *fiemap)
954 struct ofd_object *fo;
957 fo = ofd_object_find(env, ofd, fid);
959 CERROR("%s: error finding object "DFID"\n",
960 ofd_name(ofd), PFID(fid));
964 ofd_read_lock(env, fo);
965 if (ofd_object_exists(fo))
966 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
969 ofd_read_unlock(env, fo);
970 ofd_object_put(env, fo);
975 static int ofd_lock_unlock_region(const struct lu_env *env,
976 struct ldlm_namespace *ns,
977 struct ldlm_res_id *res_id,
978 unsigned long long begin,
979 unsigned long long end)
983 struct lustre_handle lh = { 0 };
985 LASSERT(begin <= end);
987 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
991 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
992 tgt_extent_unlock(&lh, LCK_PR);
998 * Lock the sparse areas of given resource.
1000 * The locking of sparse areas will cause dirty data to be flushed back from
1001 * clients. This is used when getting the FIEMAP of an object to make sure
1002 * there is no unaccounted cached data on clients.
1004 * This function goes through \a fiemap list of extents and locks only sparse
1005 * areas between extents.
1007 * \param[in] ns LDLM namespace
1008 * \param[in] res_id resource ID
1009 * \param[in] fiemap file extents mapping on disk
1010 * \param[in] locked list head of regions list
1012 * \retval 0 if successful
1013 * \retval negative value on error
1015 static int lock_zero_regions(const struct lu_env *env,
1016 struct ldlm_namespace *ns,
1017 struct ldlm_res_id *res_id,
1018 struct fiemap *fiemap)
1020 __u64 begin = fiemap->fm_start;
1023 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1027 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1028 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1029 if (fiemap_start[i].fe_logical > begin) {
1030 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1031 begin, fiemap_start[i].fe_logical);
1032 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1033 fiemap_start[i].fe_logical);
1038 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1041 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1042 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1043 begin, fiemap->fm_start + fiemap->fm_length);
1044 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1045 fiemap->fm_start + fiemap->fm_length);
1053 * OFD request handler for OST_GET_INFO RPC.
1055 * This is OFD-specific part of request handling. The OFD-specific keys are:
1056 * - KEY_LAST_ID (obsolete)
1060 * This function reads needed data from storage and fills reply with it.
1062 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1063 * and is kept for compatibility.
1065 * \param[in] tsi target session environment for this request
1067 * \retval 0 if successful
1068 * \retval negative value on error
1070 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1072 struct obd_export *exp = tsi->tsi_exp;
1073 struct ofd_device *ofd = ofd_exp(exp);
1074 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1077 int replylen, rc = 0;
1078 ktime_t kstart = ktime_get();
1082 /* this common part for get_info rpc */
1083 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1085 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1086 RETURN(err_serious(-EPROTO));
1088 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1091 if (KEY_IS(KEY_LAST_ID)) {
1093 struct ofd_seq *oseq;
1095 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1096 rc = req_capsule_server_pack(tsi->tsi_pill);
1098 RETURN(err_serious(rc));
1100 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1102 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1103 (u64)exp->exp_filter_data.fed_group);
1107 *last_id = ofd_seq_last_oid(oseq);
1108 ofd_seq_put(tsi->tsi_env, oseq);
1109 } else if (KEY_IS(KEY_FIEMAP)) {
1110 struct ll_fiemap_info_key *fm_key;
1111 struct fiemap *fiemap;
1114 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1116 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1117 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1119 RETURN(err_serious(rc));
1121 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1123 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1125 replylen = fiemap_count_to_size(
1126 fm_key->lfik_fiemap.fm_extent_count);
1127 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1128 RCL_SERVER, replylen);
1130 rc = req_capsule_server_pack(tsi->tsi_pill);
1132 RETURN(err_serious(rc));
1134 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1138 *fiemap = fm_key->lfik_fiemap;
1139 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1141 /* LU-3219: Lock the sparse areas to make sure dirty
1142 * flushed back from client, then call fiemap again. */
1143 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1144 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1145 ost_fid_build_resid(fid, &fti->fti_resid);
1146 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1147 &fti->fti_resid, fiemap);
1149 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1152 } else if (KEY_IS(KEY_LAST_FID)) {
1153 struct ofd_device *ofd = ofd_exp(exp);
1154 struct ofd_seq *oseq;
1158 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1159 rc = req_capsule_server_pack(tsi->tsi_pill);
1161 RETURN(err_serious(rc));
1163 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1165 RETURN(err_serious(-EPROTO));
1167 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1169 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1173 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1174 ostid_seq(&fti->fti_ostid));
1176 RETURN(PTR_ERR(oseq));
1178 rc = ostid_to_fid(fid, &oseq->os_oi,
1179 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1183 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1186 ofd_seq_put(tsi->tsi_env, oseq);
1188 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1192 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1193 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1199 * OFD request handler for OST_GETATTR RPC.
1201 * This is OFD-specific part of request handling. It finds the OFD object
1202 * by its FID, gets attributes from storage and packs result to the reply.
1204 * \param[in] tsi target session environment for this request
1206 * \retval 0 if successful
1207 * \retval negative value on error
1209 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1211 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1212 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1213 struct ost_body *repbody;
1214 struct lustre_handle lh = { 0 };
1215 struct ofd_object *fo;
1217 enum ldlm_mode lock_mode = LCK_PR;
1218 ktime_t kstart = ktime_get();
1223 LASSERT(tsi->tsi_ost_body != NULL);
1225 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1226 if (repbody == NULL)
1229 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1230 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1232 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1233 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1236 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1239 rc = tgt_extent_lock(tsi->tsi_env,
1240 tsi->tsi_tgt->lut_obd->obd_namespace,
1241 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1247 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1249 GOTO(out, rc = PTR_ERR(fo));
1251 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1255 obdo_from_la(&repbody->oa, &fti->fti_attr,
1256 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1258 /* Store object version in reply */
1259 curr_version = dt_version_get(tsi->tsi_env,
1260 ofd_object_child(fo));
1261 if ((__s64)curr_version != -EOPNOTSUPP) {
1262 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1263 repbody->oa.o_data_version = curr_version;
1266 if (fo->ofo_ff.ff_layout_version > 0) {
1267 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1268 repbody->oa.o_layout_version =
1269 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1271 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1272 PFID(&tsi->tsi_fid),
1273 repbody->oa.o_layout_version);
1277 ofd_object_put(tsi->tsi_env, fo);
1280 tgt_extent_unlock(&lh, lock_mode);
1282 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1283 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1285 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1286 repbody->oa.o_flags = OBD_FL_FLUSH;
1292 * OFD request handler for OST_SETATTR RPC.
1294 * This is OFD-specific part of request handling. It finds the OFD object
1295 * by its FID, sets attributes from request and packs result to the reply.
1297 * \param[in] tsi target session environment for this request
1299 * \retval 0 if successful
1300 * \retval negative value on error
1302 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1304 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1305 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1306 struct ost_body *body = tsi->tsi_ost_body;
1307 struct ost_body *repbody;
1308 struct ldlm_resource *res;
1309 struct ofd_object *fo;
1310 ktime_t kstart = ktime_get();
1315 LASSERT(body != NULL);
1317 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1318 if (repbody == NULL)
1321 repbody->oa.o_oi = body->oa.o_oi;
1322 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1324 /* This would be very bad - accidentally truncating a file when
1325 * changing the time or similar - bug 12203. */
1326 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1327 body->oa.o_size != OBD_OBJECT_EOF) {
1328 static char mdsinum[48];
1330 if (body->oa.o_valid & OBD_MD_FLFID)
1331 snprintf(mdsinum, sizeof(mdsinum) - 1,
1332 "of parent "DFID, body->oa.o_parent_seq,
1333 body->oa.o_parent_oid, 0);
1337 CERROR("%s: setattr from %s is trying to truncate object "DFID
1338 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1339 PFID(&tsi->tsi_fid), mdsinum);
1343 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1345 GOTO(out, rc = PTR_ERR(fo));
1347 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1348 fti->fti_attr.la_valid &= ~LA_TYPE;
1350 /* setting objects attributes (including owner/group) */
1351 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1355 obdo_from_la(&repbody->oa, &fti->fti_attr,
1356 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1358 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1359 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1362 ofd_object_put(tsi->tsi_env, fo);
1365 /* we do not call this before to avoid lu_object_find() in
1366 * ->lvbo_update() holding another reference on the object.
1367 * otherwise concurrent destroy can make the object unavailable
1368 * for 2nd lu_object_find() waiting for the first reference
1369 * to go... deadlock! */
1370 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1371 &tsi->tsi_resid, LDLM_EXTENT, 0);
1373 ldlm_res_lvbo_update(res, NULL, 0);
1374 ldlm_resource_putref(res);
1381 * Destroy OST orphans.
1383 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1384 * set then we must destroy possible orphaned objects.
1386 * \param[in] env execution environment
1387 * \param[in] exp OBD export
1388 * \param[in] ofd OFD device
1389 * \param[in] oa obdo structure for reply
1391 * \retval 0 if successful
1392 * \retval negative value on error
1394 static int ofd_orphans_destroy(const struct lu_env *env,
1395 struct obd_export *exp,
1396 struct ofd_device *ofd, struct obdo *oa)
1398 struct ofd_thread_info *info = ofd_info(env);
1399 struct lu_fid *fid = &info->fti_fid;
1400 struct ost_id *oi = &oa->o_oi;
1401 struct ofd_seq *oseq;
1402 u64 seq = ostid_seq(oi);
1403 u64 end_id = ostid_id(oi);
1411 oseq = ofd_seq_get(ofd, seq);
1413 CERROR("%s: Can not find seq for "DOSTID"\n",
1414 ofd_name(ofd), POSTID(oi));
1419 last = ofd_seq_last_oid(oseq);
1422 LASSERT(exp != NULL);
1423 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1425 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1428 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1429 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1431 while (oid > end_id) {
1432 rc = fid_set_id(fid, oid);
1433 if (unlikely(rc != 0))
1436 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1437 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1438 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1439 /* this is pretty fatal... */
1440 CEMERG("%s: error destroying precreated id "
1442 ofd_name(ofd), PFID(fid), rc);
1446 ofd_seq_last_oid_set(oseq, oid);
1447 /* update last_id on disk periodically so that if we
1448 * restart * we don't need to re-scan all of the just
1449 * deleted objects. */
1450 if ((oid & 511) == 0)
1451 ofd_seq_last_oid_write(env, ofd, oseq);
1455 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1456 ofd_name(ofd), seq, oid);
1460 ofd_seq_last_oid_set(oseq, oid);
1461 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1463 /* don't reuse orphan object, return last used objid */
1464 rc = ostid_set_id(oi, last);
1470 ofd_seq_put(env, oseq);
1475 * OFD request handler for OST_CREATE RPC.
1477 * This is OFD-specific part of request handling. Its main purpose is to
1478 * create new data objects on OST, but it also used to destroy orphans.
1480 * \param[in] tsi target session environment for this request
1482 * \retval 0 if successful
1483 * \retval negative value on error
1485 static int ofd_create_hdl(struct tgt_session_info *tsi)
1487 struct ptlrpc_request *req = tgt_ses_req(tsi);
1488 struct ost_body *repbody;
1489 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1490 struct obdo *rep_oa;
1491 struct obd_export *exp = tsi->tsi_exp;
1492 struct ofd_device *ofd = ofd_exp(exp);
1493 u64 seq = ostid_seq(&oa->o_oi);
1494 u64 oid = ostid_id(&oa->o_oi);
1495 struct ofd_seq *oseq;
1498 ktime_t kstart = ktime_get();
1504 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1507 if (ofd->ofd_no_precreate)
1510 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1511 if (repbody == NULL)
1514 down_read(&ofd->ofd_lastid_rwsem);
1515 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1516 * we may do that in the future.
1517 * Return -ENOSPC until the LAST_ID rebuilt. */
1518 if (unlikely(ofd->ofd_lastid_rebuilding))
1519 GOTO(out_sem, rc = -ENOSPC);
1521 rep_oa = &repbody->oa;
1522 rep_oa->o_oi = oa->o_oi;
1524 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1526 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1528 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1530 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1531 ofd_name(ofd), seq, PTR_ERR(oseq));
1532 GOTO(out_sem, rc = -EINVAL);
1535 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1536 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1537 if (!ofd_obd(ofd)->obd_recovering ||
1538 oid > ofd_seq_last_oid(oseq)) {
1539 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1540 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1541 ofd_seq_last_oid(oseq));
1542 GOTO(out_nolock, rc = -EINVAL);
1544 /* Do nothing here, we re-create objects during recovery
1545 * upon write replay, see ofd_preprw_write() */
1546 GOTO(out_nolock, rc = 0);
1548 /* former ofd_handle_precreate */
1549 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1550 (oa->o_flags & OBD_FL_DELORPHAN)) {
1551 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1553 /* destroy orphans */
1554 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1555 exp->exp_conn_cnt) {
1556 CERROR("%s: dropping old orphan cleanup request\n",
1558 GOTO(out_nolock, rc = 0);
1560 /* This causes inflight precreates to abort and drop lock */
1561 oseq->os_destroys_in_progress = 1;
1562 mutex_lock(&oseq->os_create_lock);
1563 if (!oseq->os_destroys_in_progress) {
1565 "%s:[%llu] destroys_in_progress already cleared\n",
1566 ofd_name(ofd), seq);
1567 rc = ostid_set_id(&rep_oa->o_oi,
1568 ofd_seq_last_oid(oseq));
1571 diff = oid - ofd_seq_last_oid(oseq);
1572 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1573 ofd_seq_last_oid(oseq), diff);
1574 if (-diff > OST_MAX_PRECREATE) {
1575 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1576 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1577 "(%llu), trust the OST\n",
1578 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1579 PFID(&oseq->os_oi.oi_fid),
1580 ofd_seq_last_oid(oseq));
1582 /* Let MDS know that we are so far ahead. */
1583 rc = ostid_set_id(&rep_oa->o_oi,
1584 ofd_seq_last_oid(oseq) + 1);
1585 } else if (diff < 0) {
1586 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1588 oseq->os_destroys_in_progress = 0;
1590 /* XXX: Used by MDS for the first time! */
1591 oseq->os_destroys_in_progress = 0;
1594 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1595 ofd->ofd_lastid_gen)) {
1596 /* Keep the export ref so we can send the reply. */
1597 ofd_obd_disconnect(class_export_get(exp));
1598 GOTO(out_nolock, rc = -ENOTCONN);
1601 mutex_lock(&oseq->os_create_lock);
1602 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1603 exp->exp_conn_cnt) {
1604 CERROR("%s: dropping old precreate request\n",
1608 /* only precreate if seq is 0, IDIF or normal and also o_id
1609 * must be specfied */
1610 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1611 !fid_seq_is_idif(seq)) || oid == 0) {
1612 diff = 1; /* shouldn't we create this right now? */
1614 diff = oid - ofd_seq_last_oid(oseq);
1615 /* Do sync create if the seq is about to used up */
1616 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1617 if (unlikely(oid >= IDIF_MAX_OID - 1))
1619 } else if (fid_seq_is_norm(seq)) {
1621 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1624 CERROR("%s : invalid o_seq "DOSTID"\n",
1625 ofd_name(ofd), POSTID(&oa->o_oi));
1626 GOTO(out, rc = -EINVAL);
1631 CERROR("%s: invalid precreate request for "
1632 DOSTID", last_id %llu. "
1633 "Likely MDS last_id corruption\n",
1634 ofd_name(ofd), POSTID(&oa->o_oi),
1635 ofd_seq_last_oid(oseq));
1636 GOTO(out, rc = -EINVAL);
1641 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1647 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1648 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1649 /* don't enforce grant during orphan recovery */
1650 granted = tgt_grant_create(tsi->tsi_env,
1651 ofd_obd(ofd)->obd_self_export,
1656 CDEBUG(D_HA, "%s: failed to acquire grant "
1657 "space for precreate (%lld): rc = %d\n",
1658 ofd_name(ofd), diff, rc);
1663 /* This can happen if a new OST is formatted and installed
1664 * in place of an old one at the same index. Instead of
1665 * precreating potentially millions of deleted old objects
1666 * (possibly filling the OST), only precreate the last batch.
1667 * LFSCK will eventually clean up any orphans. LU-14 */
1668 if (diff > 5 * OST_MAX_PRECREATE) {
1669 diff = OST_MAX_PRECREATE / 2;
1670 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1671 "OST replaced or reformatted: "
1672 "LFSCK will clean up",
1675 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1676 "%u larger than the LAST_ID "DOSTID", only "
1677 "precreating the last %lld objects.\n",
1678 ofd_name(ofd), POSTID(&oa->o_oi),
1679 5 * OST_MAX_PRECREATE,
1680 POSTID(&oseq->os_oi), diff);
1681 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1685 next_id = ofd_seq_last_oid(oseq) + 1;
1686 count = ofd_precreate_batch(ofd, (int)diff);
1688 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1689 " at %llu\n", ofd_name(ofd),
1690 count, seq, next_id);
1692 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1693 && ktime_get_seconds() > enough_time) {
1694 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1695 " created at a rate of %d/s\n",
1696 ofd_name(ofd), created, diff + created,
1697 created / DISK_TIMEOUT);
1701 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1702 oseq, count, sync_trans);
1706 } else if (rc < 0) {
1712 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1713 LCONSOLE_WARN("%s: can't create the same count of"
1714 " objects when replaying the request"
1715 " (diff is %lld). see LU-4621\n",
1716 ofd_name(ofd), diff);
1719 /* some objects got created, we can return
1720 * them, even if last creation failed */
1723 CERROR("%s: unable to precreate: rc = %d\n",
1726 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1727 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1728 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1733 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1737 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1738 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1739 if (unlikely(!oseq->os_last_id_synced))
1740 oseq->os_last_id_synced = 1;
1742 mutex_unlock(&oseq->os_create_lock);
1745 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1746 ofd_seq_put(tsi->tsi_env, oseq);
1749 up_read(&ofd->ofd_lastid_rwsem);
1754 * OFD request handler for OST_DESTROY RPC.
1756 * This is OFD-specific part of request handling. It destroys data objects
1757 * related to destroyed object on MDT.
1759 * \param[in] tsi target session environment for this request
1761 * \retval 0 if successful
1762 * \retval negative value on error
1764 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1766 const struct ost_body *body = tsi->tsi_ost_body;
1767 struct ost_body *repbody;
1768 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1769 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1770 struct lu_fid *fid = &fti->fti_fid;
1771 ktime_t kstart = ktime_get();
1778 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1781 /* This is old case for clients before Lustre 2.4 */
1782 /* If there's a DLM request, cancel the locks mentioned in it */
1783 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1785 struct ldlm_request *dlm;
1787 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1790 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1793 *fid = body->oa.o_oi.oi_fid;
1794 oid = ostid_id(&body->oa.o_oi);
1797 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1799 /* check that o_misc makes sense */
1800 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1801 count = body->oa.o_misc;
1803 count = 1; /* default case - single destroy */
1805 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1806 POSTID(&body->oa.o_oi), count);
1811 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1812 if (lrc == -ENOENT) {
1814 "%s: destroying non-existent object "DFID"\n",
1815 ofd_name(ofd), PFID(fid));
1816 /* rewrite rc with -ENOENT only if it is 0 */
1819 } else if (lrc != 0) {
1820 CERROR("%s: error destroying object "DFID": %d\n",
1821 ofd_name(ofd), PFID(fid), lrc);
1827 lrc = fid_set_id(fid, oid);
1828 if (unlikely(lrc != 0 && count > 0))
1829 GOTO(out, rc = lrc);
1832 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1833 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1838 fid_to_ostid(fid, &repbody->oa.o_oi);
1843 * OFD request handler for OST_STATFS RPC.
1845 * This function gets statfs data from storage as part of request
1848 * \param[in] tsi target session environment for this request
1850 * \retval 0 if successful
1851 * \retval negative value on error
1853 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1855 ktime_t kstart = ktime_get();
1856 struct obd_statfs *osfs;
1861 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1863 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1865 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1866 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1868 CERROR("%s: statfs failed: rc = %d\n",
1869 tgt_name(tsi->tsi_tgt), rc);
1871 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1874 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1875 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1881 * OFD request handler for OST_SYNC RPC.
1883 * Sync object data or all filesystem data to the disk and pack the
1886 * \param[in] tsi target session environment for this request
1888 * \retval 0 if successful
1889 * \retval negative value on error
1891 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1893 struct ost_body *body = tsi->tsi_ost_body;
1894 struct ost_body *repbody;
1895 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1896 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1897 struct ofd_object *fo = NULL;
1898 ktime_t kstart = ktime_get();
1903 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1905 /* if no objid is specified, it means "sync whole filesystem" */
1906 if (!fid_is_zero(&tsi->tsi_fid)) {
1907 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1909 RETURN(PTR_ERR(fo));
1912 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1913 fo != NULL ? ofd_object_child(fo) : NULL,
1914 repbody->oa.o_size, repbody->oa.o_blocks);
1918 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1919 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1923 repbody->oa.o_oi = body->oa.o_oi;
1924 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1926 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1928 obdo_from_la(&repbody->oa, &fti->fti_attr,
1931 /* don't return rc from getattr */
1936 ofd_object_put(tsi->tsi_env, fo);
1941 * OFD request handler for OST_FALLOCATE RPC.
1943 * This is part of request processing. Validate request fields,
1944 * preallocate the given OFD object and pack reply.
1946 * \param[in] tsi target session environment for this request
1948 * \retval 0 if successful
1949 * \retval negative value on error
1951 static int ofd_fallocate_hdl(struct tgt_session_info *tsi)
1953 struct obdo *oa = &tsi->tsi_ost_body->oa;
1954 struct ost_body *repbody;
1955 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1956 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1957 struct ldlm_resource *res;
1958 struct ofd_object *fo;
1960 struct lustre_handle lh = { 0, };
1964 ktime_t kstart = ktime_get();
1966 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1967 if (repbody == NULL)
1968 RETURN(err_serious(-ENOMEM));
1971 * fallocate start and end are passed in o_size, o_blocks
1976 mode = oa->o_falloc_mode;
1978 * Only mode == 0 (which is standard prealloc) is supported now.
1979 * Punch is not supported yet.
1981 if (mode & ~FALLOC_FL_KEEP_SIZE)
1982 RETURN(-EOPNOTSUPP);
1984 repbody->oa.o_oi = oa->o_oi;
1985 repbody->oa.o_valid = OBD_MD_FLID;
1987 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1988 oa->o_flags & OBD_FL_SRVLOCK;
1991 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid,
1992 start, end, &lh, LCK_PW, &flags);
1997 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2000 GOTO(out, rc = PTR_ERR(fo));
2002 la_from_obdo(&info->fti_attr, oa,
2003 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2005 rc = ofd_object_fallocate(tsi->tsi_env, fo, start, end, mode,
2006 &info->fti_attr, oa);
2010 rc = ofd_attr_get(tsi->tsi_env, fo, &info->fti_attr);
2012 obdo_from_la(&repbody->oa, &info->fti_attr,
2017 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PREALLOC,
2018 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2022 ofd_object_put(tsi->tsi_env, fo);
2025 tgt_extent_unlock(&lh, LCK_PW);
2027 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2030 struct ost_lvb *res_lvb;
2032 ldlm_res_lvbo_update(res, NULL, 0);
2033 res_lvb = res->lr_lvb_data;
2035 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2036 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2038 repbody->oa.o_valid |= OBD_MD_FLSIZE;
2039 repbody->oa.o_size = res_lvb->lvb_size;
2041 ldlm_resource_putref(res);
2050 * OFD request handler for OST_PUNCH RPC.
2052 * This is part of request processing. Validate request fields,
2053 * punch (truncate) the given OFD object and pack reply.
2055 * \param[in] tsi target session environment for this request
2057 * \retval 0 if successful
2058 * \retval negative value on error
2060 static int ofd_punch_hdl(struct tgt_session_info *tsi)
2062 const struct obdo *oa = &tsi->tsi_ost_body->oa;
2063 struct ost_body *repbody;
2064 struct ofd_thread_info *info = tsi2ofd_info(tsi);
2065 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
2066 struct ldlm_resource *res;
2067 struct ofd_object *fo;
2069 struct lustre_handle lh = { 0, };
2072 ktime_t kstart = ktime_get();
2077 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
2079 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
2080 BUILD_BUG_ON(!(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK));
2082 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2083 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2084 RETURN(err_serious(-EPROTO));
2086 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2087 if (repbody == NULL)
2088 RETURN(err_serious(-ENOMEM));
2090 /* punch start,end are passed in o_size,o_blocks throught wire */
2094 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2097 /* standard truncate optimization: if file body is completely
2098 * destroyed, don't send data back to the server. */
2100 flags |= LDLM_FL_AST_DISCARD_DATA;
2102 repbody->oa.o_oi = oa->o_oi;
2103 repbody->oa.o_valid = OBD_MD_FLID;
2105 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2106 oa->o_flags & OBD_FL_SRVLOCK;
2109 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
2110 end, &lh, LCK_PW, &flags);
2115 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2116 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2117 oa->o_valid, start, end);
2119 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2122 GOTO(out, rc = PTR_ERR(fo));
2124 la_from_obdo(&info->fti_attr, oa,
2125 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2126 info->fti_attr.la_size = start;
2127 info->fti_attr.la_valid |= LA_SIZE;
2129 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2134 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2135 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2138 ofd_object_put(tsi->tsi_env, fo);
2141 tgt_extent_unlock(&lh, LCK_PW);
2143 /* we do not call this before to avoid lu_object_find() in
2144 * ->lvbo_update() holding another reference on the object.
2145 * otherwise concurrent destroy can make the object unavailable
2146 * for 2nd lu_object_find() waiting for the first reference
2147 * to go... deadlock! */
2148 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2151 struct ost_lvb *res_lvb;
2153 ldlm_res_lvbo_update(res, NULL, 0);
2154 res_lvb = res->lr_lvb_data;
2155 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2156 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2157 ldlm_resource_putref(res);
2163 static int ofd_ladvise_prefetch(const struct lu_env *env,
2164 struct ofd_object *fo,
2165 struct niobuf_local *lnb,
2166 __u64 start, __u64 end, enum dt_bufs_type dbt)
2168 struct ofd_thread_info *info = ofd_info(env);
2169 pgoff_t start_index, end_index, pages;
2170 struct niobuf_remote rnb;
2171 unsigned long nr_local;
2177 ofd_read_lock(env, fo);
2178 if (!ofd_object_exists(fo))
2179 GOTO(out_unlock, rc = -ENOENT);
2181 rc = ofd_attr_get(env, fo, &info->fti_attr);
2183 GOTO(out_unlock, rc);
2185 if (end > info->fti_attr.la_size)
2186 end = info->fti_attr.la_size;
2189 GOTO(out_unlock, rc);
2191 /* We need page aligned offset and length */
2192 start_index = start >> PAGE_SHIFT;
2193 end_index = (end - 1) >> PAGE_SHIFT;
2194 pages = end_index - start_index + 1;
2196 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2197 PTLRPC_MAX_BRW_PAGES;
2198 rnb.rnb_offset = start_index << PAGE_SHIFT;
2199 rnb.rnb_len = nr_local << PAGE_SHIFT;
2200 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb,
2201 PTLRPC_MAX_BRW_PAGES, dbt);
2202 if (unlikely(rc < 0))
2205 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2206 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2209 start_index += nr_local;
2214 ofd_read_unlock(env, fo);
2219 * OFD request handler for OST_LADVISE RPC.
2221 * Tune cache or perfetch policies according to advices.
2223 * \param[in] tsi target session environment for this request
2225 * \retval 0 if successful
2226 * \retval negative errno on error
2228 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2230 struct ptlrpc_request *req = tgt_ses_req(tsi);
2231 struct obd_export *exp = tsi->tsi_exp;
2232 struct ofd_device *ofd = ofd_exp(exp);
2233 struct ost_body *body, *repbody;
2234 struct ofd_thread_info *info;
2235 struct ofd_object *fo;
2236 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2237 const struct lu_env *env = svc_thread->t_env;
2238 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2239 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2240 struct lu_ladvise *ladvise;
2242 struct ladvise_hdr *ladvise_hdr;
2243 struct obd_ioobj ioo;
2244 struct lustre_handle lockh = { 0 };
2247 struct dt_object *dob;
2253 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2254 body = tsi->tsi_ost_body;
2256 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2257 RETURN(err_serious(-EPROTO));
2259 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2260 &RMF_OST_LADVISE_HDR);
2261 if (ladvise_hdr == NULL)
2262 RETURN(err_serious(-EPROTO));
2264 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2265 ladvise_hdr->lah_count < 1)
2266 RETURN(err_serious(-EPROTO));
2268 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2269 RETURN(err_serious(-EPROTO));
2271 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2272 if (ladvise == NULL)
2273 RETURN(err_serious(-EPROTO));
2275 num_advise = req_capsule_get_size(&req->rq_pill,
2276 &RMF_OST_LADVISE, RCL_CLIENT) /
2278 if (num_advise < ladvise_hdr->lah_count)
2279 RETURN(err_serious(-EPROTO));
2281 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2282 repbody->oa = body->oa;
2284 info = ofd_info_init(env, exp);
2286 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2287 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2291 fo = ofd_object_find(env, ofd, &info->fti_fid);
2296 LASSERT(fo != NULL);
2297 dob = ofd_object_child(fo);
2299 if (ptlrpc_connection_is_local(exp->exp_connection))
2300 dbt |= DT_BUFS_TYPE_LOCAL;
2302 for (i = 0; i < num_advise; i++, ladvise++) {
2303 start = ladvise->lla_start;
2304 end = ladvise->lla_end;
2306 rc = err_serious(-EPROTO);
2310 /* Handle different advice types */
2311 switch (ladvise->lla_advice) {
2315 case LU_LADVISE_WILLREAD:
2319 ioo.ioo_oid = body->oa.o_oi;
2321 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2322 &tsi->tsi_resid, start, end - 1,
2323 &lockh, LCK_PR, &flags);
2327 req->rq_status = ofd_ladvise_prefetch(env, fo,
2330 tgt_extent_unlock(&lockh, LCK_PR);
2332 case LU_LADVISE_DONTNEED:
2333 rc = dt_ladvise(env, dob, ladvise->lla_start,
2334 ladvise->lla_end, LU_LADVISE_DONTNEED);
2341 ofd_object_put(env, fo);
2342 req->rq_status = rc;
2347 * OFD request handler for OST_QUOTACTL RPC.
2349 * This is part of request processing to validate incoming request fields,
2350 * get the requested data from OSD and pack reply.
2352 * \param[in] tsi target session environment for this request
2354 * \retval 0 if successful
2355 * \retval negative value on error
2357 static int ofd_quotactl(struct tgt_session_info *tsi)
2359 struct obd_quotactl *oqctl, *repoqc;
2360 struct lu_nodemap *nodemap;
2361 ktime_t kstart = ktime_get();
2367 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2369 RETURN(err_serious(-EPROTO));
2371 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2373 RETURN(err_serious(-ENOMEM));
2377 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2378 if (IS_ERR(nodemap))
2379 RETURN(PTR_ERR(nodemap));
2382 if (oqctl->qc_type == USRQUOTA)
2383 id = nodemap_map_id(nodemap, NODEMAP_UID,
2384 NODEMAP_CLIENT_TO_FS,
2386 else if (oqctl->qc_type == GRPQUOTA)
2387 id = nodemap_map_id(nodemap, NODEMAP_GID,
2388 NODEMAP_CLIENT_TO_FS,
2391 nodemap_putref(nodemap);
2393 if (repoqc->qc_id != id)
2394 swap(repoqc->qc_id, id);
2396 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2398 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2399 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2401 if (repoqc->qc_id != id)
2402 swap(repoqc->qc_id, id);
2408 * Prolong lock timeout for the given extent.
2410 * This function finds all locks related with incoming request and
2411 * prolongs their timeout.
2413 * If a client is holding a lock for a long time while it sends
2414 * read or write RPCs to the OST for the object under this lock,
2415 * then we don't want the OST to evict the client. Otherwise,
2416 * if the network or disk is very busy then the client may not
2417 * be able to make any progress to clear out dirty pages under
2418 * the lock and the application will fail.
2420 * Every time a Bulk Read/Write (BRW) request arrives for the object
2421 * covered by the lock, extend the timeout on that lock. The RPC should
2422 * contain a lock handle for the lock it is using, but this
2423 * isn't handled correctly by all client versions, and the
2424 * request may cover multiple locks.
2426 * \param[in] tsi target session environment for this request
2427 * \param[in] data struct of data to prolong locks
2430 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2431 struct ldlm_prolong_args *data)
2433 struct obdo *oa = &tsi->tsi_ost_body->oa;
2434 struct ldlm_lock *lock;
2438 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2439 data->lpa_export = tsi->tsi_exp;
2440 data->lpa_resid = tsi->tsi_resid;
2442 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2443 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2444 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2445 data->lpa_extent.end);
2447 if (oa->o_valid & OBD_MD_FLHANDLE) {
2448 /* mostly a request should be covered by only one lock, try
2450 lock = ldlm_handle2lock(&oa->o_handle);
2452 /* Fast path to check if the lock covers the whole IO
2453 * region exclusively. */
2454 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2455 &data->lpa_extent)) {
2457 LASSERT(lock->l_export == data->lpa_export);
2458 ldlm_lock_prolong_one(lock, data);
2459 LDLM_LOCK_PUT(lock);
2460 if (data->lpa_locks_cnt > 0)
2462 /* The lock was destroyed probably lets try
2465 lock->l_last_used = ktime_get();
2466 LDLM_LOCK_PUT(lock);
2471 ldlm_resource_prolong(data);
2476 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2478 * Determine if \a lock and the lock from request \a req are equivalent
2479 * by comparing their resource names, modes, and extents.
2481 * It is used to give priority to read and write RPCs being done
2482 * under this lock so that the client can drop the contended
2483 * lock more quickly and let other clients use it. This improves
2484 * overall performance in the case where the first client gets a
2485 * very large lock extent that prevents other clients from
2486 * submitting their writes.
2488 * \param[in] req ptlrpc_request being processed
2489 * \param[in] lock contended lock to match
2491 * \retval 1 if lock is matched
2492 * \retval 0 otherwise
2494 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2495 struct ldlm_lock *lock)
2497 struct niobuf_remote *rnb;
2498 struct obd_ioobj *ioo;
2499 enum ldlm_mode mode;
2500 struct ldlm_extent ext;
2501 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2505 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2506 LASSERT(ioo != NULL);
2508 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2509 LASSERT(rnb != NULL);
2511 ext.start = rnb->rnb_offset;
2512 rnb += ioo->ioo_bufcnt - 1;
2513 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2515 LASSERT(lock->l_resource != NULL);
2516 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2519 /* a bulk write can only hold a reference on a PW extent lock
2522 mode = LCK_PW | LCK_GROUP;
2523 if (opc == OST_READ)
2524 /* whereas a bulk read can be protected by either a PR or PW
2528 if (!(lock->l_granted_mode & mode))
2531 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2535 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2537 * Check for whether the given PTLRPC request (\a req) is blocking
2538 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2541 * \param[in] req the incoming request
2543 * \retval 1 if \a req is blocking an LDLM lock cancel
2544 * \retval 0 if it is not
2545 * \retval -ESTALE if lock is not found
2547 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2549 struct tgt_session_info *tsi;
2550 struct obd_ioobj *ioo;
2551 struct niobuf_remote *rnb;
2553 struct ldlm_prolong_args pa = { 0 };
2557 /* Don't use tgt_ses_info() to get session info, because lock_match()
2558 * can be called while request has no processing thread yet. */
2559 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2562 * Use LASSERT below because malformed RPCs should have
2563 * been filtered out in tgt_hpreq_handler().
2565 opc = lustre_msg_get_opc(req->rq_reqmsg);
2566 LASSERT(opc == OST_READ || opc == OST_WRITE);
2568 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2569 LASSERT(ioo != NULL);
2571 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2572 LASSERT(rnb != NULL);
2573 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2575 pa.lpa_mode = LCK_PW | LCK_GROUP;
2576 if (opc == OST_READ)
2577 pa.lpa_mode |= LCK_PR;
2579 pa.lpa_extent.start = rnb->rnb_offset;
2580 rnb += ioo->ioo_bufcnt - 1;
2581 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2583 DEBUG_REQ(D_RPCTRACE, req,
2584 "%s %s: refresh rw locks for "DFID" (%llu->%llu)",
2585 tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid),
2586 pa.lpa_extent.start, pa.lpa_extent.end);
2588 ofd_prolong_extent_locks(tsi, &pa);
2590 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n",
2591 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2593 if (pa.lpa_blocks_cnt > 0)
2596 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2600 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2602 * Called after the request has been handled. It refreshes lock timeout again
2603 * so that client has more time to send lock cancel RPC.
2605 * \param[in] req request which is being processed.
2607 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2609 ofd_rw_hpreq_check(req);
2613 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2615 * This function checks if the given lock is the same by its resname, mode
2616 * and extent as one taken from the request.
2617 * It is used to give priority to punch/truncate RPCs that might lead to
2618 * the fastest release of that lock when a lock is contended.
2620 * \param[in] req ptlrpc_request being processed
2621 * \param[in] lock contended lock to match
2623 * \retval 1 if lock is matched
2624 * \retval 0 otherwise
2626 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2627 struct ldlm_lock *lock)
2629 struct tgt_session_info *tsi;
2631 struct ldlm_extent ext;
2635 /* Don't use tgt_ses_info() to get session info, because lock_match()
2636 * can be called while request has no processing thread yet. */
2637 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2640 * Use LASSERT below because malformed RPCs should have
2641 * been filtered out in tgt_hpreq_handler().
2643 LASSERT(tsi->tsi_ost_body != NULL);
2644 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2645 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2648 oa = &tsi->tsi_ost_body->oa;
2649 ext.start = oa->o_size;
2650 ext.end = oa->o_blocks;
2652 LASSERT(lock->l_resource != NULL);
2653 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2656 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2659 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2663 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2665 * High-priority queue request check for whether the given punch request
2666 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2667 * covered by an LDLM lock.
2671 * \param[in] req the incoming request
2673 * \retval 1 if \a req is blocking an LDLM lock cancel
2674 * \retval 0 if it is not
2675 * \retval -ESTALE if lock is not found
2677 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2679 struct tgt_session_info *tsi;
2681 struct ldlm_prolong_args pa = { 0 };
2685 /* Don't use tgt_ses_info() to get session info, because lock_match()
2686 * can be called while request has no processing thread yet. */
2687 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2688 LASSERT(tsi != NULL);
2689 oa = &tsi->tsi_ost_body->oa;
2691 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2692 oa->o_flags & OBD_FL_SRVLOCK));
2694 pa.lpa_mode = LCK_PW | LCK_GROUP;
2695 pa.lpa_extent.start = oa->o_size;
2696 pa.lpa_extent.end = oa->o_blocks;
2699 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2700 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2701 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2703 ofd_prolong_extent_locks(tsi, &pa);
2705 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2706 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2708 if (pa.lpa_blocks_cnt > 0)
2711 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2715 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2717 * Called after the request has been handled. It refreshes lock timeout again
2718 * so that client has more time to send lock cancel RPC.
2720 * \param[in] req request which is being processed.
2722 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2724 ofd_punch_hpreq_check(req);
2727 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2728 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2729 .hpreq_check = ofd_rw_hpreq_check,
2730 .hpreq_fini = ofd_rw_hpreq_fini
2733 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2734 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2735 .hpreq_check = ofd_punch_hpreq_check,
2736 .hpreq_fini = ofd_punch_hpreq_fini
2740 * Assign high priority operations to an IO request.
2742 * Check if the incoming request is a candidate for
2743 * high-priority processing. If it is, assign it a high
2744 * priority operations table.
2746 * \param[in] tsi target session environment for this request
2748 static void ofd_hp_brw(struct tgt_session_info *tsi)
2750 struct niobuf_remote *rnb;
2751 struct obd_ioobj *ioo;
2755 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2756 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2757 if (ioo->ioo_bufcnt > 0) {
2758 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2759 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2761 /* no high priority if server lock is needed */
2762 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2763 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2767 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2771 * Assign high priority operations to an punch request.
2773 * Check if the incoming request is a candidate for
2774 * high-priority processing. If it is, assign it a high
2775 * priority operations table.
2777 * \param[in] tsi target session environment for this request
2779 static void ofd_hp_punch(struct tgt_session_info *tsi)
2781 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2782 /* no high-priority if server lock is needed */
2783 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2784 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2785 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2786 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2788 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2791 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2792 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2793 #define OST_BRW_READ OST_READ
2794 #define OST_BRW_WRITE OST_WRITE
2797 * Table of OFD-specific request handlers
2799 * This table contains all opcodes accepted by OFD and
2800 * specifies handlers for them. The tgt_request_handler()
2801 * uses such table from each target to process incoming
2804 static struct tgt_handler ofd_tgt_handlers[] = {
2805 TGT_RPC_HANDLER(OST_FIRST_OPC,
2806 0, OST_CONNECT, tgt_connect,
2807 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2808 TGT_RPC_HANDLER(OST_FIRST_OPC,
2809 0, OST_DISCONNECT, tgt_disconnect,
2810 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2811 TGT_RPC_HANDLER(OST_FIRST_OPC,
2812 0, OST_SET_INFO, ofd_set_info_hdl,
2813 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2814 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2815 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl),
2816 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2817 OST_SETATTR, ofd_setattr_hdl),
2818 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2819 OST_CREATE, ofd_create_hdl),
2820 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2821 OST_DESTROY, ofd_destroy_hdl),
2822 TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl),
2823 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
2825 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2826 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write,
2828 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2829 OST_PUNCH, ofd_punch_hdl,
2831 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl),
2832 TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl),
2833 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl),
2834 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE, ofd_fallocate_hdl)
2837 static struct tgt_opc_slice ofd_common_slice[] = {
2839 .tos_opc_start = OST_FIRST_OPC,
2840 .tos_opc_end = OST_LAST_OPC,
2841 .tos_hs = ofd_tgt_handlers
2844 .tos_opc_start = OBD_FIRST_OPC,
2845 .tos_opc_end = OBD_LAST_OPC,
2846 .tos_hs = tgt_obd_handlers
2849 .tos_opc_start = LDLM_FIRST_OPC,
2850 .tos_opc_end = LDLM_LAST_OPC,
2851 .tos_hs = tgt_dlm_handlers
2854 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2855 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2856 .tos_hs = tgt_out_handlers
2859 .tos_opc_start = SEQ_FIRST_OPC,
2860 .tos_opc_end = SEQ_LAST_OPC,
2861 .tos_hs = seq_handlers
2864 .tos_opc_start = LFSCK_FIRST_OPC,
2865 .tos_opc_end = LFSCK_LAST_OPC,
2866 .tos_hs = tgt_lfsck_handlers
2869 .tos_opc_start = SEC_FIRST_OPC,
2870 .tos_opc_end = SEC_LAST_OPC,
2871 .tos_hs = tgt_sec_ctx_handlers
2878 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2879 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2882 * Implementation of lu_context_key::lct_key_exit.
2884 * Optional method called on lu_context_exit() for all allocated
2886 * It is used in OFD to sanitize context values which may be re-used
2887 * during another request processing by the same thread.
2889 * \param[in] ctx execution context
2890 * \param[in] key context key
2891 * \param[in] data ofd_thread_info
2893 static void ofd_key_exit(const struct lu_context *ctx,
2894 struct lu_context_key *key, void *data)
2896 struct ofd_thread_info *info = data;
2898 info->fti_env = NULL;
2899 info->fti_exp = NULL;
2902 info->fti_pre_version = 0;
2904 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2907 struct lu_context_key ofd_thread_key = {
2908 .lct_tags = LCT_DT_THREAD,
2909 .lct_init = ofd_key_init,
2910 .lct_fini = ofd_key_fini,
2911 .lct_exit = ofd_key_exit
2915 * Initialize OFD device according to parameters in the config log \a cfg.
2917 * This is the main starting point of OFD initialization. It fills all OFD
2918 * parameters with their initial values and calls other initializing functions
2919 * to set up all OFD subsystems.
2921 * \param[in] env execution environment
2922 * \param[in] m OFD device
2923 * \param[in] ldt LU device type of OFD
2924 * \param[in] cfg configuration log
2926 * \retval 0 if successful
2927 * \retval negative value on error
2929 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2930 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2932 const char *dev = lustre_cfg_string(cfg, 0);
2933 struct ofd_thread_info *info = NULL;
2934 struct obd_device *obd;
2935 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2937 struct nm_config_file *nodemap_config;
2938 struct obd_device_target *obt;
2944 obd = class_name2obd(dev);
2946 CERROR("Cannot find obd with name %s\n", dev);
2950 rc = lu_env_refill((struct lu_env *)env);
2955 obt->obt_magic = OBT_MAGIC;
2957 spin_lock_init(&m->ofd_flags_lock);
2958 m->ofd_raid_degraded = 0;
2959 m->ofd_checksum_t10pi_enforce = 0;
2960 m->ofd_sync_journal = 0;
2962 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2964 m->ofd_seq_count = 0;
2965 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2966 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2967 spin_lock_init(&m->ofd_inconsistency_lock);
2969 m->ofd_access_log_mask = -1; /* Log all accesses if enabled. */
2971 spin_lock_init(&m->ofd_batch_lock);
2972 init_rwsem(&m->ofd_lastid_rwsem);
2974 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2975 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2976 /* set this lu_device to obd, because error handling need it */
2977 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2979 /* No connection accepted until configurations will finish */
2980 spin_lock(&obd->obd_dev_lock);
2981 obd->obd_no_conn = 1;
2982 spin_unlock(&obd->obd_dev_lock);
2983 obd->obd_replayable = 1;
2984 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2985 char *str = lustre_cfg_string(cfg, 4);
2987 if (strchr(str, 'n')) {
2988 CWARN("%s: recovery disabled\n", obd->obd_name);
2989 obd->obd_replayable = 0;
2993 info = ofd_info_init(env, NULL);
2997 rc = ofd_stack_init(env, m, cfg, &lmd_flags);
2999 CERROR("%s: can't init device stack, rc %d\n",
3004 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
3005 ofd_procfs_add_brw_stats_symlink(m);
3008 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
3009 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
3010 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
3011 LDLM_NAMESPACE_SERVER,
3012 LDLM_NAMESPACE_GREEDY,
3014 if (m->ofd_namespace == NULL)
3015 GOTO(err_fini_stack, rc = -ENOMEM);
3016 /* set obd_namespace for compatibility with old code */
3017 obd->obd_namespace = m->ofd_namespace;
3018 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
3019 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
3020 m->ofd_namespace->ns_lvbp = m;
3022 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3023 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
3025 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
3026 OBD_FAIL_OST_ALL_REQUEST_NET,
3027 OBD_FAIL_OST_ALL_REPLY_NET);
3029 GOTO(err_free_ns, rc);
3031 if (lmd_flags & LMD_FLG_SKIP_LFSCK)
3032 m->ofd_skip_lfsck = 1;
3033 if (lmd_flags & LMD_FLG_LOCAL_RECOV)
3034 m->ofd_lut.lut_local_recovery = 1;
3036 rc = ofd_tunables_init(m);
3038 GOTO(err_fini_lut, rc);
3040 tgd->tgd_reserved_pcnt = 0;
3042 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
3043 m->ofd_cksum_types_supported =
3044 obd_cksum_types_supported_server(obd->obd_name);
3045 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
3046 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
3047 OFD_PRECREATE_SMALL_FS)
3048 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
3049 m->ofd_atime_diff = OFD_DEF_ATIME_DIFF;
3051 rc = ofd_fs_setup(env, m, obd);
3053 GOTO(err_fini_proc, rc);
3055 fid.f_seq = FID_SEQ_LOCAL_NAME;
3058 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
3061 GOTO(err_fini_fs, rc);
3063 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
3065 if (IS_ERR(nodemap_config)) {
3066 rc = PTR_ERR(nodemap_config);
3068 GOTO(err_fini_los, rc);
3070 obt->obt_nodemap_config_file = nodemap_config;
3073 rc = ofd_start_inconsistency_verification_thread(m);
3075 GOTO(err_fini_nm, rc);
3077 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
3082 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
3083 obt->obt_nodemap_config_file = NULL;
3085 local_oid_storage_fini(env, m->ofd_los);
3088 ofd_fs_cleanup(env, m);
3092 tgt_fini(env, &m->ofd_lut);
3094 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3095 obd->obd_namespace = m->ofd_namespace = NULL;
3097 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3102 * Stop the OFD device
3104 * This function stops the OFD device and all its subsystems.
3105 * This is the end of OFD lifecycle.
3107 * \param[in] env execution environment
3108 * \param[in] m OFD device
3110 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3112 struct obd_device *obd = ofd_obd(m);
3113 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3114 struct lfsck_stop stop;
3116 stop.ls_status = LS_PAUSED;
3118 lfsck_stop(env, m->ofd_osd, &stop);
3119 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3120 target_recovery_fini(obd);
3121 if (m->ofd_namespace != NULL)
3122 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3123 d->ld_obd->obd_force);
3125 obd_exports_barrier(obd);
3126 obd_zombie_barrier();
3129 tgt_fini(env, &m->ofd_lut);
3130 ofd_stop_inconsistency_verification_thread(m);
3131 lfsck_degister(env, m->ofd_osd);
3132 ofd_fs_cleanup(env, m);
3133 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3134 obd->u.obt.obt_nodemap_config_file = NULL;
3136 if (m->ofd_namespace != NULL) {
3137 ldlm_namespace_free_post(m->ofd_namespace);
3138 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3141 ofd_access_log_delete(m->ofd_access_log);
3142 m->ofd_access_log = NULL;
3144 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3146 LASSERT(atomic_read(&d->ld_ref) == 0);
3147 server_put_mount(obd->obd_name, true);
3152 * Implementation of lu_device_type_operations::ldto_device_fini.
3154 * Finalize device. Dual to ofd_device_init(). It is called from
3155 * obd_precleanup() and stops the current device.
3157 * \param[in] env execution environment
3158 * \param[in] d LU device of OFD
3162 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3163 struct lu_device *d)
3166 ofd_fini(env, ofd_dev(d));
3171 * Implementation of lu_device_type_operations::ldto_device_free.
3173 * Free OFD device. Dual to ofd_device_alloc().
3175 * \param[in] env execution environment
3176 * \param[in] d LU device of OFD
3180 static struct lu_device *ofd_device_free(const struct lu_env *env,
3181 struct lu_device *d)
3183 struct ofd_device *m = ofd_dev(d);
3185 dt_device_fini(&m->ofd_dt_dev);
3191 * Implementation of lu_device_type_operations::ldto_device_alloc.
3193 * This function allocates the new OFD device. It is called from
3194 * obd_setup() if OBD device had lu_device_type defined.
3196 * \param[in] env execution environment
3197 * \param[in] t lu_device_type of OFD device
3198 * \param[in] cfg configuration log
3200 * \retval pointer to the lu_device of just allocated OFD
3201 * \retval ERR_PTR of return value on error
3203 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3204 struct lu_device_type *t,
3205 struct lustre_cfg *cfg)
3207 struct ofd_device *m;
3208 struct lu_device *l;
3213 return ERR_PTR(-ENOMEM);
3215 l = &m->ofd_dt_dev.dd_lu_dev;
3216 dt_device_init(&m->ofd_dt_dev, t);
3217 rc = ofd_init0(env, m, t, cfg);
3219 ofd_device_free(env, l);
3226 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3227 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3229 static struct lu_device_type_operations ofd_device_type_ops = {
3230 .ldto_init = ofd_type_init,
3231 .ldto_fini = ofd_type_fini,
3233 .ldto_start = ofd_type_start,
3234 .ldto_stop = ofd_type_stop,
3236 .ldto_device_alloc = ofd_device_alloc,
3237 .ldto_device_free = ofd_device_free,
3238 .ldto_device_fini = ofd_device_fini
3241 static struct lu_device_type ofd_device_type = {
3242 .ldt_tags = LU_DEVICE_DT,
3243 .ldt_name = LUSTRE_OST_NAME,
3244 .ldt_ops = &ofd_device_type_ops,
3245 .ldt_ctx_tags = LCT_DT_THREAD
3249 * Initialize OFD module.
3251 * This function is called upon module loading. It registers OFD device type
3252 * and prepares all in-memory structures used by all OFD devices.
3254 * \retval 0 if successful
3255 * \retval negative value on error
3257 static int __init ofd_init(void)
3261 rc = lu_kmem_init(ofd_caches);
3265 rc = ofd_access_log_module_init();
3269 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3270 LUSTRE_OST_NAME, &ofd_device_type);
3272 goto out_ofd_access_log;
3277 ofd_access_log_module_exit();
3279 lu_kmem_fini(ofd_caches);
3287 * This function is called upon OFD module unloading.
3288 * It frees all related structures and unregisters OFD device type.
3290 static void __exit ofd_exit(void)
3292 class_unregister_type(LUSTRE_OST_NAME);
3293 ofd_access_log_module_exit();
3294 lu_kmem_fini(ofd_caches);
3297 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3298 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3299 MODULE_VERSION(LUSTRE_VERSION_STRING);
3300 MODULE_LICENSE("GPL");
3302 module_init(ofd_init);
3303 module_exit(ofd_exit);