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, 2015, 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 <lustre_param.h>
73 #include <lustre_fid.h>
74 #include <lustre_lfsck.h>
75 #include <lustre/lustre_idl.h>
76 #include <lustre_dlm.h>
77 #include <lustre_quota.h>
78 #include <lustre_nodemap.h>
80 #include "ofd_internal.h"
82 /* Slab for OFD object allocation */
83 static struct kmem_cache *ofd_object_kmem;
85 static struct lu_kmem_descr ofd_caches[] = {
87 .ckd_cache = &ofd_object_kmem,
88 .ckd_name = "ofd_obj",
89 .ckd_size = sizeof(struct ofd_object)
97 * Connect OFD to the next device in the stack.
99 * This function is used for device stack configuration and links OFD
100 * device with bottom OSD device.
102 * \param[in] env execution environment
103 * \param[in] m OFD device
104 * \param[in] next name of next device in the stack
105 * \param[out] exp export to return
107 * \retval 0 and export in \a exp if successful
108 * \retval negative value on error
110 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
111 const char *next, struct obd_export **exp)
113 struct obd_connect_data *data = NULL;
114 struct obd_device *obd;
120 GOTO(out, rc = -ENOMEM);
122 obd = class_name2obd(next);
124 CERROR("%s: can't locate next device: %s\n",
126 GOTO(out, rc = -ENOTCONN);
129 data->ocd_connect_flags = OBD_CONNECT_VERSION;
130 data->ocd_version = LUSTRE_VERSION_CODE;
132 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
134 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
135 ofd_name(m), next, rc);
139 m->ofd_dt_dev.dd_lu_dev.ld_site =
140 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
141 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
142 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
143 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
152 * Initialize stack of devices.
154 * This function initializes OFD-OSD device stack to serve OST requests
156 * \param[in] env execution environment
157 * \param[in] m OFD device
158 * \param[in] cfg Lustre config for this server
160 * \retval 0 if successful
161 * \retval negative value on error
163 static int ofd_stack_init(const struct lu_env *env,
164 struct ofd_device *m, struct lustre_cfg *cfg)
166 const char *dev = lustre_cfg_string(cfg, 0);
168 struct ofd_thread_info *info = ofd_info(env);
169 struct lustre_mount_info *lmi;
170 struct lustre_mount_data *lmd;
176 lmi = server_get_mount(dev);
178 CERROR("Cannot get mount info for %s!\n", dev);
182 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
183 if (lmd != NULL && lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
184 m->ofd_skip_lfsck = 1;
186 /* find bottom osd */
187 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
191 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
192 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
193 OBD_FREE(osdname, MTI_NAME_MAXLEN);
197 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
199 m->ofd_osd = lu2dt_dev(d);
201 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
202 "%s-osd", lustre_cfg_string(cfg, 0));
208 * Finalize the device stack OFD-OSD.
210 * This function cleans OFD-OSD device stack and
211 * disconnects OFD from the OSD.
213 * \param[in] env execution environment
214 * \param[in] m OFD device
215 * \param[in] top top device of stack
217 * \retval 0 if successful
218 * \retval negative value on error
220 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
221 struct lu_device *top)
223 struct obd_device *obd = ofd_obd(m);
224 struct lustre_cfg_bufs bufs;
225 struct lustre_cfg *lcfg;
230 lu_site_purge(env, top->ld_site, ~0);
231 /* process cleanup, pass mdt obd name to get obd umount flags */
232 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
237 lustre_cfg_bufs_set_string(&bufs, 1, flags);
238 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
243 top->ld_ops->ldo_process_config(env, top, lcfg);
244 lustre_cfg_free(lcfg);
246 lu_site_purge(env, top->ld_site, ~0);
247 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
248 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
249 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
252 LASSERT(m->ofd_osd_exp);
253 obd_disconnect(m->ofd_osd_exp);
258 /* For interoperability, see mdt_interop_param[]. */
259 static struct cfg_interop_param ofd_interop_param[] = {
260 { "ost.quota_type", NULL },
265 * Check if parameters are symlinks to the OSD.
267 * Some parameters were moved from ofd to osd and only their
268 * symlinks were kept in ofd by LU-3106. They are:
269 * -writehthrough_cache_enable
270 * -readcache_max_filesize
274 * Since they are not included by the static lprocfs var list, a pre-check
275 * is added for them to avoid "unknown param" errors. If they are matched
276 * in this check, they will be passed to the OSD directly.
278 * \param[in] param parameters to check
280 * \retval true if param is symlink to OSD param
283 static bool match_symlink_param(char *param)
288 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
289 sval = strchr(param, '=');
291 paramlen = sval - param;
292 if (strncmp(param, "writethrough_cache_enable",
294 strncmp(param, "readcache_max_filesize",
296 strncmp(param, "read_cache_enable",
298 strncmp(param, "brw_stats", paramlen) == 0)
307 * Process various configuration parameters.
309 * This function is used by MGS to process specific configurations and
310 * pass them through to the next device in server stack, i.e. the OSD.
312 * \param[in] env execution environment
313 * \param[in] d LU device of OFD
314 * \param[in] cfg parameters to process
316 * \retval 0 if successful
317 * \retval negative value on error
319 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
320 struct lustre_cfg *cfg)
322 struct ofd_device *m = ofd_dev(d);
323 struct dt_device *dt_next = m->ofd_osd;
324 struct lu_device *next = &dt_next->dd_lu_dev;
329 switch (cfg->lcfg_command) {
331 struct obd_device *obd = ofd_obd(m);
332 /* For interoperability */
333 struct cfg_interop_param *ptr = NULL;
334 struct lustre_cfg *old_cfg = NULL;
337 param = lustre_cfg_string(cfg, 1);
339 CERROR("param is empty\n");
344 ptr = class_find_old_param(param, ofd_interop_param);
346 if (ptr->new_param == NULL) {
348 CWARN("For interoperability, skip this %s."
349 " It is obsolete.\n", ptr->old_param);
353 CWARN("Found old param %s, changed it to %s.\n",
354 ptr->old_param, ptr->new_param);
357 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
364 if (match_symlink_param(param)) {
365 rc = next->ld_ops->ldo_process_config(env, next, cfg);
369 rc = class_process_proc_param(PARAM_OST, obd->obd_vars, cfg,
371 if (rc > 0 || rc == -ENOSYS) {
372 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
374 /* we don't understand; pass it on */
375 rc = next->ld_ops->ldo_process_config(env, next, cfg);
379 case LCFG_SPTLRPC_CONF: {
384 /* others are passed further */
385 rc = next->ld_ops->ldo_process_config(env, next, cfg);
392 * Implementation of lu_object_operations::loo_object_init for OFD
394 * Allocate just the next object (OSD) in stack.
396 * \param[in] env execution environment
397 * \param[in] o lu_object of OFD object
398 * \param[in] conf additional configuration parameters, not used here
400 * \retval 0 if successful
401 * \retval negative value on error
403 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
404 const struct lu_object_conf *conf)
406 struct ofd_device *d = ofd_dev(o->lo_dev);
407 struct lu_device *under;
408 struct lu_object *below;
413 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
414 PFID(lu_object_fid(o)));
416 under = &d->ofd_osd->dd_lu_dev;
417 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
419 lu_object_add(o, below);
427 * Implementation of lu_object_operations::loo_object_free.
429 * Finish OFD object lifecycle and free its memory.
431 * \param[in] env execution environment
432 * \param[in] o LU object of OFD object
434 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
436 struct ofd_object *of = ofd_obj(o);
437 struct lu_object_header *h;
442 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
443 PFID(lu_object_fid(o)));
446 lu_object_header_fini(h);
447 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
452 * Implementation of lu_object_operations::loo_object_print.
454 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
455 * LU_OBJECT_DEBUG() for more details about the compound object printing.
457 * \param[in] env execution environment
458 * \param[in] cookie opaque data passed to the printer function
459 * \param[in] p printer function to use
460 * \param[in] o LU object of OFD object
462 * \retval 0 if successful
463 * \retval negative value on error
465 static int ofd_object_print(const struct lu_env *env, void *cookie,
466 lu_printer_t p, const struct lu_object *o)
468 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
471 static struct lu_object_operations ofd_obj_ops = {
472 .loo_object_init = ofd_object_init,
473 .loo_object_free = ofd_object_free,
474 .loo_object_print = ofd_object_print
478 * Implementation of lu_device_operations::lod_object_alloc.
480 * This function allocates OFD part of compound OFD-OSD object and
481 * initializes its header, because OFD is the top device in stack
483 * \param[in] env execution environment
484 * \param[in] hdr object header, NULL for OFD
485 * \param[in] d lu_device
487 * \retval allocated object if successful
488 * \retval NULL value on failed allocation
490 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
491 const struct lu_object_header *hdr,
494 struct ofd_object *of;
498 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
501 struct lu_object_header *h;
503 o = &of->ofo_obj.do_lu;
505 lu_object_header_init(h);
506 lu_object_init(o, h, d);
507 lu_object_add_top(h, o);
508 o->lo_ops = &ofd_obj_ops;
516 * Return the result of LFSCK run to the OFD.
518 * Notify OFD about result of LFSCK run. That may block the new object
519 * creation until problem is fixed by LFSCK.
521 * \param[in] env execution environment
522 * \param[in] data pointer to the OFD device
523 * \param[in] event LFSCK event type
525 * \retval 0 if successful
526 * \retval negative value on unknown event
528 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
529 enum lfsck_events event)
531 struct ofd_device *ofd = data;
532 struct obd_device *obd = ofd_obd(ofd);
535 case LE_LASTID_REBUILDING:
536 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
537 "on the device until the LAST_ID rebuilt successfully.\n",
539 down_write(&ofd->ofd_lastid_rwsem);
540 ofd->ofd_lastid_rebuilding = 1;
541 up_write(&ofd->ofd_lastid_rwsem);
543 case LE_LASTID_REBUILT: {
544 down_write(&ofd->ofd_lastid_rwsem);
545 ofd_seqs_free(env, ofd);
546 ofd->ofd_lastid_rebuilding = 0;
547 ofd->ofd_lastid_gen++;
548 up_write(&ofd->ofd_lastid_rwsem);
549 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
554 CERROR("%s: unknown lfsck event: rc = %d\n",
555 ofd_name(ofd), event);
563 * Implementation of lu_device_operations::ldo_prepare.
565 * This method is called after layer has been initialized and before it starts
566 * serving user requests. In OFD it starts lfsk check routines and initializes
569 * \param[in] env execution environment
570 * \param[in] pdev higher device in stack, NULL for OFD
571 * \param[in] dev lu_device of OFD device
573 * \retval 0 if successful
574 * \retval negative value on error
576 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
577 struct lu_device *dev)
579 struct ofd_thread_info *info;
580 struct ofd_device *ofd = ofd_dev(dev);
581 struct obd_device *obd = ofd_obd(ofd);
582 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
587 info = ofd_info_init(env, NULL);
591 /* initialize lower device */
592 rc = next->ld_ops->ldo_prepare(env, dev, next);
596 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
597 ofd_lfsck_out_notify, ofd, false);
599 CERROR("%s: failed to initialize lfsck: rc = %d\n",
604 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
605 /* The LFSCK instance is registered just now, so it must be there when
606 * register the namespace to such instance. */
607 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
609 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
610 LASSERT(obd->obd_no_conn);
611 spin_lock(&obd->obd_dev_lock);
612 obd->obd_no_conn = 0;
613 spin_unlock(&obd->obd_dev_lock);
615 if (obd->obd_recovering == 0)
616 ofd_postrecov(env, ofd);
622 * Implementation of lu_device_operations::ldo_recovery_complete.
624 * This method notifies all layers about 'recovery complete' event. That means
625 * device is in full state and consistent. An OFD calculates available grant
626 * space upon this event.
628 * \param[in] env execution environment
629 * \param[in] dev lu_device of OFD device
631 * \retval 0 if successful
632 * \retval negative value on error
634 static int ofd_recovery_complete(const struct lu_env *env,
635 struct lu_device *dev)
637 struct ofd_thread_info *oti = ofd_info(env);
638 struct ofd_device *ofd = ofd_dev(dev);
639 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
645 * Grant space for object precreation on the self export.
646 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
647 * is enough to create 10k objects. More space is then acquired for
648 * precreation in ofd_grant_create().
650 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
651 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
652 oti->fti_ocd.ocd_grant *= ofd->ofd_dt_conf.ddp_inodespace;
653 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
654 OBD_CONNECT_GRANT_PARAM;
655 ofd_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
657 rc = next->ld_ops->ldo_recovery_complete(env, next);
662 * lu_device_operations matrix for OFD device.
664 static struct lu_device_operations ofd_lu_ops = {
665 .ldo_object_alloc = ofd_object_alloc,
666 .ldo_process_config = ofd_process_config,
667 .ldo_recovery_complete = ofd_recovery_complete,
668 .ldo_prepare = ofd_prepare,
671 LPROC_SEQ_FOPS(lprocfs_nid_stats_clear);
674 * Initialize all needed procfs entries for OFD device.
676 * \param[in] ofd OFD device
678 * \retval 0 if successful
679 * \retval negative value on error
681 static int ofd_procfs_init(struct ofd_device *ofd)
683 struct obd_device *obd = ofd_obd(ofd);
684 struct proc_dir_entry *entry;
689 /* lprocfs must be setup before the ofd so state can be safely added
690 * to /proc incrementally as the ofd is setup */
691 obd->obd_vars = lprocfs_ofd_obd_vars;
692 rc = lprocfs_obd_setup(obd);
694 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
699 rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
701 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
703 GOTO(obd_cleanup, rc);
706 obd->obd_uses_nid_stats = 1;
708 entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
711 CERROR("%s: error %d setting up lprocfs for %s\n",
712 obd->obd_name, rc, "exports");
713 GOTO(obd_cleanup, rc);
715 obd->obd_proc_exports_entry = entry;
717 entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
718 obd, &lprocfs_nid_stats_clear_fops);
721 CERROR("%s: add proc entry 'clear' failed: %d.\n",
723 GOTO(obd_cleanup, rc);
726 ofd_stats_counter_init(obd->obd_stats);
728 rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
729 ofd_stats_counter_init);
731 GOTO(obd_cleanup, rc);
734 lprocfs_obd_cleanup(obd);
735 lprocfs_free_obd_stats(obd);
741 * Expose OSD statistics to OFD layer.
743 * The osd interfaces to the backend file system exposes useful data
744 * such as brw_stats and read or write cache states. This same data
745 * needs to be exposed into the obdfilter (ofd) layer to maintain
746 * backwards compatibility. This function creates the symlinks in the
747 * proc layer to enable this.
749 * \param[in] ofd OFD device
751 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
753 struct obd_device *obd = ofd_obd(ofd);
754 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
756 if (obd->obd_proc_entry == NULL)
759 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
760 "../../%s/%s/brw_stats",
761 osd_obd->obd_type->typ_name, obd->obd_name);
763 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
764 "../../%s/%s/read_cache_enable",
765 osd_obd->obd_type->typ_name, obd->obd_name);
767 lprocfs_add_symlink("readcache_max_filesize",
769 "../../%s/%s/readcache_max_filesize",
770 osd_obd->obd_type->typ_name, obd->obd_name);
772 lprocfs_add_symlink("writethrough_cache_enable",
774 "../../%s/%s/writethrough_cache_enable",
775 osd_obd->obd_type->typ_name, obd->obd_name);
779 * Cleanup all procfs entries in OFD.
781 * \param[in] ofd OFD device
783 static void ofd_procfs_fini(struct ofd_device *ofd)
785 struct obd_device *obd = ofd_obd(ofd);
787 lprocfs_free_per_client_stats(obd);
788 lprocfs_obd_cleanup(obd);
789 lprocfs_free_obd_stats(obd);
790 lprocfs_job_stats_fini(obd);
794 * Stop SEQ/FID server on OFD.
796 * \param[in] env execution environment
797 * \param[in] ofd OFD device
799 * \retval 0 if successful
800 * \retval negative value on error
802 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
804 return seq_site_fini(env, &ofd->ofd_seq_site);
808 * Start SEQ/FID server on OFD.
810 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
811 * It also connects to the master server to get own FID sequence (SEQ) range
812 * to this particular OFD. Typically that happens when the OST is first
813 * formatted or in the rare case that it exhausts the local sequence range.
815 * The sequence range is allocated out to the MDTs for OST object allocations,
816 * and not directly to the clients.
818 * \param[in] env execution environment
819 * \param[in] ofd OFD device
821 * \retval 0 if successful
822 * \retval negative value on error
824 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
826 struct seq_server_site *ss = &ofd->ofd_seq_site;
827 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
828 char *obd_name = ofd_name(ofd);
832 ss = &ofd->ofd_seq_site;
833 lu->ld_site->ld_seq_site = ss;
834 ss->ss_lu = lu->ld_site;
835 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
837 OBD_ALLOC(name, sizeof(obd_name) * 2 + 10);
841 OBD_ALLOC_PTR(ss->ss_server_seq);
842 if (ss->ss_server_seq == NULL)
843 GOTO(out_name, rc = -ENOMEM);
845 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
846 LUSTRE_SEQ_SERVER, ss);
848 CERROR("%s : seq server init error %d\n", obd_name, rc);
849 GOTO(out_server, rc);
851 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
853 OBD_ALLOC_PTR(ss->ss_client_seq);
854 if (ss->ss_client_seq == NULL)
855 GOTO(out_server, rc = -ENOMEM);
858 * It always printed as "%p", so that the name is unique in the kernel,
859 * even if the filesystem is mounted twice. So sizeof(.) * 2 is enough.
861 snprintf(name, sizeof(obd_name) * 2 + 7, "%p-super", obd_name);
862 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
865 CERROR("%s : seq client init error %d\n", obd_name, rc);
866 GOTO(out_client, rc);
869 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
873 seq_client_fini(ss->ss_client_seq);
874 OBD_FREE_PTR(ss->ss_client_seq);
875 ss->ss_client_seq = NULL;
877 seq_server_fini(ss->ss_server_seq, env);
878 OBD_FREE_PTR(ss->ss_server_seq);
879 ss->ss_server_seq = NULL;
882 OBD_FREE(name, sizeof(obd_name) * 2 + 10);
888 * OFD request handler for OST_SET_INFO RPC.
890 * This is OFD-specific part of request handling
892 * \param[in] tsi target session environment for this request
894 * \retval 0 if successful
895 * \retval negative value on error
897 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
899 struct ptlrpc_request *req = tgt_ses_req(tsi);
900 struct ost_body *body = NULL, *repbody;
901 void *key, *val = NULL;
902 int keylen, vallen, rc = 0;
903 bool is_grant_shrink;
907 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
909 DEBUG_REQ(D_HA, req, "no set_info key");
910 RETURN(err_serious(-EFAULT));
912 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
915 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
917 DEBUG_REQ(D_HA, req, "no set_info val");
918 RETURN(err_serious(-EFAULT));
920 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
923 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
925 /* In this case the value is actually an RMF_OST_BODY, so we
926 * transmutate the type of this PTLRPC */
927 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
929 rc = req_capsule_server_pack(tsi->tsi_pill);
933 if (is_grant_shrink) {
934 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
936 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
939 /** handle grant shrink, similar to a read request */
940 ofd_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
942 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
944 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
946 } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
947 rc = tgt_adapt_sptlrpc_conf(tsi->tsi_tgt, 0);
949 CERROR("%s: Unsupported key %s\n",
950 tgt_name(tsi->tsi_tgt), (char *)key);
953 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
960 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
962 * This function returns a list of extents which describes how a file's
963 * blocks are laid out on the disk.
965 * \param[in] env execution environment
966 * \param[in] ofd OFD device
967 * \param[in] fid FID of object
968 * \param[in] fiemap fiemap structure to fill with data
970 * \retval 0 if \a fiemap is filled with data successfully
971 * \retval negative value on error
973 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
974 struct lu_fid *fid, struct fiemap *fiemap)
976 struct ofd_object *fo;
979 fo = ofd_object_find(env, ofd, fid);
981 CERROR("%s: error finding object "DFID"\n",
982 ofd_name(ofd), PFID(fid));
986 ofd_read_lock(env, fo);
987 if (ofd_object_exists(fo))
988 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
991 ofd_read_unlock(env, fo);
992 ofd_object_put(env, fo);
996 struct locked_region {
997 struct list_head list;
998 struct lustre_handle lh;
1002 * Lock single extent and save lock handle in the list.
1004 * This is supplemental function for lock_zero_regions(). It allocates
1005 * new locked_region structure and locks it with extent lock, then adds
1006 * it to the list of all such regions.
1008 * \param[in] ns LDLM namespace
1009 * \param[in] res_id resource ID
1010 * \param[in] begin start of region
1011 * \param[in] end end of region
1012 * \param[in] locked list head of regions list
1014 * \retval 0 if successful locking
1015 * \retval negative value on error
1017 static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1018 unsigned long long begin, unsigned long long end,
1019 struct list_head *locked)
1021 struct locked_region *region = NULL;
1025 LASSERT(begin <= end);
1026 OBD_ALLOC_PTR(region);
1030 rc = tgt_extent_lock(ns, res_id, begin, end, ®ion->lh,
1035 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
1037 list_add(®ion->list, locked);
1043 * Lock the sparse areas of given resource.
1045 * The locking of sparse areas will cause dirty data to be flushed back from
1046 * clients. This is used when getting the FIEMAP of an object to make sure
1047 * there is no unaccounted cached data on clients.
1049 * This function goes through \a fiemap list of extents and locks only sparse
1050 * areas between extents.
1052 * \param[in] ns LDLM namespace
1053 * \param[in] res_id resource ID
1054 * \param[in] fiemap file extents mapping on disk
1055 * \param[in] locked list head of regions list
1057 * \retval 0 if successful
1058 * \retval negative value on error
1060 static int lock_zero_regions(struct ldlm_namespace *ns,
1061 struct ldlm_res_id *res_id,
1062 struct fiemap *fiemap,
1063 struct list_head *locked)
1065 __u64 begin = fiemap->fm_start;
1068 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1072 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1073 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1074 if (fiemap_start[i].fe_logical > begin) {
1075 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1076 begin, fiemap_start[i].fe_logical);
1077 rc = lock_region(ns, res_id, begin,
1078 fiemap_start[i].fe_logical, locked);
1083 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1086 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1087 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1088 begin, fiemap->fm_start + fiemap->fm_length);
1089 rc = lock_region(ns, res_id, begin,
1090 fiemap->fm_start + fiemap->fm_length, locked);
1097 * Unlock all previously locked sparse areas for given resource.
1099 * This function goes through list of locked regions, unlocking and freeing
1102 * \param[in] ns LDLM namespace
1103 * \param[in] locked list head of regions list
1106 unlock_zero_regions(struct ldlm_namespace *ns, struct list_head *locked)
1108 struct locked_region *entry, *temp;
1110 list_for_each_entry_safe(entry, temp, locked, list) {
1111 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
1112 tgt_extent_unlock(&entry->lh, LCK_PR);
1113 list_del(&entry->list);
1114 OBD_FREE_PTR(entry);
1119 * OFD request handler for OST_GET_INFO RPC.
1121 * This is OFD-specific part of request handling. The OFD-specific keys are:
1122 * - KEY_LAST_ID (obsolete)
1126 * This function reads needed data from storage and fills reply with it.
1128 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1129 * and is kept for compatibility.
1131 * \param[in] tsi target session environment for this request
1133 * \retval 0 if successful
1134 * \retval negative value on error
1136 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1138 struct obd_export *exp = tsi->tsi_exp;
1139 struct ofd_device *ofd = ofd_exp(exp);
1140 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1143 int replylen, rc = 0;
1147 /* this common part for get_info rpc */
1148 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1150 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1151 RETURN(err_serious(-EPROTO));
1153 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1156 if (KEY_IS(KEY_LAST_ID)) {
1158 struct ofd_seq *oseq;
1160 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1161 rc = req_capsule_server_pack(tsi->tsi_pill);
1163 RETURN(err_serious(rc));
1165 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1167 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1168 (u64)exp->exp_filter_data.fed_group);
1172 *last_id = ofd_seq_last_oid(oseq);
1173 ofd_seq_put(tsi->tsi_env, oseq);
1174 } else if (KEY_IS(KEY_FIEMAP)) {
1175 struct ll_fiemap_info_key *fm_key;
1176 struct fiemap *fiemap;
1179 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1181 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1182 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1184 RETURN(err_serious(rc));
1186 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1188 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1190 replylen = fiemap_count_to_size(
1191 fm_key->lfik_fiemap.fm_extent_count);
1192 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1193 RCL_SERVER, replylen);
1195 rc = req_capsule_server_pack(tsi->tsi_pill);
1197 RETURN(err_serious(rc));
1199 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1203 *fiemap = fm_key->lfik_fiemap;
1204 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1206 /* LU-3219: Lock the sparse areas to make sure dirty
1207 * flushed back from client, then call fiemap again. */
1208 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1209 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1210 struct list_head locked;
1212 INIT_LIST_HEAD(&locked);
1213 ost_fid_build_resid(fid, &fti->fti_resid);
1214 rc = lock_zero_regions(ofd->ofd_namespace,
1215 &fti->fti_resid, fiemap,
1217 if (rc == 0 && !list_empty(&locked)) {
1218 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1220 unlock_zero_regions(ofd->ofd_namespace,
1224 } else if (KEY_IS(KEY_LAST_FID)) {
1225 struct ofd_device *ofd = ofd_exp(exp);
1226 struct ofd_seq *oseq;
1230 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1231 rc = req_capsule_server_pack(tsi->tsi_pill);
1233 RETURN(err_serious(rc));
1235 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1237 RETURN(err_serious(-EPROTO));
1239 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1241 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1245 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1246 ostid_seq(&fti->fti_ostid));
1248 RETURN(PTR_ERR(oseq));
1250 rc = ostid_to_fid(fid, &oseq->os_oi,
1251 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1255 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1258 ofd_seq_put(tsi->tsi_env, oseq);
1260 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1264 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1271 * OFD request handler for OST_GETATTR RPC.
1273 * This is OFD-specific part of request handling. It finds the OFD object
1274 * by its FID, gets attributes from storage and packs result to the reply.
1276 * \param[in] tsi target session environment for this request
1278 * \retval 0 if successful
1279 * \retval negative value on error
1281 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1283 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1284 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1285 struct ost_body *repbody;
1286 struct lustre_handle lh = { 0 };
1287 struct ofd_object *fo;
1289 enum ldlm_mode lock_mode = LCK_PR;
1294 LASSERT(tsi->tsi_ost_body != NULL);
1296 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1297 if (repbody == NULL)
1300 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1301 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1303 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1304 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1307 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1310 rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace,
1311 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1317 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1319 GOTO(out, rc = PTR_ERR(fo));
1321 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1325 obdo_from_la(&repbody->oa, &fti->fti_attr,
1326 OFD_VALID_FLAGS | LA_UID | LA_GID);
1328 /* Store object version in reply */
1329 curr_version = dt_version_get(tsi->tsi_env,
1330 ofd_object_child(fo));
1331 if ((__s64)curr_version != -EOPNOTSUPP) {
1332 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1333 repbody->oa.o_data_version = curr_version;
1337 ofd_object_put(tsi->tsi_env, fo);
1340 tgt_extent_unlock(&lh, lock_mode);
1342 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1345 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1346 repbody->oa.o_flags = OBD_FL_FLUSH;
1352 * OFD request handler for OST_SETATTR RPC.
1354 * This is OFD-specific part of request handling. It finds the OFD object
1355 * by its FID, sets attributes from request and packs result to the reply.
1357 * \param[in] tsi target session environment for this request
1359 * \retval 0 if successful
1360 * \retval negative value on error
1362 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1364 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1365 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1366 struct ost_body *body = tsi->tsi_ost_body;
1367 struct ost_body *repbody;
1368 struct ldlm_resource *res;
1369 struct ofd_object *fo;
1370 struct filter_fid *ff = NULL;
1375 LASSERT(body != NULL);
1377 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1378 if (repbody == NULL)
1381 repbody->oa.o_oi = body->oa.o_oi;
1382 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1384 /* This would be very bad - accidentally truncating a file when
1385 * changing the time or similar - bug 12203. */
1386 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1387 body->oa.o_size != OBD_OBJECT_EOF) {
1388 static char mdsinum[48];
1390 if (body->oa.o_valid & OBD_MD_FLFID)
1391 snprintf(mdsinum, sizeof(mdsinum) - 1,
1392 "of parent "DFID, body->oa.o_parent_seq,
1393 body->oa.o_parent_oid, 0);
1397 CERROR("%s: setattr from %s is trying to truncate object "DFID
1398 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1399 PFID(&tsi->tsi_fid), mdsinum);
1403 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1405 GOTO(out, rc = PTR_ERR(fo));
1407 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1408 fti->fti_attr.la_valid &= ~LA_TYPE;
1410 if (body->oa.o_valid & OBD_MD_FLFID) {
1411 ff = &fti->fti_mds_fid;
1412 ofd_prepare_fidea(ff, &body->oa);
1415 /* setting objects attributes (including owner/group) */
1416 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff);
1420 obdo_from_la(&repbody->oa, &fti->fti_attr,
1421 OFD_VALID_FLAGS | LA_UID | LA_GID);
1423 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1427 ofd_object_put(tsi->tsi_env, fo);
1430 /* we do not call this before to avoid lu_object_find() in
1431 * ->lvbo_update() holding another reference on the object.
1432 * otherwise concurrent destroy can make the object unavailable
1433 * for 2nd lu_object_find() waiting for the first reference
1434 * to go... deadlock! */
1435 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1436 &tsi->tsi_resid, LDLM_EXTENT, 0);
1438 ldlm_res_lvbo_update(res, NULL, 0);
1439 ldlm_resource_putref(res);
1446 * Destroy OST orphans.
1448 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1449 * set then we must destroy possible orphaned objects.
1451 * \param[in] env execution environment
1452 * \param[in] exp OBD export
1453 * \param[in] ofd OFD device
1454 * \param[in] oa obdo structure for reply
1456 * \retval 0 if successful
1457 * \retval negative value on error
1459 static int ofd_orphans_destroy(const struct lu_env *env,
1460 struct obd_export *exp,
1461 struct ofd_device *ofd, struct obdo *oa)
1463 struct ofd_thread_info *info = ofd_info(env);
1464 struct lu_fid *fid = &info->fti_fid;
1465 struct ost_id *oi = &oa->o_oi;
1466 struct ofd_seq *oseq;
1467 u64 seq = ostid_seq(oi);
1468 u64 end_id = ostid_id(oi);
1476 oseq = ofd_seq_get(ofd, seq);
1478 CERROR("%s: Can not find seq for "DOSTID"\n",
1479 ofd_name(ofd), POSTID(oi));
1484 last = ofd_seq_last_oid(oseq);
1487 LASSERT(exp != NULL);
1488 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1490 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1493 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1494 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1496 while (oid > end_id) {
1497 rc = fid_set_id(fid, oid);
1498 if (unlikely(rc != 0))
1501 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1502 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1503 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1504 /* this is pretty fatal... */
1505 CEMERG("%s: error destroying precreated id "
1507 ofd_name(ofd), PFID(fid), rc);
1511 ofd_seq_last_oid_set(oseq, oid);
1512 /* update last_id on disk periodically so that if we
1513 * restart * we don't need to re-scan all of the just
1514 * deleted objects. */
1515 if ((oid & 511) == 0)
1516 ofd_seq_last_oid_write(env, ofd, oseq);
1520 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1521 ofd_name(ofd), seq, oid);
1525 ofd_seq_last_oid_set(oseq, oid);
1526 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1528 /* don't reuse orphan object, return last used objid */
1529 ostid_set_id(oi, last);
1536 ofd_seq_put(env, oseq);
1541 * OFD request handler for OST_CREATE RPC.
1543 * This is OFD-specific part of request handling. Its main purpose is to
1544 * create new data objects on OST, but it also used to destroy orphans.
1546 * \param[in] tsi target session environment for this request
1548 * \retval 0 if successful
1549 * \retval negative value on error
1551 static int ofd_create_hdl(struct tgt_session_info *tsi)
1553 struct ptlrpc_request *req = tgt_ses_req(tsi);
1554 struct ost_body *repbody;
1555 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1556 struct obdo *rep_oa;
1557 struct obd_export *exp = tsi->tsi_exp;
1558 struct ofd_device *ofd = ofd_exp(exp);
1559 u64 seq = ostid_seq(&oa->o_oi);
1560 u64 oid = ostid_id(&oa->o_oi);
1561 struct ofd_seq *oseq;
1568 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1571 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1572 if (repbody == NULL)
1575 down_read(&ofd->ofd_lastid_rwsem);
1576 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1577 * we may do that in the future.
1578 * Return -ENOSPC until the LAST_ID rebuilt. */
1579 if (unlikely(ofd->ofd_lastid_rebuilding))
1580 GOTO(out_sem, rc = -ENOSPC);
1582 rep_oa = &repbody->oa;
1583 rep_oa->o_oi = oa->o_oi;
1585 LASSERT(seq >= FID_SEQ_OST_MDT0);
1586 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1588 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1590 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1592 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1593 ofd_name(ofd), seq, PTR_ERR(oseq));
1594 GOTO(out_sem, rc = -EINVAL);
1597 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1598 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1599 if (!ofd_obd(ofd)->obd_recovering ||
1600 oid > ofd_seq_last_oid(oseq)) {
1601 CERROR("%s: recreate objid "DOSTID" > last id "LPU64
1602 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1603 ofd_seq_last_oid(oseq));
1604 GOTO(out_nolock, rc = -EINVAL);
1606 /* Do nothing here, we re-create objects during recovery
1607 * upon write replay, see ofd_preprw_write() */
1608 GOTO(out_nolock, rc = 0);
1610 /* former ofd_handle_precreate */
1611 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1612 (oa->o_flags & OBD_FL_DELORPHAN)) {
1613 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1615 /* destroy orphans */
1616 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1617 exp->exp_conn_cnt) {
1618 CERROR("%s: dropping old orphan cleanup request\n",
1620 GOTO(out_nolock, rc = 0);
1622 /* This causes inflight precreates to abort and drop lock */
1623 oseq->os_destroys_in_progress = 1;
1624 mutex_lock(&oseq->os_create_lock);
1625 if (!oseq->os_destroys_in_progress) {
1626 CERROR("%s:["LPU64"] destroys_in_progress already"
1627 " cleared\n", ofd_name(ofd), seq);
1628 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1631 diff = oid - ofd_seq_last_oid(oseq);
1632 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1633 ofd_seq_last_oid(oseq), diff);
1634 if (-diff > OST_MAX_PRECREATE) {
1635 /* Let MDS know that we are so far ahead. */
1636 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq) + 1);
1638 } else if (diff < 0) {
1639 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1641 oseq->os_destroys_in_progress = 0;
1643 /* XXX: Used by MDS for the first time! */
1644 oseq->os_destroys_in_progress = 0;
1647 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1648 ofd->ofd_lastid_gen)) {
1649 /* Keep the export ref so we can send the reply. */
1650 ofd_obd_disconnect(class_export_get(exp));
1651 GOTO(out_nolock, rc = -ENOTCONN);
1654 mutex_lock(&oseq->os_create_lock);
1655 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1656 exp->exp_conn_cnt) {
1657 CERROR("%s: dropping old precreate request\n",
1661 /* only precreate if seq is 0, IDIF or normal and also o_id
1662 * must be specfied */
1663 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1664 !fid_seq_is_idif(seq)) || oid == 0) {
1665 diff = 1; /* shouldn't we create this right now? */
1667 diff = oid - ofd_seq_last_oid(oseq);
1668 /* Do sync create if the seq is about to used up */
1669 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1670 if (unlikely(oid >= IDIF_MAX_OID - 1))
1672 } else if (fid_seq_is_norm(seq)) {
1674 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1677 CERROR("%s : invalid o_seq "DOSTID"\n",
1678 ofd_name(ofd), POSTID(&oa->o_oi));
1679 GOTO(out, rc = -EINVAL);
1684 CERROR("%s: invalid precreate request for "
1685 DOSTID", last_id " LPU64 ". "
1686 "Likely MDS last_id corruption\n",
1687 ofd_name(ofd), POSTID(&oa->o_oi),
1688 ofd_seq_last_oid(oseq));
1689 GOTO(out, rc = -EINVAL);
1694 cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
1699 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1700 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1701 /* don't enforce grant during orphan recovery */
1702 granted = ofd_grant_create(tsi->tsi_env,
1703 ofd_obd(ofd)->obd_self_export,
1708 CDEBUG(D_HA, "%s: failed to acquire grant "
1709 "space for precreate (%d): rc = %d\n",
1710 ofd_name(ofd), diff, rc);
1715 /* This can happen if a new OST is formatted and installed
1716 * in place of an old one at the same index. Instead of
1717 * precreating potentially millions of deleted old objects
1718 * (possibly filling the OST), only precreate the last batch.
1719 * LFSCK will eventually clean up any orphans. LU-14 */
1720 if (diff > 5 * OST_MAX_PRECREATE) {
1721 diff = OST_MAX_PRECREATE / 2;
1722 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1723 "OST replaced or reformatted: "
1724 "LFSCK will clean up",
1727 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1728 "%u larger than the LAST_ID "DOSTID", only "
1729 "precreating the last %u objects.\n",
1730 ofd_name(ofd), POSTID(&oa->o_oi),
1731 5 * OST_MAX_PRECREATE,
1732 POSTID(&oseq->os_oi), diff);
1733 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1737 next_id = ofd_seq_last_oid(oseq) + 1;
1738 count = ofd_precreate_batch(ofd, diff);
1740 CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1741 " at "LPU64"\n", ofd_name(ofd),
1742 count, seq, next_id);
1744 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1745 && cfs_time_after(jiffies, enough_time)) {
1746 CDEBUG(D_HA, "%s: Slow creates, %d/%d objects"
1747 " created at a rate of %d/s\n",
1748 ofd_name(ofd), created, diff + created,
1749 created / DISK_TIMEOUT);
1753 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1754 oseq, count, sync_trans);
1758 } else if (rc < 0) {
1764 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1765 LCONSOLE_WARN("%s: can't create the same count of"
1766 " objects when replaying the request"
1767 " (diff is %d). see LU-4621\n",
1768 ofd_name(ofd), diff);
1771 /* some objects got created, we can return
1772 * them, even if last creation failed */
1775 CERROR("%s: unable to precreate: rc = %d\n",
1778 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1779 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1780 ofd_grant_commit(ofd_obd(ofd)->obd_self_export, granted,
1785 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1788 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1791 mutex_unlock(&oseq->os_create_lock);
1794 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1795 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1796 struct lu_fid *fid = &info->fti_fid;
1798 /* For compatible purpose, it needs to convert back to
1799 * OST ID before put it on wire. */
1800 *fid = rep_oa->o_oi.oi_fid;
1801 fid_to_ostid(fid, &rep_oa->o_oi);
1803 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1805 ofd_seq_put(tsi->tsi_env, oseq);
1808 up_read(&ofd->ofd_lastid_rwsem);
1813 * OFD request handler for OST_DESTROY RPC.
1815 * This is OFD-specific part of request handling. It destroys data objects
1816 * related to destroyed object on MDT.
1818 * \param[in] tsi target session environment for this request
1820 * \retval 0 if successful
1821 * \retval negative value on error
1823 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1825 const struct ost_body *body = tsi->tsi_ost_body;
1826 struct ost_body *repbody;
1827 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1828 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1829 struct lu_fid *fid = &fti->fti_fid;
1836 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1839 /* This is old case for clients before Lustre 2.4 */
1840 /* If there's a DLM request, cancel the locks mentioned in it */
1841 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1843 struct ldlm_request *dlm;
1845 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1848 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1851 *fid = body->oa.o_oi.oi_fid;
1852 oid = ostid_id(&body->oa.o_oi);
1855 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1857 /* check that o_misc makes sense */
1858 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1859 count = body->oa.o_misc;
1861 count = 1; /* default case - single destroy */
1863 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1864 POSTID(&body->oa.o_oi), count);
1869 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1870 if (lrc == -ENOENT) {
1872 "%s: destroying non-existent object "DFID"\n",
1873 ofd_name(ofd), PFID(fid));
1874 /* rewrite rc with -ENOENT only if it is 0 */
1877 } else if (lrc != 0) {
1878 CERROR("%s: error destroying object "DFID": %d\n",
1879 ofd_name(ofd), PFID(fid), lrc);
1885 lrc = fid_set_id(fid, oid);
1886 if (unlikely(lrc != 0 && count > 0))
1887 GOTO(out, rc = lrc);
1890 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1896 fid_to_ostid(fid, &repbody->oa.o_oi);
1901 * OFD request handler for OST_STATFS RPC.
1903 * This function gets statfs data from storage as part of request
1906 * \param[in] tsi target session environment for this request
1908 * \retval 0 if successful
1909 * \retval negative value on error
1911 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1913 struct obd_statfs *osfs;
1918 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1920 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1921 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
1923 CERROR("%s: statfs failed: rc = %d\n",
1924 tgt_name(tsi->tsi_tgt), rc);
1926 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1929 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1936 * OFD request handler for OST_SYNC RPC.
1938 * Sync object data or all filesystem data to the disk and pack the
1941 * \param[in] tsi target session environment for this request
1943 * \retval 0 if successful
1944 * \retval negative value on error
1946 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1948 struct ost_body *body = tsi->tsi_ost_body;
1949 struct ost_body *repbody;
1950 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1951 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1952 struct ofd_object *fo = NULL;
1957 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1959 /* if no objid is specified, it means "sync whole filesystem" */
1960 if (!fid_is_zero(&tsi->tsi_fid)) {
1961 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1963 RETURN(PTR_ERR(fo));
1966 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1967 fo != NULL ? ofd_object_child(fo) : NULL,
1968 repbody->oa.o_size, repbody->oa.o_blocks);
1972 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1977 repbody->oa.o_oi = body->oa.o_oi;
1978 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1980 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1982 obdo_from_la(&repbody->oa, &fti->fti_attr,
1985 /* don't return rc from getattr */
1990 ofd_object_put(tsi->tsi_env, fo);
1995 * OFD request handler for OST_PUNCH RPC.
1997 * This is part of request processing. Validate request fields,
1998 * punch (truncate) the given OFD object and pack reply.
2000 * \param[in] tsi target session environment for this request
2002 * \retval 0 if successful
2003 * \retval negative value on error
2005 static int ofd_punch_hdl(struct tgt_session_info *tsi)
2007 const struct obdo *oa = &tsi->tsi_ost_body->oa;
2008 struct ost_body *repbody;
2009 struct ofd_thread_info *info = tsi2ofd_info(tsi);
2010 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
2011 struct ldlm_resource *res;
2012 struct ofd_object *fo;
2013 struct filter_fid *ff = NULL;
2015 struct lustre_handle lh = { 0, };
2022 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
2024 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
2025 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
2027 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2028 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2029 RETURN(err_serious(-EPROTO));
2031 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2032 if (repbody == NULL)
2033 RETURN(err_serious(-ENOMEM));
2035 /* punch start,end are passed in o_size,o_blocks throught wire */
2039 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2042 /* standard truncate optimization: if file body is completely
2043 * destroyed, don't send data back to the server. */
2045 flags |= LDLM_FL_AST_DISCARD_DATA;
2047 repbody->oa.o_oi = oa->o_oi;
2048 repbody->oa.o_valid = OBD_MD_FLID;
2050 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2051 oa->o_flags & OBD_FL_SRVLOCK;
2054 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
2060 CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
2061 ", start = "LPD64", end = "LPD64"\n", PFID(&tsi->tsi_fid),
2062 oa->o_valid, start, end);
2064 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2067 GOTO(out, rc = PTR_ERR(fo));
2069 la_from_obdo(&info->fti_attr, oa,
2070 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2071 info->fti_attr.la_size = start;
2072 info->fti_attr.la_valid |= LA_SIZE;
2074 if (oa->o_valid & OBD_MD_FLFID) {
2075 ff = &info->fti_mds_fid;
2076 ofd_prepare_fidea(ff, oa);
2079 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2080 ff, (struct obdo *)oa);
2084 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2088 ofd_object_put(tsi->tsi_env, fo);
2091 tgt_extent_unlock(&lh, LCK_PW);
2093 /* we do not call this before to avoid lu_object_find() in
2094 * ->lvbo_update() holding another reference on the object.
2095 * otherwise concurrent destroy can make the object unavailable
2096 * for 2nd lu_object_find() waiting for the first reference
2097 * to go... deadlock! */
2098 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2101 ldlm_res_lvbo_update(res, NULL, 0);
2102 ldlm_resource_putref(res);
2108 static int ofd_ladvise_prefetch(const struct lu_env *env,
2109 struct ofd_object *fo,
2110 __u64 start, __u64 end)
2112 struct ofd_thread_info *info = ofd_info(env);
2113 pgoff_t start_index, end_index, pages;
2114 struct niobuf_remote rnb;
2115 unsigned long nr_local;
2116 struct niobuf_local *lnb;
2122 OBD_ALLOC_LARGE(lnb, sizeof(*lnb) * PTLRPC_MAX_BRW_PAGES);
2126 ofd_read_lock(env, fo);
2127 if (!ofd_object_exists(fo))
2128 GOTO(out_unlock, rc = -ENOENT);
2130 rc = ofd_attr_get(env, fo, &info->fti_attr);
2132 GOTO(out_unlock, rc);
2134 if (end > info->fti_attr.la_size)
2135 end = info->fti_attr.la_size;
2138 GOTO(out_unlock, rc);
2140 /* We need page aligned offset and length */
2141 start_index = start >> PAGE_CACHE_SHIFT;
2142 end_index = (end - 1) >> PAGE_CACHE_SHIFT;
2143 pages = end_index - start_index + 1;
2145 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2146 PTLRPC_MAX_BRW_PAGES;
2147 rnb.rnb_offset = start_index << PAGE_CACHE_SHIFT;
2148 rnb.rnb_len = nr_local << PAGE_CACHE_SHIFT;
2149 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, 0);
2150 if (unlikely(rc < 0))
2153 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2154 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2157 start_index += nr_local;
2162 ofd_read_unlock(env, fo);
2163 OBD_FREE_LARGE(lnb, sizeof(*lnb) * PTLRPC_MAX_BRW_PAGES);
2168 * OFD request handler for OST_LADVISE RPC.
2170 * Tune cache or perfetch policies according to advices.
2172 * \param[in] tsi target session environment for this request
2174 * \retval 0 if successful
2175 * \retval negative errno on error
2177 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2179 struct ptlrpc_request *req = tgt_ses_req(tsi);
2180 struct obd_export *exp = tsi->tsi_exp;
2181 struct ofd_device *ofd = ofd_exp(exp);
2182 struct ost_body *body, *repbody;
2183 struct ofd_thread_info *info;
2184 struct ofd_object *fo;
2185 const struct lu_env *env = req->rq_svc_thread->t_env;
2187 struct lu_ladvise *ladvise;
2189 struct ladvise_hdr *ladvise_hdr;
2190 struct obd_ioobj ioo;
2191 struct lustre_handle lockh = { 0 };
2196 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2197 body = tsi->tsi_ost_body;
2199 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2200 RETURN(err_serious(-EPROTO));
2202 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2203 &RMF_OST_LADVISE_HDR);
2204 if (ladvise_hdr == NULL)
2205 RETURN(err_serious(-EPROTO));
2207 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2208 ladvise_hdr->lah_count < 1)
2209 RETURN(err_serious(-EPROTO));
2211 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2212 RETURN(err_serious(-EPROTO));
2214 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2215 if (ladvise == NULL)
2216 RETURN(err_serious(-EPROTO));
2218 num_advise = req_capsule_get_size(&req->rq_pill,
2219 &RMF_OST_LADVISE, RCL_CLIENT) /
2221 if (num_advise < ladvise_hdr->lah_count)
2222 RETURN(err_serious(-EPROTO));
2224 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2225 repbody->oa = body->oa;
2227 info = ofd_info_init(env, exp);
2229 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2230 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2234 fo = ofd_object_find(env, ofd, &info->fti_fid);
2239 LASSERT(fo != NULL);
2241 for (i = 0; i < num_advise; i++, ladvise++) {
2242 if (ladvise->lla_end <= ladvise->lla_start) {
2243 rc = err_serious(-EPROTO);
2247 /* Handle different advice types */
2248 switch (ladvise->lla_advice) {
2252 case LU_LADVISE_WILLREAD:
2253 ioo.ioo_oid = body->oa.o_oi;
2255 rc = tgt_extent_lock(exp->exp_obd->obd_namespace,
2258 ladvise->lla_end - 1,
2259 &lockh, LCK_PR, &flags);
2263 req->rq_status = ofd_ladvise_prefetch(env,
2267 tgt_extent_unlock(&lockh, LCK_PR);
2274 ofd_object_put(env, fo);
2275 req->rq_status = rc;
2280 * OFD request handler for OST_QUOTACTL RPC.
2282 * This is part of request processing to validate incoming request fields,
2283 * get the requested data from OSD and pack reply.
2285 * \param[in] tsi target session environment for this request
2287 * \retval 0 if successful
2288 * \retval negative value on error
2290 static int ofd_quotactl(struct tgt_session_info *tsi)
2292 struct obd_quotactl *oqctl, *repoqc;
2293 struct lu_nodemap *nodemap;
2299 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2301 RETURN(err_serious(-EPROTO));
2303 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2305 RETURN(err_serious(-ENOMEM));
2309 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2310 if (IS_ERR(nodemap))
2311 RETURN(PTR_ERR(nodemap));
2314 if (oqctl->qc_type == USRQUOTA)
2315 id = nodemap_map_id(nodemap, NODEMAP_UID,
2316 NODEMAP_CLIENT_TO_FS,
2318 else if (oqctl->qc_type == GRPQUOTA)
2319 id = nodemap_map_id(nodemap, NODEMAP_GID,
2320 NODEMAP_CLIENT_TO_FS,
2323 nodemap_putref(nodemap);
2325 if (repoqc->qc_id != id)
2326 swap(repoqc->qc_id, id);
2328 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2330 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2333 if (repoqc->qc_id != id)
2334 swap(repoqc->qc_id, id);
2340 * Calculate the amount of time for lock prolongation.
2342 * This is helper for ofd_prolong_extent_locks() function to get
2343 * the timeout extra time.
2345 * \param[in] req current request
2347 * \retval amount of time to extend the timeout with
2349 static inline int prolong_timeout(struct ptlrpc_request *req)
2351 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2354 return obd_timeout / 2;
2356 return 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"LPU64
2395 " ext("LPU64"->"LPU64")\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);
2414 lock->l_last_used = cfs_time_current();
2415 LDLM_LOCK_PUT(lock);
2419 ldlm_resource_prolong(data);
2424 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2426 * Determine if \a lock and the lock from request \a req are equivalent
2427 * by comparing their resource names, modes, and extents.
2429 * It is used to give priority to read and write RPCs being done
2430 * under this lock so that the client can drop the contended
2431 * lock more quickly and let other clients use it. This improves
2432 * overall performance in the case where the first client gets a
2433 * very large lock extent that prevents other clients from
2434 * submitting their writes.
2436 * \param[in] req ptlrpc_request being processed
2437 * \param[in] lock contended lock to match
2439 * \retval 1 if lock is matched
2440 * \retval 0 otherwise
2442 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2443 struct ldlm_lock *lock)
2445 struct niobuf_remote *rnb;
2446 struct obd_ioobj *ioo;
2447 enum ldlm_mode mode;
2448 struct ldlm_extent ext;
2449 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2453 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2454 LASSERT(ioo != NULL);
2456 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2457 LASSERT(rnb != NULL);
2459 ext.start = rnb->rnb_offset;
2460 rnb += ioo->ioo_bufcnt - 1;
2461 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2463 LASSERT(lock->l_resource != NULL);
2464 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2467 /* a bulk write can only hold a reference on a PW extent lock
2470 mode = LCK_PW | LCK_GROUP;
2471 if (opc == OST_READ)
2472 /* whereas a bulk read can be protected by either a PR or PW
2476 if (!(lock->l_granted_mode & mode))
2479 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2483 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2485 * Check for whether the given PTLRPC request (\a req) is blocking
2486 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2489 * \param[in] req the incoming request
2491 * \retval 1 if \a req is blocking an LDLM lock cancel
2492 * \retval 0 if it is not
2493 * \retval -ESTALE if lock is not found
2495 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2497 struct tgt_session_info *tsi;
2498 struct obd_ioobj *ioo;
2499 struct niobuf_remote *rnb;
2501 struct ldlm_prolong_args pa = { 0 };
2505 /* Don't use tgt_ses_info() to get session info, because lock_match()
2506 * can be called while request has no processing thread yet. */
2507 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2510 * Use LASSERT below because malformed RPCs should have
2511 * been filtered out in tgt_hpreq_handler().
2513 opc = lustre_msg_get_opc(req->rq_reqmsg);
2514 LASSERT(opc == OST_READ || opc == OST_WRITE);
2516 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2517 LASSERT(ioo != NULL);
2519 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2520 LASSERT(rnb != NULL);
2521 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2523 pa.lpa_mode = LCK_PW | LCK_GROUP;
2524 if (opc == OST_READ)
2525 pa.lpa_mode |= LCK_PR;
2527 pa.lpa_extent.start = rnb->rnb_offset;
2528 rnb += ioo->ioo_bufcnt - 1;
2529 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2531 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
2532 " ("LPU64"->"LPU64")\n", tgt_name(tsi->tsi_tgt),
2533 current->comm, PFID(&tsi->tsi_fid), pa.lpa_extent.start,
2536 ofd_prolong_extent_locks(tsi, &pa);
2538 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2539 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2541 if (pa.lpa_blocks_cnt > 0)
2544 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2548 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2550 * Called after the request has been handled. It refreshes lock timeout again
2551 * so that client has more time to send lock cancel RPC.
2553 * \param[in] req request which is being processed.
2555 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2557 ofd_rw_hpreq_check(req);
2561 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2563 * This function checks if the given lock is the same by its resname, mode
2564 * and extent as one taken from the request.
2565 * It is used to give priority to punch/truncate RPCs that might lead to
2566 * the fastest release of that lock when a lock is contended.
2568 * \param[in] req ptlrpc_request being processed
2569 * \param[in] lock contended lock to match
2571 * \retval 1 if lock is matched
2572 * \retval 0 otherwise
2574 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2575 struct ldlm_lock *lock)
2577 struct tgt_session_info *tsi;
2579 struct ldlm_extent ext;
2583 /* Don't use tgt_ses_info() to get session info, because lock_match()
2584 * can be called while request has no processing thread yet. */
2585 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2588 * Use LASSERT below because malformed RPCs should have
2589 * been filtered out in tgt_hpreq_handler().
2591 LASSERT(tsi->tsi_ost_body != NULL);
2592 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2593 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2596 oa = &tsi->tsi_ost_body->oa;
2597 ext.start = oa->o_size;
2598 ext.end = oa->o_blocks;
2600 LASSERT(lock->l_resource != NULL);
2601 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2604 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2607 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2611 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2613 * High-priority queue request check for whether the given punch request
2614 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2615 * covered by an LDLM lock.
2619 * \param[in] req the incoming request
2621 * \retval 1 if \a req is blocking an LDLM lock cancel
2622 * \retval 0 if it is not
2623 * \retval -ESTALE if lock is not found
2625 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2627 struct tgt_session_info *tsi;
2629 struct ldlm_prolong_args pa = { 0 };
2633 /* Don't use tgt_ses_info() to get session info, because lock_match()
2634 * can be called while request has no processing thread yet. */
2635 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2636 LASSERT(tsi != NULL);
2637 oa = &tsi->tsi_ost_body->oa;
2639 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2640 oa->o_flags & OBD_FL_SRVLOCK));
2642 pa.lpa_mode = LCK_PW | LCK_GROUP;
2643 pa.lpa_extent.start = oa->o_size;
2644 pa.lpa_extent.end = oa->o_blocks;
2647 "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
2648 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2649 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2651 ofd_prolong_extent_locks(tsi, &pa);
2653 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2654 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2656 if (pa.lpa_blocks_cnt > 0)
2659 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2663 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2665 * Called after the request has been handled. It refreshes lock timeout again
2666 * so that client has more time to send lock cancel RPC.
2668 * \param[in] req request which is being processed.
2670 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2672 ofd_punch_hpreq_check(req);
2675 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2676 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2677 .hpreq_check = ofd_rw_hpreq_check,
2678 .hpreq_fini = ofd_rw_hpreq_fini
2681 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2682 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2683 .hpreq_check = ofd_punch_hpreq_check,
2684 .hpreq_fini = ofd_punch_hpreq_fini
2688 * Assign high priority operations to an IO request.
2690 * Check if the incoming request is a candidate for
2691 * high-priority processing. If it is, assign it a high
2692 * priority operations table.
2694 * \param[in] tsi target session environment for this request
2696 static void ofd_hp_brw(struct tgt_session_info *tsi)
2698 struct niobuf_remote *rnb;
2699 struct obd_ioobj *ioo;
2703 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2704 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2705 if (ioo->ioo_bufcnt > 0) {
2706 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2707 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2709 /* no high priority if server lock is needed */
2710 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2711 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2715 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2719 * Assign high priority operations to an punch request.
2721 * Check if the incoming request is a candidate for
2722 * high-priority processing. If it is, assign it a high
2723 * priority operations table.
2725 * \param[in] tsi target session environment for this request
2727 static void ofd_hp_punch(struct tgt_session_info *tsi)
2729 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2730 /* no high-priority if server lock is needed */
2731 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2732 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2733 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2734 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2736 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2739 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2740 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2741 #define OST_BRW_READ OST_READ
2742 #define OST_BRW_WRITE OST_WRITE
2745 * Table of OFD-specific request handlers
2747 * This table contains all opcodes accepted by OFD and
2748 * specifies handlers for them. The tgt_request_handler()
2749 * uses such table from each target to process incoming
2752 static struct tgt_handler ofd_tgt_handlers[] = {
2753 TGT_RPC_HANDLER(OST_FIRST_OPC,
2754 0, OST_CONNECT, tgt_connect,
2755 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2756 TGT_RPC_HANDLER(OST_FIRST_OPC,
2757 0, OST_DISCONNECT, tgt_disconnect,
2758 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2759 TGT_RPC_HANDLER(OST_FIRST_OPC,
2760 0, OST_SET_INFO, ofd_set_info_hdl,
2761 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2762 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2763 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
2764 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2765 OST_SETATTR, ofd_setattr_hdl),
2766 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2767 OST_CREATE, ofd_create_hdl),
2768 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2769 OST_DESTROY, ofd_destroy_hdl),
2770 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
2771 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2772 OST_BRW_READ, tgt_brw_read,
2774 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2775 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
2777 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2778 OST_PUNCH, ofd_punch_hdl,
2780 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
2781 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
2782 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_LADVISE, ofd_ladvise_hdl),
2785 static struct tgt_opc_slice ofd_common_slice[] = {
2787 .tos_opc_start = OST_FIRST_OPC,
2788 .tos_opc_end = OST_LAST_OPC,
2789 .tos_hs = ofd_tgt_handlers
2792 .tos_opc_start = OBD_FIRST_OPC,
2793 .tos_opc_end = OBD_LAST_OPC,
2794 .tos_hs = tgt_obd_handlers
2797 .tos_opc_start = LDLM_FIRST_OPC,
2798 .tos_opc_end = LDLM_LAST_OPC,
2799 .tos_hs = tgt_dlm_handlers
2802 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2803 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2804 .tos_hs = tgt_out_handlers
2807 .tos_opc_start = SEQ_FIRST_OPC,
2808 .tos_opc_end = SEQ_LAST_OPC,
2809 .tos_hs = seq_handlers
2812 .tos_opc_start = LFSCK_FIRST_OPC,
2813 .tos_opc_end = LFSCK_LAST_OPC,
2814 .tos_hs = tgt_lfsck_handlers
2817 .tos_opc_start = SEC_FIRST_OPC,
2818 .tos_opc_end = SEC_LAST_OPC,
2819 .tos_hs = tgt_sec_ctx_handlers
2826 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2827 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2830 * Implementation of lu_context_key::lct_key_exit.
2832 * Optional method called on lu_context_exit() for all allocated
2834 * It is used in OFD to sanitize context values which may be re-used
2835 * during another request processing by the same thread.
2837 * \param[in] ctx execution context
2838 * \param[in] key context key
2839 * \param[in] data ofd_thread_info
2841 static void ofd_key_exit(const struct lu_context *ctx,
2842 struct lu_context_key *key, void *data)
2844 struct ofd_thread_info *info = data;
2846 info->fti_env = NULL;
2847 info->fti_exp = NULL;
2850 info->fti_pre_version = 0;
2853 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2856 struct lu_context_key ofd_thread_key = {
2857 .lct_tags = LCT_DT_THREAD,
2858 .lct_init = ofd_key_init,
2859 .lct_fini = ofd_key_fini,
2860 .lct_exit = ofd_key_exit
2864 * Initialize OFD device according to parameters in the config log \a cfg.
2866 * This is the main starting point of OFD initialization. It fills all OFD
2867 * parameters with their initial values and calls other initializing functions
2868 * to set up all OFD subsystems.
2870 * \param[in] env execution environment
2871 * \param[in] m OFD device
2872 * \param[in] ldt LU device type of OFD
2873 * \param[in] cfg configuration log
2875 * \retval 0 if successful
2876 * \retval negative value on error
2878 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2879 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2881 const char *dev = lustre_cfg_string(cfg, 0);
2882 struct ofd_thread_info *info = NULL;
2883 struct obd_device *obd;
2884 struct obd_statfs *osfs;
2890 obd = class_name2obd(dev);
2892 CERROR("Cannot find obd with name %s\n", dev);
2896 rc = lu_env_refill((struct lu_env *)env);
2900 obd->u.obt.obt_magic = OBT_MAGIC;
2902 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2903 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2905 spin_lock_init(&m->ofd_flags_lock);
2906 m->ofd_raid_degraded = 0;
2907 m->ofd_syncjournal = 0;
2909 m->ofd_grant_compat_disable = 0;
2910 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2913 spin_lock_init(&m->ofd_osfs_lock);
2914 m->ofd_osfs_age = cfs_time_shift_64(-1000);
2915 m->ofd_osfs_unstable = 0;
2916 m->ofd_statfs_inflight = 0;
2917 m->ofd_osfs_inflight = 0;
2920 spin_lock_init(&m->ofd_grant_lock);
2921 m->ofd_tot_dirty = 0;
2922 m->ofd_tot_granted = 0;
2923 m->ofd_tot_pending = 0;
2924 m->ofd_seq_count = 0;
2925 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2926 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2927 spin_lock_init(&m->ofd_inconsistency_lock);
2929 spin_lock_init(&m->ofd_batch_lock);
2930 init_rwsem(&m->ofd_lastid_rwsem);
2932 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2933 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2934 /* set this lu_device to obd, because error handling need it */
2935 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2937 rc = ofd_procfs_init(m);
2939 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2943 /* No connection accepted until configurations will finish */
2944 spin_lock(&obd->obd_dev_lock);
2945 obd->obd_no_conn = 1;
2946 spin_unlock(&obd->obd_dev_lock);
2947 obd->obd_replayable = 1;
2948 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2949 char *str = lustre_cfg_string(cfg, 4);
2951 if (strchr(str, 'n')) {
2952 CWARN("%s: recovery disabled\n", obd->obd_name);
2953 obd->obd_replayable = 0;
2957 info = ofd_info_init(env, NULL);
2959 GOTO(err_fini_proc, rc = -EFAULT);
2961 rc = ofd_stack_init(env, m, cfg);
2963 CERROR("Can't init device stack, rc %d\n", rc);
2964 GOTO(err_fini_proc, rc);
2967 ofd_procfs_add_brw_stats_symlink(m);
2969 /* populate cached statfs data */
2970 osfs = &ofd_info(env)->fti_u.osfs;
2971 rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
2973 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
2974 GOTO(err_fini_stack, rc);
2976 if (!IS_PO2(osfs->os_bsize)) {
2977 CERROR("%s: blocksize (%d) is not a power of 2\n",
2978 obd->obd_name, osfs->os_bsize);
2979 GOTO(err_fini_stack, rc = -EPROTO);
2981 m->ofd_blockbits = fls(osfs->os_bsize) - 1;
2983 if (ONE_MB_BRW_SIZE < (1U << m->ofd_blockbits))
2984 m->ofd_brw_size = 1U << m->ofd_blockbits;
2986 m->ofd_brw_size = ONE_MB_BRW_SIZE;
2988 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2989 if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
2990 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2992 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2993 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2994 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2995 LDLM_NAMESPACE_SERVER,
2996 LDLM_NAMESPACE_GREEDY,
2998 if (m->ofd_namespace == NULL)
2999 GOTO(err_fini_stack, rc = -ENOMEM);
3000 /* set obd_namespace for compatibility with old code */
3001 obd->obd_namespace = m->ofd_namespace;
3002 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
3003 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
3004 m->ofd_namespace->ns_lvbp = m;
3006 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3007 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
3009 dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
3011 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
3012 OBD_FAIL_OST_ALL_REQUEST_NET,
3013 OBD_FAIL_OST_ALL_REPLY_NET);
3015 GOTO(err_free_ns, rc);
3017 rc = ofd_fs_setup(env, m, obd);
3019 GOTO(err_fini_lut, rc);
3021 fid.f_seq = FID_SEQ_LOCAL_NAME;
3024 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
3027 GOTO(err_fini_fs, rc);
3029 rc = nodemap_fs_init(env, m->ofd_osd, obd, m->ofd_los);
3031 GOTO(err_fini_los, rc);
3033 rc = ofd_start_inconsistency_verification_thread(m);
3035 GOTO(err_fini_nm, rc);
3037 tgt_adapt_sptlrpc_conf(&m->ofd_lut, 1);
3042 nodemap_fs_fini(env, obd);
3044 local_oid_storage_fini(env, m->ofd_los);
3047 ofd_fs_cleanup(env, m);
3049 tgt_fini(env, &m->ofd_lut);
3051 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3052 obd->obd_namespace = m->ofd_namespace = NULL;
3054 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3061 * Stop the OFD device
3063 * This function stops the OFD device and all its subsystems.
3064 * This is the end of OFD lifecycle.
3066 * \param[in] env execution environment
3067 * \param[in] m OFD device
3069 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3071 struct obd_device *obd = ofd_obd(m);
3072 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3073 struct lfsck_stop stop;
3075 stop.ls_status = LS_PAUSED;
3077 lfsck_stop(env, m->ofd_osd, &stop);
3078 target_recovery_fini(obd);
3079 if (m->ofd_namespace != NULL)
3080 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3081 d->ld_obd->obd_force);
3083 obd_exports_barrier(obd);
3084 obd_zombie_barrier();
3086 tgt_fini(env, &m->ofd_lut);
3087 ofd_stop_inconsistency_verification_thread(m);
3088 lfsck_degister(env, m->ofd_osd);
3089 ofd_fs_cleanup(env, m);
3090 nodemap_fs_fini(env, obd);
3092 if (m->ofd_los != NULL) {
3093 local_oid_storage_fini(env, m->ofd_los);
3097 if (m->ofd_namespace != NULL) {
3098 ldlm_namespace_free_post(m->ofd_namespace);
3099 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3102 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3104 LASSERT(atomic_read(&d->ld_ref) == 0);
3105 server_put_mount(obd->obd_name, true);
3110 * Implementation of lu_device_type_operations::ldto_device_fini.
3112 * Finalize device. Dual to ofd_device_init(). It is called from
3113 * obd_precleanup() and stops the current device.
3115 * \param[in] env execution environment
3116 * \param[in] d LU device of OFD
3120 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3121 struct lu_device *d)
3124 ofd_fini(env, ofd_dev(d));
3129 * Implementation of lu_device_type_operations::ldto_device_free.
3131 * Free OFD device. Dual to ofd_device_alloc().
3133 * \param[in] env execution environment
3134 * \param[in] d LU device of OFD
3138 static struct lu_device *ofd_device_free(const struct lu_env *env,
3139 struct lu_device *d)
3141 struct ofd_device *m = ofd_dev(d);
3143 dt_device_fini(&m->ofd_dt_dev);
3149 * Implementation of lu_device_type_operations::ldto_device_alloc.
3151 * This function allocates the new OFD device. It is called from
3152 * obd_setup() if OBD device had lu_device_type defined.
3154 * \param[in] env execution environment
3155 * \param[in] t lu_device_type of OFD device
3156 * \param[in] cfg configuration log
3158 * \retval pointer to the lu_device of just allocated OFD
3159 * \retval ERR_PTR of return value on error
3161 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3162 struct lu_device_type *t,
3163 struct lustre_cfg *cfg)
3165 struct ofd_device *m;
3166 struct lu_device *l;
3171 return ERR_PTR(-ENOMEM);
3173 l = &m->ofd_dt_dev.dd_lu_dev;
3174 dt_device_init(&m->ofd_dt_dev, t);
3175 rc = ofd_init0(env, m, t, cfg);
3177 ofd_device_free(env, l);
3184 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3185 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3187 static struct lu_device_type_operations ofd_device_type_ops = {
3188 .ldto_init = ofd_type_init,
3189 .ldto_fini = ofd_type_fini,
3191 .ldto_start = ofd_type_start,
3192 .ldto_stop = ofd_type_stop,
3194 .ldto_device_alloc = ofd_device_alloc,
3195 .ldto_device_free = ofd_device_free,
3196 .ldto_device_fini = ofd_device_fini
3199 static struct lu_device_type ofd_device_type = {
3200 .ldt_tags = LU_DEVICE_DT,
3201 .ldt_name = LUSTRE_OST_NAME,
3202 .ldt_ops = &ofd_device_type_ops,
3203 .ldt_ctx_tags = LCT_DT_THREAD
3207 * Initialize OFD module.
3209 * This function is called upon module loading. It registers OFD device type
3210 * and prepares all in-memory structures used by all OFD devices.
3212 * \retval 0 if successful
3213 * \retval negative value on error
3215 static int __init ofd_init(void)
3219 rc = lu_kmem_init(ofd_caches);
3223 rc = ofd_fmd_init();
3225 lu_kmem_fini(ofd_caches);
3229 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3230 LUSTRE_OST_NAME, &ofd_device_type);
3237 * This function is called upon OFD module unloading.
3238 * It frees all related structures and unregisters OFD device type.
3240 static void __exit ofd_exit(void)
3243 lu_kmem_fini(ofd_caches);
3244 class_unregister_type(LUSTRE_OST_NAME);
3247 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3248 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3249 MODULE_VERSION(LUSTRE_VERSION_STRING);
3250 MODULE_LICENSE("GPL");
3252 module_init(ofd_init);
3253 module_exit(ofd_exit);