4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/ofd/ofd_dev.c
34 * This file contains OSD API methods for OBD Filter Device (OFD),
35 * request handlers and supplemental functions to set OFD up and clean it up.
37 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
38 * Author: Mike Pershin <mike.pershin@intel.com>
39 * Author: Johann Lombardi <johann.lombardi@intel.com>
42 * The OBD Filter Device (OFD) module belongs to the Object Storage
43 * Server stack and connects the RPC oriented Unified Target (TGT)
44 * layer (see lustre/include/lu_target.h) to the storage oriented OSD
45 * layer (see Documentation/osd-api.txt).
53 * OFD implements the LU and OBD device APIs and is responsible for:
55 * - Handling client requests (create, destroy, bulk IO, setattr,
56 * get_info, set_info, statfs) for the objects belonging to the OST
57 * (together with TGT).
59 * - Providing grant space management which allows clients to reserve
60 * disk space for data writeback. OFD tracks grants on global and
63 * - Handling object precreation requests from MDTs.
65 * - Operating the LDLM service that allows clients to maintain object
66 * data cache coherence.
69 #define DEBUG_SUBSYSTEM S_FILTER
71 #include <obd_class.h>
72 #include <obd_cksum.h>
73 #include <uapi/linux/lustre/lustre_param.h>
74 #include <lustre_fid.h>
75 #include <lustre_lfsck.h>
76 #include <lustre_dlm.h>
77 #include <lustre_quota.h>
78 #include <lustre_nodemap.h>
79 #include <lustre_log.h>
81 #include "ofd_internal.h"
83 /* Slab for OFD object allocation */
84 static struct kmem_cache *ofd_object_kmem;
86 static struct lu_kmem_descr ofd_caches[] = {
88 .ckd_cache = &ofd_object_kmem,
89 .ckd_name = "ofd_obj",
90 .ckd_size = sizeof(struct ofd_object)
98 * Connect OFD to the next device in the stack.
100 * This function is used for device stack configuration and links OFD
101 * device with bottom OSD device.
103 * \param[in] env execution environment
104 * \param[in] m OFD device
105 * \param[in] next name of next device in the stack
106 * \param[out] exp export to return
108 * \retval 0 and export in \a exp if successful
109 * \retval negative value on error
111 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
112 const char *next, struct obd_export **exp)
114 struct obd_connect_data *data = NULL;
115 struct obd_device *obd;
121 GOTO(out, rc = -ENOMEM);
123 obd = class_name2obd(next);
125 CERROR("%s: can't locate next device: %s\n",
127 GOTO(out, rc = -ENOTCONN);
130 data->ocd_connect_flags = OBD_CONNECT_VERSION;
131 data->ocd_version = LUSTRE_VERSION_CODE;
133 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
135 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
136 ofd_name(m), next, rc);
140 m->ofd_dt_dev.dd_lu_dev.ld_site =
141 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
142 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
143 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
144 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
153 * Initialize stack of devices.
155 * This function initializes OFD-OSD device stack to serve OST requests
157 * \param[in] env execution environment
158 * \param[in] m OFD device
159 * \param[in] cfg Lustre config for this server
161 * \retval 0 if successful
162 * \retval negative value on error
164 static int ofd_stack_init(const struct lu_env *env,
165 struct ofd_device *m, struct lustre_cfg *cfg)
167 const char *dev = lustre_cfg_string(cfg, 0);
169 struct ofd_thread_info *info = ofd_info(env);
170 struct lustre_mount_info *lmi;
171 struct lustre_mount_data *lmd;
177 lmi = server_get_mount(dev);
179 CERROR("Cannot get mount info for %s!\n", dev);
183 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
184 if (lmd != NULL && lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
185 m->ofd_skip_lfsck = 1;
187 /* find bottom osd */
188 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
192 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
193 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
194 OBD_FREE(osdname, MTI_NAME_MAXLEN);
198 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
200 m->ofd_osd = lu2dt_dev(d);
202 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
203 "%s-osd", lustre_cfg_string(cfg, 0));
209 * Finalize the device stack OFD-OSD.
211 * This function cleans OFD-OSD device stack and
212 * disconnects OFD from the OSD.
214 * \param[in] env execution environment
215 * \param[in] m OFD device
216 * \param[in] top top device of stack
218 * \retval 0 if successful
219 * \retval negative value on error
221 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
222 struct lu_device *top)
224 struct obd_device *obd = ofd_obd(m);
225 struct lustre_cfg_bufs bufs;
226 struct lustre_cfg *lcfg;
231 lu_site_purge(env, top->ld_site, ~0);
232 /* process cleanup, pass mdt obd name to get obd umount flags */
233 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
238 lustre_cfg_bufs_set_string(&bufs, 1, flags);
239 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
242 lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs);
245 top->ld_ops->ldo_process_config(env, top, lcfg);
246 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
248 if (m->ofd_los != NULL) {
249 local_oid_storage_fini(env, m->ofd_los);
253 lu_site_purge(env, top->ld_site, ~0);
254 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
255 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
256 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
259 LASSERT(m->ofd_osd_exp);
260 obd_disconnect(m->ofd_osd_exp);
265 static void ofd_stack_pre_fini(const struct lu_env *env, struct ofd_device *m,
266 struct lu_device *top)
268 struct lustre_cfg_bufs bufs;
269 struct lustre_cfg *lcfg;
274 lustre_cfg_bufs_reset(&bufs, ofd_name(m));
275 lustre_cfg_bufs_set_string(&bufs, 1, NULL);
276 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
278 CERROR("%s: failed to trigger LCFG_PRE_CLEANUP\n", ofd_name(m));
280 lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, &bufs);
281 top->ld_ops->ldo_process_config(env, top, lcfg);
282 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
283 lcfg->lcfg_buflens));
289 /* For interoperability, see mdt_interop_param[]. */
290 static struct cfg_interop_param ofd_interop_param[] = {
291 { "ost.quota_type", NULL },
296 * Check if parameters are symlinks to the OSD.
298 * Some parameters were moved from ofd to osd and only their
299 * symlinks were kept in ofd by LU-3106. They are:
300 * -writehthrough_cache_enable
301 * -readcache_max_filesize
305 * Since they are not included by the static lprocfs var list, a pre-check
306 * is added for them to avoid "unknown param" errors. If they are matched
307 * in this check, they will be passed to the OSD directly.
309 * \param[in] param parameters to check
311 * \retval true if param is symlink to OSD param
314 static bool match_symlink_param(char *param)
319 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
320 sval = strchr(param, '=');
322 paramlen = sval - param;
323 if (strncmp(param, "writethrough_cache_enable",
325 strncmp(param, "readcache_max_filesize",
327 strncmp(param, "read_cache_enable",
329 strncmp(param, "brw_stats", paramlen) == 0)
338 * Process various configuration parameters.
340 * This function is used by MGS to process specific configurations and
341 * pass them through to the next device in server stack, i.e. the OSD.
343 * \param[in] env execution environment
344 * \param[in] d LU device of OFD
345 * \param[in] cfg parameters to process
347 * \retval 0 if successful
348 * \retval negative value on error
350 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
351 struct lustre_cfg *cfg)
353 struct ofd_device *m = ofd_dev(d);
354 struct dt_device *dt_next = m->ofd_osd;
355 struct lu_device *next = &dt_next->dd_lu_dev;
360 switch (cfg->lcfg_command) {
362 struct obd_device *obd = ofd_obd(m);
363 /* For interoperability */
364 struct cfg_interop_param *ptr = NULL;
365 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 rc = class_process_proc_param(PARAM_OST, obd->obd_vars, cfg,
402 if (rc > 0 || rc == -ENOSYS) {
403 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
405 /* we don't understand; pass it on */
406 rc = next->ld_ops->ldo_process_config(env, next, cfg);
410 case LCFG_SPTLRPC_CONF: {
415 /* others are passed further */
416 rc = next->ld_ops->ldo_process_config(env, next, cfg);
423 * Implementation of lu_object_operations::loo_object_init for OFD
425 * Allocate just the next object (OSD) in stack.
427 * \param[in] env execution environment
428 * \param[in] o lu_object of OFD object
429 * \param[in] conf additional configuration parameters, not used here
431 * \retval 0 if successful
432 * \retval negative value on error
434 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
435 const struct lu_object_conf *conf)
437 struct ofd_device *d = ofd_dev(o->lo_dev);
438 struct lu_device *under;
439 struct lu_object *below;
444 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
445 PFID(lu_object_fid(o)));
447 under = &d->ofd_osd->dd_lu_dev;
448 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
450 lu_object_add(o, below);
458 * Implementation of lu_object_operations::loo_object_free.
460 * Finish OFD object lifecycle and free its memory.
462 * \param[in] env execution environment
463 * \param[in] o LU object of OFD object
465 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
467 struct ofd_object *of = ofd_obj(o);
468 struct lu_object_header *h;
473 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
474 PFID(lu_object_fid(o)));
477 lu_object_header_fini(h);
478 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
483 * Implementation of lu_object_operations::loo_object_print.
485 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
486 * LU_OBJECT_DEBUG() for more details about the compound object printing.
488 * \param[in] env execution environment
489 * \param[in] cookie opaque data passed to the printer function
490 * \param[in] p printer function to use
491 * \param[in] o LU object of OFD object
493 * \retval 0 if successful
494 * \retval negative value on error
496 static int ofd_object_print(const struct lu_env *env, void *cookie,
497 lu_printer_t p, const struct lu_object *o)
499 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
502 static struct lu_object_operations ofd_obj_ops = {
503 .loo_object_init = ofd_object_init,
504 .loo_object_free = ofd_object_free,
505 .loo_object_print = ofd_object_print
509 * Implementation of lu_device_operations::lod_object_alloc.
511 * This function allocates OFD part of compound OFD-OSD object and
512 * initializes its header, because OFD is the top device in stack
514 * \param[in] env execution environment
515 * \param[in] hdr object header, NULL for OFD
516 * \param[in] d lu_device
518 * \retval allocated object if successful
519 * \retval NULL value on failed allocation
521 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
522 const struct lu_object_header *hdr,
525 struct ofd_object *of;
529 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
532 struct lu_object_header *h;
534 o = &of->ofo_obj.do_lu;
536 lu_object_header_init(h);
537 lu_object_init(o, h, d);
538 lu_object_add_top(h, o);
539 o->lo_ops = &ofd_obj_ops;
547 * Return the result of LFSCK run to the OFD.
549 * Notify OFD about result of LFSCK run. That may block the new object
550 * creation until problem is fixed by LFSCK.
552 * \param[in] env execution environment
553 * \param[in] data pointer to the OFD device
554 * \param[in] event LFSCK event type
556 * \retval 0 if successful
557 * \retval negative value on unknown event
559 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
560 enum lfsck_events event)
562 struct ofd_device *ofd = data;
563 struct obd_device *obd = ofd_obd(ofd);
566 case LE_LASTID_REBUILDING:
567 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
568 "on the device until the LAST_ID rebuilt successfully.\n",
570 down_write(&ofd->ofd_lastid_rwsem);
571 ofd->ofd_lastid_rebuilding = 1;
572 up_write(&ofd->ofd_lastid_rwsem);
574 case LE_LASTID_REBUILT: {
575 down_write(&ofd->ofd_lastid_rwsem);
576 ofd_seqs_free(env, ofd);
577 ofd->ofd_lastid_rebuilding = 0;
578 ofd->ofd_lastid_gen++;
579 up_write(&ofd->ofd_lastid_rwsem);
580 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
585 CERROR("%s: unknown lfsck event: rc = %d\n",
586 ofd_name(ofd), event);
594 * Implementation of lu_device_operations::ldo_prepare.
596 * This method is called after layer has been initialized and before it starts
597 * serving user requests. In OFD it starts lfsk check routines and initializes
600 * \param[in] env execution environment
601 * \param[in] pdev higher device in stack, NULL for OFD
602 * \param[in] dev lu_device of OFD device
604 * \retval 0 if successful
605 * \retval negative value on error
607 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
608 struct lu_device *dev)
610 struct ofd_thread_info *info;
611 struct ofd_device *ofd = ofd_dev(dev);
612 struct obd_device *obd = ofd_obd(ofd);
613 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
618 info = ofd_info_init(env, NULL);
622 /* initialize lower device */
623 rc = next->ld_ops->ldo_prepare(env, dev, next);
627 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
628 ofd_lfsck_out_notify, ofd, false);
630 CERROR("%s: failed to initialize lfsck: rc = %d\n",
635 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
636 /* The LFSCK instance is registered just now, so it must be there when
637 * register the namespace to such instance. */
638 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
640 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
641 LASSERT(obd->obd_no_conn);
642 spin_lock(&obd->obd_dev_lock);
643 obd->obd_no_conn = 0;
644 spin_unlock(&obd->obd_dev_lock);
646 if (obd->obd_recovering == 0)
647 ofd_postrecov(env, ofd);
653 * Implementation of lu_device_operations::ldo_recovery_complete.
655 * This method notifies all layers about 'recovery complete' event. That means
656 * device is in full state and consistent. An OFD calculates available grant
657 * space upon this event.
659 * \param[in] env execution environment
660 * \param[in] dev lu_device of OFD device
662 * \retval 0 if successful
663 * \retval negative value on error
665 static int ofd_recovery_complete(const struct lu_env *env,
666 struct lu_device *dev)
668 struct ofd_thread_info *oti = ofd_info(env);
669 struct ofd_device *ofd = ofd_dev(dev);
670 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
676 * Grant space for object precreation on the self export.
677 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
678 * is enough to create 10k objects. More space is then acquired for
679 * precreation in tgt_grant_create().
681 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
682 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
683 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
684 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
685 OBD_CONNECT_GRANT_PARAM;
686 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
688 rc = next->ld_ops->ldo_recovery_complete(env, next);
693 * lu_device_operations matrix for OFD device.
695 static struct lu_device_operations ofd_lu_ops = {
696 .ldo_object_alloc = ofd_object_alloc,
697 .ldo_process_config = ofd_process_config,
698 .ldo_recovery_complete = ofd_recovery_complete,
699 .ldo_prepare = ofd_prepare,
702 LPROC_SEQ_FOPS(lprocfs_nid_stats_clear);
705 * Initialize all needed procfs entries for OFD device.
707 * \param[in] ofd OFD device
709 * \retval 0 if successful
710 * \retval negative value on error
712 static int ofd_procfs_init(struct ofd_device *ofd)
714 struct obd_device *obd = ofd_obd(ofd);
715 struct proc_dir_entry *entry;
720 /* lprocfs must be setup before the ofd so state can be safely added
721 * to /proc incrementally as the ofd is setup */
722 obd->obd_vars = lprocfs_ofd_obd_vars;
723 rc = lprocfs_obd_setup(obd, false);
725 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
730 rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
732 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
734 GOTO(obd_cleanup, rc);
737 entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
740 CERROR("%s: error %d setting up lprocfs for %s\n",
741 obd->obd_name, rc, "exports");
742 GOTO(obd_cleanup, rc);
744 obd->obd_proc_exports_entry = entry;
746 entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
747 obd, &lprocfs_nid_stats_clear_fops);
750 CERROR("%s: add proc entry 'clear' failed: %d.\n",
752 GOTO(obd_cleanup, rc);
755 ofd_stats_counter_init(obd->obd_stats);
757 rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
758 ofd_stats_counter_init);
760 GOTO(obd_cleanup, rc);
763 lprocfs_obd_cleanup(obd);
764 lprocfs_free_obd_stats(obd);
770 * Expose OSD statistics to OFD layer.
772 * The osd interfaces to the backend file system exposes useful data
773 * such as brw_stats and read or write cache states. This same data
774 * needs to be exposed into the obdfilter (ofd) layer to maintain
775 * backwards compatibility. This function creates the symlinks in the
776 * proc layer to enable this.
778 * \param[in] ofd OFD device
780 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
782 struct obd_device *obd = ofd_obd(ofd);
783 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
785 if (obd->obd_proc_entry == NULL)
788 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
789 "../../%s/%s/brw_stats",
790 osd_obd->obd_type->typ_name, obd->obd_name);
792 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
793 "../../%s/%s/read_cache_enable",
794 osd_obd->obd_type->typ_name, obd->obd_name);
796 lprocfs_add_symlink("readcache_max_filesize",
798 "../../%s/%s/readcache_max_filesize",
799 osd_obd->obd_type->typ_name, obd->obd_name);
801 lprocfs_add_symlink("writethrough_cache_enable",
803 "../../%s/%s/writethrough_cache_enable",
804 osd_obd->obd_type->typ_name, obd->obd_name);
808 * Cleanup all procfs entries in OFD.
810 * \param[in] ofd OFD device
812 static void ofd_procfs_fini(struct ofd_device *ofd)
814 struct obd_device *obd = ofd_obd(ofd);
816 lprocfs_free_per_client_stats(obd);
817 lprocfs_obd_cleanup(obd);
818 lprocfs_free_obd_stats(obd);
819 lprocfs_job_stats_fini(obd);
823 * Stop SEQ/FID server on OFD.
825 * \param[in] env execution environment
826 * \param[in] ofd OFD device
828 * \retval 0 if successful
829 * \retval negative value on error
831 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
833 return seq_site_fini(env, &ofd->ofd_seq_site);
837 * Start SEQ/FID server on OFD.
839 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
840 * It also connects to the master server to get own FID sequence (SEQ) range
841 * to this particular OFD. Typically that happens when the OST is first
842 * formatted or in the rare case that it exhausts the local sequence range.
844 * The sequence range is allocated out to the MDTs for OST object allocations,
845 * and not directly to the clients.
847 * \param[in] env execution environment
848 * \param[in] ofd OFD device
850 * \retval 0 if successful
851 * \retval negative value on error
853 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
855 struct seq_server_site *ss = &ofd->ofd_seq_site;
856 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
857 char *obd_name = ofd_name(ofd);
861 ss = &ofd->ofd_seq_site;
862 lu->ld_site->ld_seq_site = ss;
863 ss->ss_lu = lu->ld_site;
864 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
866 OBD_ALLOC(name, sizeof(obd_name) * 2 + 10);
870 OBD_ALLOC_PTR(ss->ss_server_seq);
871 if (ss->ss_server_seq == NULL)
872 GOTO(out_name, rc = -ENOMEM);
874 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
875 LUSTRE_SEQ_SERVER, ss);
877 CERROR("%s : seq server init error %d\n", obd_name, rc);
878 GOTO(out_server, rc);
880 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
882 OBD_ALLOC_PTR(ss->ss_client_seq);
883 if (ss->ss_client_seq == NULL)
884 GOTO(out_server, rc = -ENOMEM);
887 * It always printed as "%p", so that the name is unique in the kernel,
888 * even if the filesystem is mounted twice. So sizeof(.) * 2 is enough.
890 snprintf(name, sizeof(obd_name) * 2 + 7, "%p-super", obd_name);
891 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
894 CERROR("%s : seq client init error %d\n", obd_name, rc);
895 GOTO(out_client, rc);
898 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
902 seq_client_fini(ss->ss_client_seq);
903 OBD_FREE_PTR(ss->ss_client_seq);
904 ss->ss_client_seq = NULL;
906 seq_server_fini(ss->ss_server_seq, env);
907 OBD_FREE_PTR(ss->ss_server_seq);
908 ss->ss_server_seq = NULL;
911 OBD_FREE(name, sizeof(obd_name) * 2 + 10);
917 * OFD request handler for OST_SET_INFO RPC.
919 * This is OFD-specific part of request handling
921 * \param[in] tsi target session environment for this request
923 * \retval 0 if successful
924 * \retval negative value on error
926 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
928 struct ptlrpc_request *req = tgt_ses_req(tsi);
929 struct ost_body *body = NULL, *repbody;
930 void *key, *val = NULL;
931 int keylen, vallen, rc = 0;
932 bool is_grant_shrink;
936 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
938 DEBUG_REQ(D_HA, req, "no set_info key");
939 RETURN(err_serious(-EFAULT));
941 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
944 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
946 DEBUG_REQ(D_HA, req, "no set_info val");
947 RETURN(err_serious(-EFAULT));
949 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
952 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
954 /* In this case the value is actually an RMF_OST_BODY, so we
955 * transmutate the type of this PTLRPC */
956 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
958 rc = req_capsule_server_pack(tsi->tsi_pill);
962 if (is_grant_shrink) {
963 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
965 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
968 /** handle grant shrink, similar to a read request */
969 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
971 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
973 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
976 CERROR("%s: Unsupported key %s\n",
977 tgt_name(tsi->tsi_tgt), (char *)key);
980 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
987 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
989 * This function returns a list of extents which describes how a file's
990 * blocks are laid out on the disk.
992 * \param[in] env execution environment
993 * \param[in] ofd OFD device
994 * \param[in] fid FID of object
995 * \param[in] fiemap fiemap structure to fill with data
997 * \retval 0 if \a fiemap is filled with data successfully
998 * \retval negative value on error
1000 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
1001 struct lu_fid *fid, struct fiemap *fiemap)
1003 struct ofd_object *fo;
1006 fo = ofd_object_find(env, ofd, fid);
1008 CERROR("%s: error finding object "DFID"\n",
1009 ofd_name(ofd), PFID(fid));
1013 ofd_read_lock(env, fo);
1014 if (ofd_object_exists(fo))
1015 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
1018 ofd_read_unlock(env, fo);
1019 ofd_object_put(env, fo);
1024 static int ofd_lock_unlock_region(struct ldlm_namespace *ns,
1025 struct ldlm_res_id *res_id,
1026 unsigned long long begin,
1027 unsigned long long end)
1031 struct lustre_handle lh = { 0 };
1033 LASSERT(begin <= end);
1035 rc = tgt_extent_lock(ns, res_id, begin, end, &lh, LCK_PR, &flags);
1039 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
1040 tgt_extent_unlock(&lh, LCK_PR);
1046 * Lock the sparse areas of given resource.
1048 * The locking of sparse areas will cause dirty data to be flushed back from
1049 * clients. This is used when getting the FIEMAP of an object to make sure
1050 * there is no unaccounted cached data on clients.
1052 * This function goes through \a fiemap list of extents and locks only sparse
1053 * areas between extents.
1055 * \param[in] ns LDLM namespace
1056 * \param[in] res_id resource ID
1057 * \param[in] fiemap file extents mapping on disk
1058 * \param[in] locked list head of regions list
1060 * \retval 0 if successful
1061 * \retval negative value on error
1063 static int lock_zero_regions(struct ldlm_namespace *ns,
1064 struct ldlm_res_id *res_id,
1065 struct fiemap *fiemap)
1067 __u64 begin = fiemap->fm_start;
1070 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1074 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1075 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1076 if (fiemap_start[i].fe_logical > begin) {
1077 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1078 begin, fiemap_start[i].fe_logical);
1079 rc = ofd_lock_unlock_region(ns, res_id, begin,
1080 fiemap_start[i].fe_logical);
1085 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1088 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1089 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1090 begin, fiemap->fm_start + fiemap->fm_length);
1091 rc = ofd_lock_unlock_region(ns, res_id, begin,
1092 fiemap->fm_start + fiemap->fm_length);
1100 * OFD request handler for OST_GET_INFO RPC.
1102 * This is OFD-specific part of request handling. The OFD-specific keys are:
1103 * - KEY_LAST_ID (obsolete)
1107 * This function reads needed data from storage and fills reply with it.
1109 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1110 * and is kept for compatibility.
1112 * \param[in] tsi target session environment for this request
1114 * \retval 0 if successful
1115 * \retval negative value on error
1117 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1119 struct obd_export *exp = tsi->tsi_exp;
1120 struct ofd_device *ofd = ofd_exp(exp);
1121 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1124 int replylen, rc = 0;
1128 /* this common part for get_info rpc */
1129 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1131 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1132 RETURN(err_serious(-EPROTO));
1134 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1137 if (KEY_IS(KEY_LAST_ID)) {
1139 struct ofd_seq *oseq;
1141 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1142 rc = req_capsule_server_pack(tsi->tsi_pill);
1144 RETURN(err_serious(rc));
1146 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1148 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1149 (u64)exp->exp_filter_data.fed_group);
1153 *last_id = ofd_seq_last_oid(oseq);
1154 ofd_seq_put(tsi->tsi_env, oseq);
1155 } else if (KEY_IS(KEY_FIEMAP)) {
1156 struct ll_fiemap_info_key *fm_key;
1157 struct fiemap *fiemap;
1160 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1162 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1163 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1165 RETURN(err_serious(rc));
1167 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1169 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1171 replylen = fiemap_count_to_size(
1172 fm_key->lfik_fiemap.fm_extent_count);
1173 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1174 RCL_SERVER, replylen);
1176 rc = req_capsule_server_pack(tsi->tsi_pill);
1178 RETURN(err_serious(rc));
1180 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1184 *fiemap = fm_key->lfik_fiemap;
1185 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1187 /* LU-3219: Lock the sparse areas to make sure dirty
1188 * flushed back from client, then call fiemap again. */
1189 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1190 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1191 ost_fid_build_resid(fid, &fti->fti_resid);
1192 rc = lock_zero_regions(ofd->ofd_namespace,
1193 &fti->fti_resid, fiemap);
1195 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1198 } else if (KEY_IS(KEY_LAST_FID)) {
1199 struct ofd_device *ofd = ofd_exp(exp);
1200 struct ofd_seq *oseq;
1204 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1205 rc = req_capsule_server_pack(tsi->tsi_pill);
1207 RETURN(err_serious(rc));
1209 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1211 RETURN(err_serious(-EPROTO));
1213 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1215 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1219 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1220 ostid_seq(&fti->fti_ostid));
1222 RETURN(PTR_ERR(oseq));
1224 rc = ostid_to_fid(fid, &oseq->os_oi,
1225 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1229 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1232 ofd_seq_put(tsi->tsi_env, oseq);
1234 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1238 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1245 * OFD request handler for OST_GETATTR RPC.
1247 * This is OFD-specific part of request handling. It finds the OFD object
1248 * by its FID, gets attributes from storage and packs result to the reply.
1250 * \param[in] tsi target session environment for this request
1252 * \retval 0 if successful
1253 * \retval negative value on error
1255 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1257 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1258 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1259 struct ost_body *repbody;
1260 struct lustre_handle lh = { 0 };
1261 struct ofd_object *fo;
1263 enum ldlm_mode lock_mode = LCK_PR;
1268 LASSERT(tsi->tsi_ost_body != NULL);
1270 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1271 if (repbody == NULL)
1274 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1275 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1277 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1278 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1281 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1284 rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace,
1285 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1291 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1293 GOTO(out, rc = PTR_ERR(fo));
1295 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1299 obdo_from_la(&repbody->oa, &fti->fti_attr,
1300 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1302 /* Store object version in reply */
1303 curr_version = dt_version_get(tsi->tsi_env,
1304 ofd_object_child(fo));
1305 if ((__s64)curr_version != -EOPNOTSUPP) {
1306 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1307 repbody->oa.o_data_version = curr_version;
1310 if (fo->ofo_ff.ff_layout_version > 0) {
1311 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1312 repbody->oa.o_layout_version =
1313 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1315 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1316 PFID(&tsi->tsi_fid),
1317 repbody->oa.o_layout_version);
1321 ofd_object_put(tsi->tsi_env, fo);
1324 tgt_extent_unlock(&lh, lock_mode);
1326 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1329 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1330 repbody->oa.o_flags = OBD_FL_FLUSH;
1336 * OFD request handler for OST_SETATTR RPC.
1338 * This is OFD-specific part of request handling. It finds the OFD object
1339 * by its FID, sets attributes from request and packs result to the reply.
1341 * \param[in] tsi target session environment for this request
1343 * \retval 0 if successful
1344 * \retval negative value on error
1346 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1348 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1349 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1350 struct ost_body *body = tsi->tsi_ost_body;
1351 struct ost_body *repbody;
1352 struct ldlm_resource *res;
1353 struct ofd_object *fo;
1358 LASSERT(body != NULL);
1360 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1361 if (repbody == NULL)
1364 repbody->oa.o_oi = body->oa.o_oi;
1365 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1367 /* This would be very bad - accidentally truncating a file when
1368 * changing the time or similar - bug 12203. */
1369 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1370 body->oa.o_size != OBD_OBJECT_EOF) {
1371 static char mdsinum[48];
1373 if (body->oa.o_valid & OBD_MD_FLFID)
1374 snprintf(mdsinum, sizeof(mdsinum) - 1,
1375 "of parent "DFID, body->oa.o_parent_seq,
1376 body->oa.o_parent_oid, 0);
1380 CERROR("%s: setattr from %s is trying to truncate object "DFID
1381 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1382 PFID(&tsi->tsi_fid), mdsinum);
1386 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1388 GOTO(out, rc = PTR_ERR(fo));
1390 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1391 fti->fti_attr.la_valid &= ~LA_TYPE;
1393 /* setting objects attributes (including owner/group) */
1394 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1398 obdo_from_la(&repbody->oa, &fti->fti_attr,
1399 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1401 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1405 ofd_object_put(tsi->tsi_env, fo);
1408 /* we do not call this before to avoid lu_object_find() in
1409 * ->lvbo_update() holding another reference on the object.
1410 * otherwise concurrent destroy can make the object unavailable
1411 * for 2nd lu_object_find() waiting for the first reference
1412 * to go... deadlock! */
1413 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1414 &tsi->tsi_resid, LDLM_EXTENT, 0);
1416 ldlm_res_lvbo_update(res, NULL, 0);
1417 ldlm_resource_putref(res);
1424 * Destroy OST orphans.
1426 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1427 * set then we must destroy possible orphaned objects.
1429 * \param[in] env execution environment
1430 * \param[in] exp OBD export
1431 * \param[in] ofd OFD device
1432 * \param[in] oa obdo structure for reply
1434 * \retval 0 if successful
1435 * \retval negative value on error
1437 static int ofd_orphans_destroy(const struct lu_env *env,
1438 struct obd_export *exp,
1439 struct ofd_device *ofd, struct obdo *oa)
1441 struct ofd_thread_info *info = ofd_info(env);
1442 struct lu_fid *fid = &info->fti_fid;
1443 struct ost_id *oi = &oa->o_oi;
1444 struct ofd_seq *oseq;
1445 u64 seq = ostid_seq(oi);
1446 u64 end_id = ostid_id(oi);
1454 oseq = ofd_seq_get(ofd, seq);
1456 CERROR("%s: Can not find seq for "DOSTID"\n",
1457 ofd_name(ofd), POSTID(oi));
1462 last = ofd_seq_last_oid(oseq);
1465 LASSERT(exp != NULL);
1466 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1468 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1471 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1472 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1474 while (oid > end_id) {
1475 rc = fid_set_id(fid, oid);
1476 if (unlikely(rc != 0))
1479 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1480 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1481 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1482 /* this is pretty fatal... */
1483 CEMERG("%s: error destroying precreated id "
1485 ofd_name(ofd), PFID(fid), rc);
1489 ofd_seq_last_oid_set(oseq, oid);
1490 /* update last_id on disk periodically so that if we
1491 * restart * we don't need to re-scan all of the just
1492 * deleted objects. */
1493 if ((oid & 511) == 0)
1494 ofd_seq_last_oid_write(env, ofd, oseq);
1498 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1499 ofd_name(ofd), seq, oid);
1503 ofd_seq_last_oid_set(oseq, oid);
1504 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1506 /* don't reuse orphan object, return last used objid */
1507 rc = ostid_set_id(oi, last);
1513 ofd_seq_put(env, oseq);
1518 * OFD request handler for OST_CREATE RPC.
1520 * This is OFD-specific part of request handling. Its main purpose is to
1521 * create new data objects on OST, but it also used to destroy orphans.
1523 * \param[in] tsi target session environment for this request
1525 * \retval 0 if successful
1526 * \retval negative value on error
1528 static int ofd_create_hdl(struct tgt_session_info *tsi)
1530 struct ptlrpc_request *req = tgt_ses_req(tsi);
1531 struct ost_body *repbody;
1532 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1533 struct obdo *rep_oa;
1534 struct obd_export *exp = tsi->tsi_exp;
1535 struct ofd_device *ofd = ofd_exp(exp);
1536 u64 seq = ostid_seq(&oa->o_oi);
1537 u64 oid = ostid_id(&oa->o_oi);
1538 struct ofd_seq *oseq;
1546 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1549 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1550 if (repbody == NULL)
1553 down_read(&ofd->ofd_lastid_rwsem);
1554 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1555 * we may do that in the future.
1556 * Return -ENOSPC until the LAST_ID rebuilt. */
1557 if (unlikely(ofd->ofd_lastid_rebuilding))
1558 GOTO(out_sem, rc = -ENOSPC);
1560 rep_oa = &repbody->oa;
1561 rep_oa->o_oi = oa->o_oi;
1563 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1565 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1567 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1569 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1570 ofd_name(ofd), seq, PTR_ERR(oseq));
1571 GOTO(out_sem, rc = -EINVAL);
1574 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1575 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1576 if (!ofd_obd(ofd)->obd_recovering ||
1577 oid > ofd_seq_last_oid(oseq)) {
1578 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1579 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1580 ofd_seq_last_oid(oseq));
1581 GOTO(out_nolock, rc = -EINVAL);
1583 /* Do nothing here, we re-create objects during recovery
1584 * upon write replay, see ofd_preprw_write() */
1585 GOTO(out_nolock, rc = 0);
1587 /* former ofd_handle_precreate */
1588 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1589 (oa->o_flags & OBD_FL_DELORPHAN)) {
1590 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1592 /* destroy orphans */
1593 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1594 exp->exp_conn_cnt) {
1595 CERROR("%s: dropping old orphan cleanup request\n",
1597 GOTO(out_nolock, rc = 0);
1599 /* This causes inflight precreates to abort and drop lock */
1600 oseq->os_destroys_in_progress = 1;
1601 mutex_lock(&oseq->os_create_lock);
1602 if (!oseq->os_destroys_in_progress) {
1603 CERROR("%s:[%llu] destroys_in_progress already"
1604 " cleared\n", ofd_name(ofd), seq);
1605 rc = ostid_set_id(&rep_oa->o_oi,
1606 ofd_seq_last_oid(oseq));
1609 diff = oid - ofd_seq_last_oid(oseq);
1610 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1611 ofd_seq_last_oid(oseq), diff);
1612 if (-diff > OST_MAX_PRECREATE) {
1613 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1614 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1615 "(%llu), trust the OST\n",
1616 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1617 PFID(&oseq->os_oi.oi_fid),
1618 ofd_seq_last_oid(oseq));
1620 /* Let MDS know that we are so far ahead. */
1621 rc = ostid_set_id(&rep_oa->o_oi,
1622 ofd_seq_last_oid(oseq) + 1);
1623 } else if (diff < 0) {
1624 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1626 oseq->os_destroys_in_progress = 0;
1628 /* XXX: Used by MDS for the first time! */
1629 oseq->os_destroys_in_progress = 0;
1632 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1633 ofd->ofd_lastid_gen)) {
1634 /* Keep the export ref so we can send the reply. */
1635 ofd_obd_disconnect(class_export_get(exp));
1636 GOTO(out_nolock, rc = -ENOTCONN);
1639 mutex_lock(&oseq->os_create_lock);
1640 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1641 exp->exp_conn_cnt) {
1642 CERROR("%s: dropping old precreate request\n",
1646 /* only precreate if seq is 0, IDIF or normal and also o_id
1647 * must be specfied */
1648 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1649 !fid_seq_is_idif(seq)) || oid == 0) {
1650 diff = 1; /* shouldn't we create this right now? */
1652 diff = oid - ofd_seq_last_oid(oseq);
1653 /* Do sync create if the seq is about to used up */
1654 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1655 if (unlikely(oid >= IDIF_MAX_OID - 1))
1657 } else if (fid_seq_is_norm(seq)) {
1659 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1662 CERROR("%s : invalid o_seq "DOSTID"\n",
1663 ofd_name(ofd), POSTID(&oa->o_oi));
1664 GOTO(out, rc = -EINVAL);
1669 CERROR("%s: invalid precreate request for "
1670 DOSTID", last_id %llu. "
1671 "Likely MDS last_id corruption\n",
1672 ofd_name(ofd), POSTID(&oa->o_oi),
1673 ofd_seq_last_oid(oseq));
1674 GOTO(out, rc = -EINVAL);
1679 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1684 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1685 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1686 /* don't enforce grant during orphan recovery */
1687 granted = tgt_grant_create(tsi->tsi_env,
1688 ofd_obd(ofd)->obd_self_export,
1693 CDEBUG(D_HA, "%s: failed to acquire grant "
1694 "space for precreate (%lld): rc = %d\n",
1695 ofd_name(ofd), diff, rc);
1700 /* This can happen if a new OST is formatted and installed
1701 * in place of an old one at the same index. Instead of
1702 * precreating potentially millions of deleted old objects
1703 * (possibly filling the OST), only precreate the last batch.
1704 * LFSCK will eventually clean up any orphans. LU-14 */
1705 if (diff > 5 * OST_MAX_PRECREATE) {
1706 diff = OST_MAX_PRECREATE / 2;
1707 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1708 "OST replaced or reformatted: "
1709 "LFSCK will clean up",
1712 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1713 "%u larger than the LAST_ID "DOSTID", only "
1714 "precreating the last %lld objects.\n",
1715 ofd_name(ofd), POSTID(&oa->o_oi),
1716 5 * OST_MAX_PRECREATE,
1717 POSTID(&oseq->os_oi), diff);
1718 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1722 next_id = ofd_seq_last_oid(oseq) + 1;
1723 count = ofd_precreate_batch(ofd, (int)diff);
1725 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1726 " at %llu\n", ofd_name(ofd),
1727 count, seq, next_id);
1729 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1730 && ktime_get_seconds() > enough_time) {
1731 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1732 " created at a rate of %d/s\n",
1733 ofd_name(ofd), created, diff + created,
1734 created / DISK_TIMEOUT);
1738 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1739 oseq, count, sync_trans);
1743 } else if (rc < 0) {
1749 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1750 LCONSOLE_WARN("%s: can't create the same count of"
1751 " objects when replaying the request"
1752 " (diff is %lld). see LU-4621\n",
1753 ofd_name(ofd), diff);
1756 /* some objects got created, we can return
1757 * them, even if last creation failed */
1760 CERROR("%s: unable to precreate: rc = %d\n",
1763 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1764 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1765 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1770 rc = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1773 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1776 mutex_unlock(&oseq->os_create_lock);
1779 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1780 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1781 struct lu_fid *fid = &info->fti_fid;
1783 /* For compatible purpose, it needs to convert back to
1784 * OST ID before put it on wire. */
1785 *fid = rep_oa->o_oi.oi_fid;
1786 fid_to_ostid(fid, &rep_oa->o_oi);
1788 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1790 ofd_seq_put(tsi->tsi_env, oseq);
1793 up_read(&ofd->ofd_lastid_rwsem);
1798 * OFD request handler for OST_DESTROY RPC.
1800 * This is OFD-specific part of request handling. It destroys data objects
1801 * related to destroyed object on MDT.
1803 * \param[in] tsi target session environment for this request
1805 * \retval 0 if successful
1806 * \retval negative value on error
1808 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1810 const struct ost_body *body = tsi->tsi_ost_body;
1811 struct ost_body *repbody;
1812 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1813 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1814 struct lu_fid *fid = &fti->fti_fid;
1821 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1824 /* This is old case for clients before Lustre 2.4 */
1825 /* If there's a DLM request, cancel the locks mentioned in it */
1826 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1828 struct ldlm_request *dlm;
1830 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1833 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1836 *fid = body->oa.o_oi.oi_fid;
1837 oid = ostid_id(&body->oa.o_oi);
1840 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1842 /* check that o_misc makes sense */
1843 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1844 count = body->oa.o_misc;
1846 count = 1; /* default case - single destroy */
1848 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1849 POSTID(&body->oa.o_oi), count);
1854 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1855 if (lrc == -ENOENT) {
1857 "%s: destroying non-existent object "DFID"\n",
1858 ofd_name(ofd), PFID(fid));
1859 /* rewrite rc with -ENOENT only if it is 0 */
1862 } else if (lrc != 0) {
1863 CERROR("%s: error destroying object "DFID": %d\n",
1864 ofd_name(ofd), PFID(fid), lrc);
1870 lrc = fid_set_id(fid, oid);
1871 if (unlikely(lrc != 0 && count > 0))
1872 GOTO(out, rc = lrc);
1875 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1881 fid_to_ostid(fid, &repbody->oa.o_oi);
1886 * OFD request handler for OST_STATFS RPC.
1888 * This function gets statfs data from storage as part of request
1891 * \param[in] tsi target session environment for this request
1893 * \retval 0 if successful
1894 * \retval negative value on error
1896 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1898 struct obd_statfs *osfs;
1903 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1905 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1907 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1908 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1910 CERROR("%s: statfs failed: rc = %d\n",
1911 tgt_name(tsi->tsi_tgt), rc);
1913 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1916 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1923 * OFD request handler for OST_SYNC RPC.
1925 * Sync object data or all filesystem data to the disk and pack the
1928 * \param[in] tsi target session environment for this request
1930 * \retval 0 if successful
1931 * \retval negative value on error
1933 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1935 struct ost_body *body = tsi->tsi_ost_body;
1936 struct ost_body *repbody;
1937 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1938 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1939 struct ofd_object *fo = NULL;
1944 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1946 /* if no objid is specified, it means "sync whole filesystem" */
1947 if (!fid_is_zero(&tsi->tsi_fid)) {
1948 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1950 RETURN(PTR_ERR(fo));
1953 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1954 fo != NULL ? ofd_object_child(fo) : NULL,
1955 repbody->oa.o_size, repbody->oa.o_blocks);
1959 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1964 repbody->oa.o_oi = body->oa.o_oi;
1965 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1967 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1969 obdo_from_la(&repbody->oa, &fti->fti_attr,
1972 /* don't return rc from getattr */
1977 ofd_object_put(tsi->tsi_env, fo);
1982 * OFD request handler for OST_PUNCH RPC.
1984 * This is part of request processing. Validate request fields,
1985 * punch (truncate) the given OFD object and pack reply.
1987 * \param[in] tsi target session environment for this request
1989 * \retval 0 if successful
1990 * \retval negative value on error
1992 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1994 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1995 struct ost_body *repbody;
1996 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1997 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1998 struct ldlm_resource *res;
1999 struct ofd_object *fo;
2001 struct lustre_handle lh = { 0, };
2008 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
2010 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
2011 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
2013 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2014 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2015 RETURN(err_serious(-EPROTO));
2017 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2018 if (repbody == NULL)
2019 RETURN(err_serious(-ENOMEM));
2021 /* punch start,end are passed in o_size,o_blocks throught wire */
2025 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2028 /* standard truncate optimization: if file body is completely
2029 * destroyed, don't send data back to the server. */
2031 flags |= LDLM_FL_AST_DISCARD_DATA;
2033 repbody->oa.o_oi = oa->o_oi;
2034 repbody->oa.o_valid = OBD_MD_FLID;
2036 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2037 oa->o_flags & OBD_FL_SRVLOCK;
2040 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
2046 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2047 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2048 oa->o_valid, start, end);
2050 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2053 GOTO(out, rc = PTR_ERR(fo));
2055 la_from_obdo(&info->fti_attr, oa,
2056 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2057 info->fti_attr.la_size = start;
2058 info->fti_attr.la_valid |= LA_SIZE;
2060 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2065 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2069 ofd_object_put(tsi->tsi_env, fo);
2072 tgt_extent_unlock(&lh, LCK_PW);
2074 /* we do not call this before to avoid lu_object_find() in
2075 * ->lvbo_update() holding another reference on the object.
2076 * otherwise concurrent destroy can make the object unavailable
2077 * for 2nd lu_object_find() waiting for the first reference
2078 * to go... deadlock! */
2079 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2082 struct ost_lvb *res_lvb;
2084 ldlm_res_lvbo_update(res, NULL, 0);
2085 res_lvb = res->lr_lvb_data;
2086 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2087 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2088 ldlm_resource_putref(res);
2094 static int ofd_ladvise_prefetch(const struct lu_env *env,
2095 struct ofd_object *fo,
2096 struct niobuf_local *lnb,
2097 __u64 start, __u64 end, enum dt_bufs_type dbt)
2099 struct ofd_thread_info *info = ofd_info(env);
2100 pgoff_t start_index, end_index, pages;
2101 struct niobuf_remote rnb;
2102 unsigned long nr_local;
2108 ofd_read_lock(env, fo);
2109 if (!ofd_object_exists(fo))
2110 GOTO(out_unlock, rc = -ENOENT);
2112 rc = ofd_attr_get(env, fo, &info->fti_attr);
2114 GOTO(out_unlock, rc);
2116 if (end > info->fti_attr.la_size)
2117 end = info->fti_attr.la_size;
2120 GOTO(out_unlock, rc);
2122 /* We need page aligned offset and length */
2123 start_index = start >> PAGE_SHIFT;
2124 end_index = (end - 1) >> PAGE_SHIFT;
2125 pages = end_index - start_index + 1;
2127 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2128 PTLRPC_MAX_BRW_PAGES;
2129 rnb.rnb_offset = start_index << PAGE_SHIFT;
2130 rnb.rnb_len = nr_local << PAGE_SHIFT;
2131 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, dbt);
2132 if (unlikely(rc < 0))
2135 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2136 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2139 start_index += nr_local;
2144 ofd_read_unlock(env, fo);
2149 * OFD request handler for OST_LADVISE RPC.
2151 * Tune cache or perfetch policies according to advices.
2153 * \param[in] tsi target session environment for this request
2155 * \retval 0 if successful
2156 * \retval negative errno on error
2158 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2160 struct ptlrpc_request *req = tgt_ses_req(tsi);
2161 struct obd_export *exp = tsi->tsi_exp;
2162 struct ofd_device *ofd = ofd_exp(exp);
2163 struct ost_body *body, *repbody;
2164 struct ofd_thread_info *info;
2165 struct ofd_object *fo;
2166 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2167 const struct lu_env *env = svc_thread->t_env;
2168 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2169 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2170 struct lu_ladvise *ladvise;
2172 struct ladvise_hdr *ladvise_hdr;
2173 struct obd_ioobj ioo;
2174 struct lustre_handle lockh = { 0 };
2177 struct dt_object *dob;
2183 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2184 body = tsi->tsi_ost_body;
2186 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2187 RETURN(err_serious(-EPROTO));
2189 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2190 &RMF_OST_LADVISE_HDR);
2191 if (ladvise_hdr == NULL)
2192 RETURN(err_serious(-EPROTO));
2194 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2195 ladvise_hdr->lah_count < 1)
2196 RETURN(err_serious(-EPROTO));
2198 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2199 RETURN(err_serious(-EPROTO));
2201 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2202 if (ladvise == NULL)
2203 RETURN(err_serious(-EPROTO));
2205 num_advise = req_capsule_get_size(&req->rq_pill,
2206 &RMF_OST_LADVISE, RCL_CLIENT) /
2208 if (num_advise < ladvise_hdr->lah_count)
2209 RETURN(err_serious(-EPROTO));
2211 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2212 repbody->oa = body->oa;
2214 info = ofd_info_init(env, exp);
2216 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2217 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2221 fo = ofd_object_find(env, ofd, &info->fti_fid);
2226 LASSERT(fo != NULL);
2227 dob = ofd_object_child(fo);
2229 if (ptlrpc_connection_is_local(exp->exp_connection))
2230 dbt |= DT_BUFS_TYPE_LOCAL;
2232 for (i = 0; i < num_advise; i++, ladvise++) {
2233 start = ladvise->lla_start;
2234 end = ladvise->lla_end;
2236 rc = err_serious(-EPROTO);
2240 /* Handle different advice types */
2241 switch (ladvise->lla_advice) {
2245 case LU_LADVISE_WILLREAD:
2249 ioo.ioo_oid = body->oa.o_oi;
2251 rc = tgt_extent_lock(exp->exp_obd->obd_namespace,
2252 &tsi->tsi_resid, start, end - 1,
2253 &lockh, LCK_PR, &flags);
2257 req->rq_status = ofd_ladvise_prefetch(env, fo,
2260 tgt_extent_unlock(&lockh, LCK_PR);
2262 case LU_LADVISE_DONTNEED:
2263 rc = dt_ladvise(env, dob, ladvise->lla_start,
2264 ladvise->lla_end, LU_LADVISE_DONTNEED);
2271 ofd_object_put(env, fo);
2272 req->rq_status = rc;
2277 * OFD request handler for OST_QUOTACTL RPC.
2279 * This is part of request processing to validate incoming request fields,
2280 * get the requested data from OSD and pack reply.
2282 * \param[in] tsi target session environment for this request
2284 * \retval 0 if successful
2285 * \retval negative value on error
2287 static int ofd_quotactl(struct tgt_session_info *tsi)
2289 struct obd_quotactl *oqctl, *repoqc;
2290 struct lu_nodemap *nodemap;
2296 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2298 RETURN(err_serious(-EPROTO));
2300 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2302 RETURN(err_serious(-ENOMEM));
2306 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2307 if (IS_ERR(nodemap))
2308 RETURN(PTR_ERR(nodemap));
2311 if (oqctl->qc_type == USRQUOTA)
2312 id = nodemap_map_id(nodemap, NODEMAP_UID,
2313 NODEMAP_CLIENT_TO_FS,
2315 else if (oqctl->qc_type == GRPQUOTA)
2316 id = nodemap_map_id(nodemap, NODEMAP_GID,
2317 NODEMAP_CLIENT_TO_FS,
2320 nodemap_putref(nodemap);
2322 if (repoqc->qc_id != id)
2323 swap(repoqc->qc_id, id);
2325 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2327 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2330 if (repoqc->qc_id != id)
2331 swap(repoqc->qc_id, id);
2337 * Calculate the amount of time for lock prolongation.
2339 * This is helper for ofd_prolong_extent_locks() function to get
2340 * the timeout extra time.
2342 * \param[in] req current request
2344 * \retval amount of time to extend the timeout with
2346 static inline time64_t prolong_timeout(struct ptlrpc_request *req)
2348 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2349 time64_t req_timeout;
2352 return obd_timeout / 2;
2354 req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec;
2355 return max_t(time64_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
2360 * Prolong lock timeout for the given extent.
2362 * This function finds all locks related with incoming request and
2363 * prolongs their timeout.
2365 * If a client is holding a lock for a long time while it sends
2366 * read or write RPCs to the OST for the object under this lock,
2367 * then we don't want the OST to evict the client. Otherwise,
2368 * if the network or disk is very busy then the client may not
2369 * be able to make any progress to clear out dirty pages under
2370 * the lock and the application will fail.
2372 * Every time a Bulk Read/Write (BRW) request arrives for the object
2373 * covered by the lock, extend the timeout on that lock. The RPC should
2374 * contain a lock handle for the lock it is using, but this
2375 * isn't handled correctly by all client versions, and the
2376 * request may cover multiple locks.
2378 * \param[in] tsi target session environment for this request
2379 * \param[in] data struct of data to prolong locks
2382 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2383 struct ldlm_prolong_args *data)
2385 struct obdo *oa = &tsi->tsi_ost_body->oa;
2386 struct ldlm_lock *lock;
2390 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2391 data->lpa_export = tsi->tsi_exp;
2392 data->lpa_resid = tsi->tsi_resid;
2394 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2395 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2396 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2397 data->lpa_extent.end);
2399 if (oa->o_valid & OBD_MD_FLHANDLE) {
2400 /* mostly a request should be covered by only one lock, try
2402 lock = ldlm_handle2lock(&oa->o_handle);
2404 /* Fast path to check if the lock covers the whole IO
2405 * region exclusively. */
2406 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2407 &data->lpa_extent)) {
2409 LASSERT(lock->l_export == data->lpa_export);
2410 ldlm_lock_prolong_one(lock, data);
2411 LDLM_LOCK_PUT(lock);
2412 if (data->lpa_locks_cnt > 0)
2414 /* The lock was destroyed probably lets try
2417 lock->l_last_used = ktime_get();
2418 LDLM_LOCK_PUT(lock);
2423 ldlm_resource_prolong(data);
2428 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2430 * Determine if \a lock and the lock from request \a req are equivalent
2431 * by comparing their resource names, modes, and extents.
2433 * It is used to give priority to read and write RPCs being done
2434 * under this lock so that the client can drop the contended
2435 * lock more quickly and let other clients use it. This improves
2436 * overall performance in the case where the first client gets a
2437 * very large lock extent that prevents other clients from
2438 * submitting their writes.
2440 * \param[in] req ptlrpc_request being processed
2441 * \param[in] lock contended lock to match
2443 * \retval 1 if lock is matched
2444 * \retval 0 otherwise
2446 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2447 struct ldlm_lock *lock)
2449 struct niobuf_remote *rnb;
2450 struct obd_ioobj *ioo;
2451 enum ldlm_mode mode;
2452 struct ldlm_extent ext;
2453 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2457 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2458 LASSERT(ioo != NULL);
2460 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2461 LASSERT(rnb != NULL);
2463 ext.start = rnb->rnb_offset;
2464 rnb += ioo->ioo_bufcnt - 1;
2465 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2467 LASSERT(lock->l_resource != NULL);
2468 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2471 /* a bulk write can only hold a reference on a PW extent lock
2474 mode = LCK_PW | LCK_GROUP;
2475 if (opc == OST_READ)
2476 /* whereas a bulk read can be protected by either a PR or PW
2480 if (!(lock->l_granted_mode & mode))
2483 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2487 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2489 * Check for whether the given PTLRPC request (\a req) is blocking
2490 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2493 * \param[in] req the incoming request
2495 * \retval 1 if \a req is blocking an LDLM lock cancel
2496 * \retval 0 if it is not
2497 * \retval -ESTALE if lock is not found
2499 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2501 struct tgt_session_info *tsi;
2502 struct obd_ioobj *ioo;
2503 struct niobuf_remote *rnb;
2505 struct ldlm_prolong_args pa = { 0 };
2509 /* Don't use tgt_ses_info() to get session info, because lock_match()
2510 * can be called while request has no processing thread yet. */
2511 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2514 * Use LASSERT below because malformed RPCs should have
2515 * been filtered out in tgt_hpreq_handler().
2517 opc = lustre_msg_get_opc(req->rq_reqmsg);
2518 LASSERT(opc == OST_READ || opc == OST_WRITE);
2520 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2521 LASSERT(ioo != NULL);
2523 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2524 LASSERT(rnb != NULL);
2525 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2527 pa.lpa_mode = LCK_PW | LCK_GROUP;
2528 if (opc == OST_READ)
2529 pa.lpa_mode |= LCK_PR;
2531 pa.lpa_extent.start = rnb->rnb_offset;
2532 rnb += ioo->ioo_bufcnt - 1;
2533 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2535 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
2536 " (%llu->%llu)\n", tgt_name(tsi->tsi_tgt),
2537 current->comm, PFID(&tsi->tsi_fid), pa.lpa_extent.start,
2540 ofd_prolong_extent_locks(tsi, &pa);
2542 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2543 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2545 if (pa.lpa_blocks_cnt > 0)
2548 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2552 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2554 * Called after the request has been handled. It refreshes lock timeout again
2555 * so that client has more time to send lock cancel RPC.
2557 * \param[in] req request which is being processed.
2559 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2561 ofd_rw_hpreq_check(req);
2565 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2567 * This function checks if the given lock is the same by its resname, mode
2568 * and extent as one taken from the request.
2569 * It is used to give priority to punch/truncate RPCs that might lead to
2570 * the fastest release of that lock when a lock is contended.
2572 * \param[in] req ptlrpc_request being processed
2573 * \param[in] lock contended lock to match
2575 * \retval 1 if lock is matched
2576 * \retval 0 otherwise
2578 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2579 struct ldlm_lock *lock)
2581 struct tgt_session_info *tsi;
2583 struct ldlm_extent ext;
2587 /* Don't use tgt_ses_info() to get session info, because lock_match()
2588 * can be called while request has no processing thread yet. */
2589 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2592 * Use LASSERT below because malformed RPCs should have
2593 * been filtered out in tgt_hpreq_handler().
2595 LASSERT(tsi->tsi_ost_body != NULL);
2596 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2597 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2600 oa = &tsi->tsi_ost_body->oa;
2601 ext.start = oa->o_size;
2602 ext.end = oa->o_blocks;
2604 LASSERT(lock->l_resource != NULL);
2605 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2608 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2611 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2615 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2617 * High-priority queue request check for whether the given punch request
2618 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2619 * covered by an LDLM lock.
2623 * \param[in] req the incoming request
2625 * \retval 1 if \a req is blocking an LDLM lock cancel
2626 * \retval 0 if it is not
2627 * \retval -ESTALE if lock is not found
2629 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2631 struct tgt_session_info *tsi;
2633 struct ldlm_prolong_args pa = { 0 };
2637 /* Don't use tgt_ses_info() to get session info, because lock_match()
2638 * can be called while request has no processing thread yet. */
2639 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2640 LASSERT(tsi != NULL);
2641 oa = &tsi->tsi_ost_body->oa;
2643 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2644 oa->o_flags & OBD_FL_SRVLOCK));
2646 pa.lpa_mode = LCK_PW | LCK_GROUP;
2647 pa.lpa_extent.start = oa->o_size;
2648 pa.lpa_extent.end = oa->o_blocks;
2651 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2652 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2653 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2655 ofd_prolong_extent_locks(tsi, &pa);
2657 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2658 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2660 if (pa.lpa_blocks_cnt > 0)
2663 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2667 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2669 * Called after the request has been handled. It refreshes lock timeout again
2670 * so that client has more time to send lock cancel RPC.
2672 * \param[in] req request which is being processed.
2674 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2676 ofd_punch_hpreq_check(req);
2679 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2680 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2681 .hpreq_check = ofd_rw_hpreq_check,
2682 .hpreq_fini = ofd_rw_hpreq_fini
2685 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2686 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2687 .hpreq_check = ofd_punch_hpreq_check,
2688 .hpreq_fini = ofd_punch_hpreq_fini
2692 * Assign high priority operations to an IO request.
2694 * Check if the incoming request is a candidate for
2695 * high-priority processing. If it is, assign it a high
2696 * priority operations table.
2698 * \param[in] tsi target session environment for this request
2700 static void ofd_hp_brw(struct tgt_session_info *tsi)
2702 struct niobuf_remote *rnb;
2703 struct obd_ioobj *ioo;
2707 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2708 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2709 if (ioo->ioo_bufcnt > 0) {
2710 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2711 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2713 /* no high priority if server lock is needed */
2714 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2715 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2719 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2723 * Assign high priority operations to an punch request.
2725 * Check if the incoming request is a candidate for
2726 * high-priority processing. If it is, assign it a high
2727 * priority operations table.
2729 * \param[in] tsi target session environment for this request
2731 static void ofd_hp_punch(struct tgt_session_info *tsi)
2733 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2734 /* no high-priority if server lock is needed */
2735 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2736 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2737 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2738 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2740 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2743 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2744 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2745 #define OST_BRW_READ OST_READ
2746 #define OST_BRW_WRITE OST_WRITE
2749 * Table of OFD-specific request handlers
2751 * This table contains all opcodes accepted by OFD and
2752 * specifies handlers for them. The tgt_request_handler()
2753 * uses such table from each target to process incoming
2756 static struct tgt_handler ofd_tgt_handlers[] = {
2757 TGT_RPC_HANDLER(OST_FIRST_OPC,
2758 0, OST_CONNECT, tgt_connect,
2759 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2760 TGT_RPC_HANDLER(OST_FIRST_OPC,
2761 0, OST_DISCONNECT, tgt_disconnect,
2762 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2763 TGT_RPC_HANDLER(OST_FIRST_OPC,
2764 0, OST_SET_INFO, ofd_set_info_hdl,
2765 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2766 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2767 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
2768 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2769 OST_SETATTR, ofd_setattr_hdl),
2770 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2771 OST_CREATE, ofd_create_hdl),
2772 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2773 OST_DESTROY, ofd_destroy_hdl),
2774 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
2775 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2776 OST_BRW_READ, tgt_brw_read,
2778 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2779 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
2781 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2782 OST_PUNCH, ofd_punch_hdl,
2784 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
2785 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
2786 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_LADVISE, ofd_ladvise_hdl),
2789 static struct tgt_opc_slice ofd_common_slice[] = {
2791 .tos_opc_start = OST_FIRST_OPC,
2792 .tos_opc_end = OST_LAST_OPC,
2793 .tos_hs = ofd_tgt_handlers
2796 .tos_opc_start = OBD_FIRST_OPC,
2797 .tos_opc_end = OBD_LAST_OPC,
2798 .tos_hs = tgt_obd_handlers
2801 .tos_opc_start = LDLM_FIRST_OPC,
2802 .tos_opc_end = LDLM_LAST_OPC,
2803 .tos_hs = tgt_dlm_handlers
2806 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2807 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2808 .tos_hs = tgt_out_handlers
2811 .tos_opc_start = SEQ_FIRST_OPC,
2812 .tos_opc_end = SEQ_LAST_OPC,
2813 .tos_hs = seq_handlers
2816 .tos_opc_start = LFSCK_FIRST_OPC,
2817 .tos_opc_end = LFSCK_LAST_OPC,
2818 .tos_hs = tgt_lfsck_handlers
2821 .tos_opc_start = SEC_FIRST_OPC,
2822 .tos_opc_end = SEC_LAST_OPC,
2823 .tos_hs = tgt_sec_ctx_handlers
2830 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2831 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2834 * Implementation of lu_context_key::lct_key_exit.
2836 * Optional method called on lu_context_exit() for all allocated
2838 * It is used in OFD to sanitize context values which may be re-used
2839 * during another request processing by the same thread.
2841 * \param[in] ctx execution context
2842 * \param[in] key context key
2843 * \param[in] data ofd_thread_info
2845 static void ofd_key_exit(const struct lu_context *ctx,
2846 struct lu_context_key *key, void *data)
2848 struct ofd_thread_info *info = data;
2850 info->fti_env = NULL;
2851 info->fti_exp = NULL;
2854 info->fti_pre_version = 0;
2856 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2859 struct lu_context_key ofd_thread_key = {
2860 .lct_tags = LCT_DT_THREAD,
2861 .lct_init = ofd_key_init,
2862 .lct_fini = ofd_key_fini,
2863 .lct_exit = ofd_key_exit
2867 * Initialize OFD device according to parameters in the config log \a cfg.
2869 * This is the main starting point of OFD initialization. It fills all OFD
2870 * parameters with their initial values and calls other initializing functions
2871 * to set up all OFD subsystems.
2873 * \param[in] env execution environment
2874 * \param[in] m OFD device
2875 * \param[in] ldt LU device type of OFD
2876 * \param[in] cfg configuration log
2878 * \retval 0 if successful
2879 * \retval negative value on error
2881 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2882 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2884 const char *dev = lustre_cfg_string(cfg, 0);
2885 struct ofd_thread_info *info = NULL;
2886 struct obd_device *obd;
2887 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2889 struct nm_config_file *nodemap_config;
2890 struct obd_device_target *obt;
2895 obd = class_name2obd(dev);
2897 CERROR("Cannot find obd with name %s\n", dev);
2901 rc = lu_env_refill((struct lu_env *)env);
2906 obt->obt_magic = OBT_MAGIC;
2908 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2909 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2911 spin_lock_init(&m->ofd_flags_lock);
2912 m->ofd_raid_degraded = 0;
2913 m->ofd_checksum_t10pi_enforce = 0;
2914 m->ofd_syncjournal = 0;
2916 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2918 m->ofd_seq_count = 0;
2919 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2920 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2921 spin_lock_init(&m->ofd_inconsistency_lock);
2923 spin_lock_init(&m->ofd_batch_lock);
2924 init_rwsem(&m->ofd_lastid_rwsem);
2926 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2927 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2928 /* set this lu_device to obd, because error handling need it */
2929 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2931 rc = ofd_procfs_init(m);
2933 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2937 /* No connection accepted until configurations will finish */
2938 spin_lock(&obd->obd_dev_lock);
2939 obd->obd_no_conn = 1;
2940 spin_unlock(&obd->obd_dev_lock);
2941 obd->obd_replayable = 1;
2942 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2943 char *str = lustre_cfg_string(cfg, 4);
2945 if (strchr(str, 'n')) {
2946 CWARN("%s: recovery disabled\n", obd->obd_name);
2947 obd->obd_replayable = 0;
2951 info = ofd_info_init(env, NULL);
2953 GOTO(err_fini_proc, rc = -EFAULT);
2955 rc = ofd_stack_init(env, m, cfg);
2957 CERROR("Can't init device stack, rc %d\n", rc);
2958 GOTO(err_fini_proc, rc);
2961 ofd_procfs_add_brw_stats_symlink(m);
2963 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2964 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2965 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2966 LDLM_NAMESPACE_SERVER,
2967 LDLM_NAMESPACE_GREEDY,
2969 if (m->ofd_namespace == NULL)
2970 GOTO(err_fini_stack, rc = -ENOMEM);
2971 /* set obd_namespace for compatibility with old code */
2972 obd->obd_namespace = m->ofd_namespace;
2973 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2974 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2975 m->ofd_namespace->ns_lvbp = m;
2977 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2978 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2980 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2981 OBD_FAIL_OST_ALL_REQUEST_NET,
2982 OBD_FAIL_OST_ALL_REPLY_NET);
2984 GOTO(err_free_ns, rc);
2986 tgd->tgd_reserved_pcnt = 0;
2988 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
2989 m->ofd_cksum_types_supported =
2990 obd_cksum_types_supported_server(obd->obd_name);
2991 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2992 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
2993 OFD_PRECREATE_SMALL_FS)
2994 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2996 rc = ofd_fs_setup(env, m, obd);
2998 GOTO(err_fini_lut, rc);
3000 fid.f_seq = FID_SEQ_LOCAL_NAME;
3003 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
3006 GOTO(err_fini_fs, rc);
3008 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
3010 if (IS_ERR(nodemap_config)) {
3011 rc = PTR_ERR(nodemap_config);
3013 GOTO(err_fini_los, rc);
3015 obt->obt_nodemap_config_file = nodemap_config;
3018 rc = ofd_start_inconsistency_verification_thread(m);
3020 GOTO(err_fini_nm, rc);
3022 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
3027 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
3028 obt->obt_nodemap_config_file = NULL;
3030 local_oid_storage_fini(env, m->ofd_los);
3033 ofd_fs_cleanup(env, m);
3035 tgt_fini(env, &m->ofd_lut);
3037 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3038 obd->obd_namespace = m->ofd_namespace = NULL;
3040 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3047 * Stop the OFD device
3049 * This function stops the OFD device and all its subsystems.
3050 * This is the end of OFD lifecycle.
3052 * \param[in] env execution environment
3053 * \param[in] m OFD device
3055 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3057 struct obd_device *obd = ofd_obd(m);
3058 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3059 struct lfsck_stop stop;
3061 stop.ls_status = LS_PAUSED;
3063 lfsck_stop(env, m->ofd_osd, &stop);
3064 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3065 target_recovery_fini(obd);
3066 if (m->ofd_namespace != NULL)
3067 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3068 d->ld_obd->obd_force);
3070 obd_exports_barrier(obd);
3071 obd_zombie_barrier();
3073 tgt_fini(env, &m->ofd_lut);
3074 ofd_stop_inconsistency_verification_thread(m);
3075 lfsck_degister(env, m->ofd_osd);
3076 ofd_fs_cleanup(env, m);
3077 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3078 obd->u.obt.obt_nodemap_config_file = NULL;
3080 if (m->ofd_namespace != NULL) {
3081 ldlm_namespace_free_post(m->ofd_namespace);
3082 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3085 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3087 LASSERT(atomic_read(&d->ld_ref) == 0);
3088 server_put_mount(obd->obd_name, true);
3093 * Implementation of lu_device_type_operations::ldto_device_fini.
3095 * Finalize device. Dual to ofd_device_init(). It is called from
3096 * obd_precleanup() and stops the current device.
3098 * \param[in] env execution environment
3099 * \param[in] d LU device of OFD
3103 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3104 struct lu_device *d)
3107 ofd_fini(env, ofd_dev(d));
3112 * Implementation of lu_device_type_operations::ldto_device_free.
3114 * Free OFD device. Dual to ofd_device_alloc().
3116 * \param[in] env execution environment
3117 * \param[in] d LU device of OFD
3121 static struct lu_device *ofd_device_free(const struct lu_env *env,
3122 struct lu_device *d)
3124 struct ofd_device *m = ofd_dev(d);
3126 dt_device_fini(&m->ofd_dt_dev);
3132 * Implementation of lu_device_type_operations::ldto_device_alloc.
3134 * This function allocates the new OFD device. It is called from
3135 * obd_setup() if OBD device had lu_device_type defined.
3137 * \param[in] env execution environment
3138 * \param[in] t lu_device_type of OFD device
3139 * \param[in] cfg configuration log
3141 * \retval pointer to the lu_device of just allocated OFD
3142 * \retval ERR_PTR of return value on error
3144 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3145 struct lu_device_type *t,
3146 struct lustre_cfg *cfg)
3148 struct ofd_device *m;
3149 struct lu_device *l;
3154 return ERR_PTR(-ENOMEM);
3156 l = &m->ofd_dt_dev.dd_lu_dev;
3157 dt_device_init(&m->ofd_dt_dev, t);
3158 rc = ofd_init0(env, m, t, cfg);
3160 ofd_device_free(env, l);
3167 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3168 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3170 static struct lu_device_type_operations ofd_device_type_ops = {
3171 .ldto_init = ofd_type_init,
3172 .ldto_fini = ofd_type_fini,
3174 .ldto_start = ofd_type_start,
3175 .ldto_stop = ofd_type_stop,
3177 .ldto_device_alloc = ofd_device_alloc,
3178 .ldto_device_free = ofd_device_free,
3179 .ldto_device_fini = ofd_device_fini
3182 static struct lu_device_type ofd_device_type = {
3183 .ldt_tags = LU_DEVICE_DT,
3184 .ldt_name = LUSTRE_OST_NAME,
3185 .ldt_ops = &ofd_device_type_ops,
3186 .ldt_ctx_tags = LCT_DT_THREAD
3190 * Initialize OFD module.
3192 * This function is called upon module loading. It registers OFD device type
3193 * and prepares all in-memory structures used by all OFD devices.
3195 * \retval 0 if successful
3196 * \retval negative value on error
3198 static int __init ofd_init(void)
3202 rc = lu_kmem_init(ofd_caches);
3206 rc = ofd_fmd_init();
3208 lu_kmem_fini(ofd_caches);
3212 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3213 LUSTRE_OST_NAME, &ofd_device_type);
3220 * This function is called upon OFD module unloading.
3221 * It frees all related structures and unregisters OFD device type.
3223 static void __exit ofd_exit(void)
3226 lu_kmem_fini(ofd_caches);
3227 class_unregister_type(LUSTRE_OST_NAME);
3230 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3231 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3232 MODULE_VERSION(LUSTRE_VERSION_STRING);
3233 MODULE_LICENSE("GPL");
3235 module_init(ofd_init);
3236 module_exit(ofd_exit);