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, 2016, 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_param.h>
74 #include <lustre_fid.h>
75 #include <lustre_lfsck.h>
76 #include <lustre/lustre_idl.h>
77 #include <lustre_dlm.h>
78 #include <lustre_quota.h>
79 #include <lustre_nodemap.h>
80 #include <lustre_log.h>
82 #include "ofd_internal.h"
84 /* Slab for OFD object allocation */
85 static struct kmem_cache *ofd_object_kmem;
87 static struct lu_kmem_descr ofd_caches[] = {
89 .ckd_cache = &ofd_object_kmem,
90 .ckd_name = "ofd_obj",
91 .ckd_size = sizeof(struct ofd_object)
99 * Connect OFD to the next device in the stack.
101 * This function is used for device stack configuration and links OFD
102 * device with bottom OSD device.
104 * \param[in] env execution environment
105 * \param[in] m OFD device
106 * \param[in] next name of next device in the stack
107 * \param[out] exp export to return
109 * \retval 0 and export in \a exp if successful
110 * \retval negative value on error
112 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
113 const char *next, struct obd_export **exp)
115 struct obd_connect_data *data = NULL;
116 struct obd_device *obd;
122 GOTO(out, rc = -ENOMEM);
124 obd = class_name2obd(next);
126 CERROR("%s: can't locate next device: %s\n",
128 GOTO(out, rc = -ENOTCONN);
131 data->ocd_connect_flags = OBD_CONNECT_VERSION;
132 data->ocd_version = LUSTRE_VERSION_CODE;
134 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
136 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
137 ofd_name(m), next, rc);
141 m->ofd_dt_dev.dd_lu_dev.ld_site =
142 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
143 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
144 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
145 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
154 * Initialize stack of devices.
156 * This function initializes OFD-OSD device stack to serve OST requests
158 * \param[in] env execution environment
159 * \param[in] m OFD device
160 * \param[in] cfg Lustre config for this server
162 * \retval 0 if successful
163 * \retval negative value on error
165 static int ofd_stack_init(const struct lu_env *env,
166 struct ofd_device *m, struct lustre_cfg *cfg)
168 const char *dev = lustre_cfg_string(cfg, 0);
170 struct ofd_thread_info *info = ofd_info(env);
171 struct lustre_mount_info *lmi;
172 struct lustre_mount_data *lmd;
178 lmi = server_get_mount(dev);
180 CERROR("Cannot get mount info for %s!\n", dev);
184 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
185 if (lmd != NULL && lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
186 m->ofd_skip_lfsck = 1;
188 /* find bottom osd */
189 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
193 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
194 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
195 OBD_FREE(osdname, MTI_NAME_MAXLEN);
199 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
201 m->ofd_osd = lu2dt_dev(d);
203 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
204 "%s-osd", lustre_cfg_string(cfg, 0));
210 * Finalize the device stack OFD-OSD.
212 * This function cleans OFD-OSD device stack and
213 * disconnects OFD from the OSD.
215 * \param[in] env execution environment
216 * \param[in] m OFD device
217 * \param[in] top top device of stack
219 * \retval 0 if successful
220 * \retval negative value on error
222 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
223 struct lu_device *top)
225 struct obd_device *obd = ofd_obd(m);
226 struct lustre_cfg_bufs bufs;
227 struct lustre_cfg *lcfg;
232 lu_site_purge(env, top->ld_site, ~0);
233 /* process cleanup, pass mdt obd name to get obd umount flags */
234 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
239 lustre_cfg_bufs_set_string(&bufs, 1, flags);
240 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
243 lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs);
246 top->ld_ops->ldo_process_config(env, top, lcfg);
247 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
249 lu_site_purge(env, top->ld_site, ~0);
250 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
251 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
252 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
255 LASSERT(m->ofd_osd_exp);
256 obd_disconnect(m->ofd_osd_exp);
261 /* For interoperability, see mdt_interop_param[]. */
262 static struct cfg_interop_param ofd_interop_param[] = {
263 { "ost.quota_type", NULL },
268 * Check if parameters are symlinks to the OSD.
270 * Some parameters were moved from ofd to osd and only their
271 * symlinks were kept in ofd by LU-3106. They are:
272 * -writehthrough_cache_enable
273 * -readcache_max_filesize
277 * Since they are not included by the static lprocfs var list, a pre-check
278 * is added for them to avoid "unknown param" errors. If they are matched
279 * in this check, they will be passed to the OSD directly.
281 * \param[in] param parameters to check
283 * \retval true if param is symlink to OSD param
286 static bool match_symlink_param(char *param)
291 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
292 sval = strchr(param, '=');
294 paramlen = sval - param;
295 if (strncmp(param, "writethrough_cache_enable",
297 strncmp(param, "readcache_max_filesize",
299 strncmp(param, "read_cache_enable",
301 strncmp(param, "brw_stats", paramlen) == 0)
310 * Process various configuration parameters.
312 * This function is used by MGS to process specific configurations and
313 * pass them through to the next device in server stack, i.e. the OSD.
315 * \param[in] env execution environment
316 * \param[in] d LU device of OFD
317 * \param[in] cfg parameters to process
319 * \retval 0 if successful
320 * \retval negative value on error
322 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
323 struct lustre_cfg *cfg)
325 struct ofd_device *m = ofd_dev(d);
326 struct dt_device *dt_next = m->ofd_osd;
327 struct lu_device *next = &dt_next->dd_lu_dev;
332 switch (cfg->lcfg_command) {
334 struct obd_device *obd = ofd_obd(m);
335 /* For interoperability */
336 struct cfg_interop_param *ptr = NULL;
337 struct lustre_cfg *old_cfg = NULL;
340 param = lustre_cfg_string(cfg, 1);
342 CERROR("param is empty\n");
347 ptr = class_find_old_param(param, ofd_interop_param);
349 if (ptr->new_param == NULL) {
351 CWARN("For interoperability, skip this %s."
352 " It is obsolete.\n", ptr->old_param);
356 CWARN("Found old param %s, changed it to %s.\n",
357 ptr->old_param, ptr->new_param);
360 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
367 if (match_symlink_param(param)) {
368 rc = next->ld_ops->ldo_process_config(env, next, cfg);
372 rc = class_process_proc_param(PARAM_OST, obd->obd_vars, cfg,
374 if (rc > 0 || rc == -ENOSYS) {
375 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
377 /* we don't understand; pass it on */
378 rc = next->ld_ops->ldo_process_config(env, next, cfg);
382 case LCFG_SPTLRPC_CONF: {
387 /* others are passed further */
388 rc = next->ld_ops->ldo_process_config(env, next, cfg);
395 * Implementation of lu_object_operations::loo_object_init for OFD
397 * Allocate just the next object (OSD) in stack.
399 * \param[in] env execution environment
400 * \param[in] o lu_object of OFD object
401 * \param[in] conf additional configuration parameters, not used here
403 * \retval 0 if successful
404 * \retval negative value on error
406 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
407 const struct lu_object_conf *conf)
409 struct ofd_device *d = ofd_dev(o->lo_dev);
410 struct lu_device *under;
411 struct lu_object *below;
416 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
417 PFID(lu_object_fid(o)));
419 under = &d->ofd_osd->dd_lu_dev;
420 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
422 lu_object_add(o, below);
430 * Implementation of lu_object_operations::loo_object_free.
432 * Finish OFD object lifecycle and free its memory.
434 * \param[in] env execution environment
435 * \param[in] o LU object of OFD object
437 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
439 struct ofd_object *of = ofd_obj(o);
440 struct lu_object_header *h;
445 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
446 PFID(lu_object_fid(o)));
449 lu_object_header_fini(h);
450 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
455 * Implementation of lu_object_operations::loo_object_print.
457 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
458 * LU_OBJECT_DEBUG() for more details about the compound object printing.
460 * \param[in] env execution environment
461 * \param[in] cookie opaque data passed to the printer function
462 * \param[in] p printer function to use
463 * \param[in] o LU object of OFD object
465 * \retval 0 if successful
466 * \retval negative value on error
468 static int ofd_object_print(const struct lu_env *env, void *cookie,
469 lu_printer_t p, const struct lu_object *o)
471 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
474 static struct lu_object_operations ofd_obj_ops = {
475 .loo_object_init = ofd_object_init,
476 .loo_object_free = ofd_object_free,
477 .loo_object_print = ofd_object_print
481 * Implementation of lu_device_operations::lod_object_alloc.
483 * This function allocates OFD part of compound OFD-OSD object and
484 * initializes its header, because OFD is the top device in stack
486 * \param[in] env execution environment
487 * \param[in] hdr object header, NULL for OFD
488 * \param[in] d lu_device
490 * \retval allocated object if successful
491 * \retval NULL value on failed allocation
493 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
494 const struct lu_object_header *hdr,
497 struct ofd_object *of;
501 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
504 struct lu_object_header *h;
506 o = &of->ofo_obj.do_lu;
508 lu_object_header_init(h);
509 lu_object_init(o, h, d);
510 lu_object_add_top(h, o);
511 o->lo_ops = &ofd_obj_ops;
519 * Return the result of LFSCK run to the OFD.
521 * Notify OFD about result of LFSCK run. That may block the new object
522 * creation until problem is fixed by LFSCK.
524 * \param[in] env execution environment
525 * \param[in] data pointer to the OFD device
526 * \param[in] event LFSCK event type
528 * \retval 0 if successful
529 * \retval negative value on unknown event
531 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
532 enum lfsck_events event)
534 struct ofd_device *ofd = data;
535 struct obd_device *obd = ofd_obd(ofd);
538 case LE_LASTID_REBUILDING:
539 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
540 "on the device until the LAST_ID rebuilt successfully.\n",
542 down_write(&ofd->ofd_lastid_rwsem);
543 ofd->ofd_lastid_rebuilding = 1;
544 up_write(&ofd->ofd_lastid_rwsem);
546 case LE_LASTID_REBUILT: {
547 down_write(&ofd->ofd_lastid_rwsem);
548 ofd_seqs_free(env, ofd);
549 ofd->ofd_lastid_rebuilding = 0;
550 ofd->ofd_lastid_gen++;
551 up_write(&ofd->ofd_lastid_rwsem);
552 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
557 CERROR("%s: unknown lfsck event: rc = %d\n",
558 ofd_name(ofd), event);
566 * Implementation of lu_device_operations::ldo_prepare.
568 * This method is called after layer has been initialized and before it starts
569 * serving user requests. In OFD it starts lfsk check routines and initializes
572 * \param[in] env execution environment
573 * \param[in] pdev higher device in stack, NULL for OFD
574 * \param[in] dev lu_device of OFD device
576 * \retval 0 if successful
577 * \retval negative value on error
579 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
580 struct lu_device *dev)
582 struct ofd_thread_info *info;
583 struct ofd_device *ofd = ofd_dev(dev);
584 struct obd_device *obd = ofd_obd(ofd);
585 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
590 info = ofd_info_init(env, NULL);
594 /* initialize lower device */
595 rc = next->ld_ops->ldo_prepare(env, dev, next);
599 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
600 ofd_lfsck_out_notify, ofd, false);
602 CERROR("%s: failed to initialize lfsck: rc = %d\n",
607 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
608 /* The LFSCK instance is registered just now, so it must be there when
609 * register the namespace to such instance. */
610 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
612 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
613 LASSERT(obd->obd_no_conn);
614 spin_lock(&obd->obd_dev_lock);
615 obd->obd_no_conn = 0;
616 spin_unlock(&obd->obd_dev_lock);
618 if (obd->obd_recovering == 0)
619 ofd_postrecov(env, ofd);
625 * Implementation of lu_device_operations::ldo_recovery_complete.
627 * This method notifies all layers about 'recovery complete' event. That means
628 * device is in full state and consistent. An OFD calculates available grant
629 * space upon this event.
631 * \param[in] env execution environment
632 * \param[in] dev lu_device of OFD device
634 * \retval 0 if successful
635 * \retval negative value on error
637 static int ofd_recovery_complete(const struct lu_env *env,
638 struct lu_device *dev)
640 struct ofd_thread_info *oti = ofd_info(env);
641 struct ofd_device *ofd = ofd_dev(dev);
642 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
648 * Grant space for object precreation on the self export.
649 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
650 * is enough to create 10k objects. More space is then acquired for
651 * precreation in tgt_grant_create().
653 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
654 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
655 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
656 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
657 OBD_CONNECT_GRANT_PARAM;
658 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
660 rc = next->ld_ops->ldo_recovery_complete(env, next);
665 * lu_device_operations matrix for OFD device.
667 static struct lu_device_operations ofd_lu_ops = {
668 .ldo_object_alloc = ofd_object_alloc,
669 .ldo_process_config = ofd_process_config,
670 .ldo_recovery_complete = ofd_recovery_complete,
671 .ldo_prepare = ofd_prepare,
674 LPROC_SEQ_FOPS(lprocfs_nid_stats_clear);
677 * Initialize all needed procfs entries for OFD device.
679 * \param[in] ofd OFD device
681 * \retval 0 if successful
682 * \retval negative value on error
684 static int ofd_procfs_init(struct ofd_device *ofd)
686 struct obd_device *obd = ofd_obd(ofd);
687 struct proc_dir_entry *entry;
692 /* lprocfs must be setup before the ofd so state can be safely added
693 * to /proc incrementally as the ofd is setup */
694 obd->obd_vars = lprocfs_ofd_obd_vars;
695 rc = lprocfs_obd_setup(obd, false);
697 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
702 rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
704 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
706 GOTO(obd_cleanup, rc);
709 obd->obd_uses_nid_stats = 1;
711 entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
714 CERROR("%s: error %d setting up lprocfs for %s\n",
715 obd->obd_name, rc, "exports");
716 GOTO(obd_cleanup, rc);
718 obd->obd_proc_exports_entry = entry;
720 entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
721 obd, &lprocfs_nid_stats_clear_fops);
724 CERROR("%s: add proc entry 'clear' failed: %d.\n",
726 GOTO(obd_cleanup, rc);
729 ofd_stats_counter_init(obd->obd_stats);
731 rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
732 ofd_stats_counter_init);
734 GOTO(obd_cleanup, rc);
737 lprocfs_obd_cleanup(obd);
738 lprocfs_free_obd_stats(obd);
744 * Expose OSD statistics to OFD layer.
746 * The osd interfaces to the backend file system exposes useful data
747 * such as brw_stats and read or write cache states. This same data
748 * needs to be exposed into the obdfilter (ofd) layer to maintain
749 * backwards compatibility. This function creates the symlinks in the
750 * proc layer to enable this.
752 * \param[in] ofd OFD device
754 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
756 struct obd_device *obd = ofd_obd(ofd);
757 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
759 if (obd->obd_proc_entry == NULL)
762 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
763 "../../%s/%s/brw_stats",
764 osd_obd->obd_type->typ_name, obd->obd_name);
766 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
767 "../../%s/%s/read_cache_enable",
768 osd_obd->obd_type->typ_name, obd->obd_name);
770 lprocfs_add_symlink("readcache_max_filesize",
772 "../../%s/%s/readcache_max_filesize",
773 osd_obd->obd_type->typ_name, obd->obd_name);
775 lprocfs_add_symlink("writethrough_cache_enable",
777 "../../%s/%s/writethrough_cache_enable",
778 osd_obd->obd_type->typ_name, obd->obd_name);
782 * Cleanup all procfs entries in OFD.
784 * \param[in] ofd OFD device
786 static void ofd_procfs_fini(struct ofd_device *ofd)
788 struct obd_device *obd = ofd_obd(ofd);
790 lprocfs_free_per_client_stats(obd);
791 lprocfs_obd_cleanup(obd);
792 lprocfs_free_obd_stats(obd);
793 lprocfs_job_stats_fini(obd);
797 * Stop SEQ/FID server on OFD.
799 * \param[in] env execution environment
800 * \param[in] ofd OFD device
802 * \retval 0 if successful
803 * \retval negative value on error
805 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
807 return seq_site_fini(env, &ofd->ofd_seq_site);
811 * Start SEQ/FID server on OFD.
813 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
814 * It also connects to the master server to get own FID sequence (SEQ) range
815 * to this particular OFD. Typically that happens when the OST is first
816 * formatted or in the rare case that it exhausts the local sequence range.
818 * The sequence range is allocated out to the MDTs for OST object allocations,
819 * and not directly to the clients.
821 * \param[in] env execution environment
822 * \param[in] ofd OFD device
824 * \retval 0 if successful
825 * \retval negative value on error
827 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
829 struct seq_server_site *ss = &ofd->ofd_seq_site;
830 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
831 char *obd_name = ofd_name(ofd);
835 ss = &ofd->ofd_seq_site;
836 lu->ld_site->ld_seq_site = ss;
837 ss->ss_lu = lu->ld_site;
838 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
840 OBD_ALLOC(name, sizeof(obd_name) * 2 + 10);
844 OBD_ALLOC_PTR(ss->ss_server_seq);
845 if (ss->ss_server_seq == NULL)
846 GOTO(out_name, rc = -ENOMEM);
848 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
849 LUSTRE_SEQ_SERVER, ss);
851 CERROR("%s : seq server init error %d\n", obd_name, rc);
852 GOTO(out_server, rc);
854 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
856 OBD_ALLOC_PTR(ss->ss_client_seq);
857 if (ss->ss_client_seq == NULL)
858 GOTO(out_server, rc = -ENOMEM);
861 * It always printed as "%p", so that the name is unique in the kernel,
862 * even if the filesystem is mounted twice. So sizeof(.) * 2 is enough.
864 snprintf(name, sizeof(obd_name) * 2 + 7, "%p-super", obd_name);
865 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
868 CERROR("%s : seq client init error %d\n", obd_name, rc);
869 GOTO(out_client, rc);
872 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
876 seq_client_fini(ss->ss_client_seq);
877 OBD_FREE_PTR(ss->ss_client_seq);
878 ss->ss_client_seq = NULL;
880 seq_server_fini(ss->ss_server_seq, env);
881 OBD_FREE_PTR(ss->ss_server_seq);
882 ss->ss_server_seq = NULL;
885 OBD_FREE(name, sizeof(obd_name) * 2 + 10);
891 * OFD request handler for OST_SET_INFO RPC.
893 * This is OFD-specific part of request handling
895 * \param[in] tsi target session environment for this request
897 * \retval 0 if successful
898 * \retval negative value on error
900 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
902 struct ptlrpc_request *req = tgt_ses_req(tsi);
903 struct ost_body *body = NULL, *repbody;
904 void *key, *val = NULL;
905 int keylen, vallen, rc = 0;
906 bool is_grant_shrink;
910 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
912 DEBUG_REQ(D_HA, req, "no set_info key");
913 RETURN(err_serious(-EFAULT));
915 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
918 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
920 DEBUG_REQ(D_HA, req, "no set_info val");
921 RETURN(err_serious(-EFAULT));
923 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
926 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
928 /* In this case the value is actually an RMF_OST_BODY, so we
929 * transmutate the type of this PTLRPC */
930 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
932 rc = req_capsule_server_pack(tsi->tsi_pill);
936 if (is_grant_shrink) {
937 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
939 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
942 /** handle grant shrink, similar to a read request */
943 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
945 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
947 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
950 CERROR("%s: Unsupported key %s\n",
951 tgt_name(tsi->tsi_tgt), (char *)key);
954 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
961 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
963 * This function returns a list of extents which describes how a file's
964 * blocks are laid out on the disk.
966 * \param[in] env execution environment
967 * \param[in] ofd OFD device
968 * \param[in] fid FID of object
969 * \param[in] fiemap fiemap structure to fill with data
971 * \retval 0 if \a fiemap is filled with data successfully
972 * \retval negative value on error
974 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
975 struct lu_fid *fid, struct fiemap *fiemap)
977 struct ofd_object *fo;
980 fo = ofd_object_find(env, ofd, fid);
982 CERROR("%s: error finding object "DFID"\n",
983 ofd_name(ofd), PFID(fid));
987 ofd_read_lock(env, fo);
988 if (ofd_object_exists(fo))
989 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
992 ofd_read_unlock(env, fo);
993 ofd_object_put(env, fo);
997 struct locked_region {
998 struct list_head list;
999 struct lustre_handle lh;
1003 * Lock single extent and save lock handle in the list.
1005 * This is supplemental function for lock_zero_regions(). It allocates
1006 * new locked_region structure and locks it with extent lock, then adds
1007 * it to the list of all such regions.
1009 * \param[in] ns LDLM namespace
1010 * \param[in] res_id resource ID
1011 * \param[in] begin start of region
1012 * \param[in] end end of region
1013 * \param[in] locked list head of regions list
1015 * \retval 0 if successful locking
1016 * \retval negative value on error
1018 static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1019 unsigned long long begin, unsigned long long end,
1020 struct list_head *locked)
1022 struct locked_region *region = NULL;
1026 LASSERT(begin <= end);
1027 OBD_ALLOC_PTR(region);
1031 rc = tgt_extent_lock(ns, res_id, begin, end, ®ion->lh,
1036 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
1038 list_add(®ion->list, locked);
1044 * Lock the sparse areas of given resource.
1046 * The locking of sparse areas will cause dirty data to be flushed back from
1047 * clients. This is used when getting the FIEMAP of an object to make sure
1048 * there is no unaccounted cached data on clients.
1050 * This function goes through \a fiemap list of extents and locks only sparse
1051 * areas between extents.
1053 * \param[in] ns LDLM namespace
1054 * \param[in] res_id resource ID
1055 * \param[in] fiemap file extents mapping on disk
1056 * \param[in] locked list head of regions list
1058 * \retval 0 if successful
1059 * \retval negative value on error
1061 static int lock_zero_regions(struct ldlm_namespace *ns,
1062 struct ldlm_res_id *res_id,
1063 struct fiemap *fiemap,
1064 struct list_head *locked)
1066 __u64 begin = fiemap->fm_start;
1069 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1073 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1074 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1075 if (fiemap_start[i].fe_logical > begin) {
1076 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1077 begin, fiemap_start[i].fe_logical);
1078 rc = lock_region(ns, res_id, begin,
1079 fiemap_start[i].fe_logical, locked);
1084 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1087 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1088 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1089 begin, fiemap->fm_start + fiemap->fm_length);
1090 rc = lock_region(ns, res_id, begin,
1091 fiemap->fm_start + fiemap->fm_length, locked);
1098 * Unlock all previously locked sparse areas for given resource.
1100 * This function goes through list of locked regions, unlocking and freeing
1103 * \param[in] ns LDLM namespace
1104 * \param[in] locked list head of regions list
1107 unlock_zero_regions(struct ldlm_namespace *ns, struct list_head *locked)
1109 struct locked_region *entry, *temp;
1111 list_for_each_entry_safe(entry, temp, locked, list) {
1112 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
1113 tgt_extent_unlock(&entry->lh, LCK_PR);
1114 list_del(&entry->list);
1115 OBD_FREE_PTR(entry);
1120 * OFD request handler for OST_GET_INFO RPC.
1122 * This is OFD-specific part of request handling. The OFD-specific keys are:
1123 * - KEY_LAST_ID (obsolete)
1127 * This function reads needed data from storage and fills reply with it.
1129 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1130 * and is kept for compatibility.
1132 * \param[in] tsi target session environment for this request
1134 * \retval 0 if successful
1135 * \retval negative value on error
1137 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1139 struct obd_export *exp = tsi->tsi_exp;
1140 struct ofd_device *ofd = ofd_exp(exp);
1141 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1144 int replylen, rc = 0;
1148 /* this common part for get_info rpc */
1149 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1151 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1152 RETURN(err_serious(-EPROTO));
1154 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1157 if (KEY_IS(KEY_LAST_ID)) {
1159 struct ofd_seq *oseq;
1161 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1162 rc = req_capsule_server_pack(tsi->tsi_pill);
1164 RETURN(err_serious(rc));
1166 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1168 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1169 (u64)exp->exp_filter_data.fed_group);
1173 *last_id = ofd_seq_last_oid(oseq);
1174 ofd_seq_put(tsi->tsi_env, oseq);
1175 } else if (KEY_IS(KEY_FIEMAP)) {
1176 struct ll_fiemap_info_key *fm_key;
1177 struct fiemap *fiemap;
1180 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1182 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1183 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1185 RETURN(err_serious(rc));
1187 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1189 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1191 replylen = fiemap_count_to_size(
1192 fm_key->lfik_fiemap.fm_extent_count);
1193 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1194 RCL_SERVER, replylen);
1196 rc = req_capsule_server_pack(tsi->tsi_pill);
1198 RETURN(err_serious(rc));
1200 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1204 *fiemap = fm_key->lfik_fiemap;
1205 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1207 /* LU-3219: Lock the sparse areas to make sure dirty
1208 * flushed back from client, then call fiemap again. */
1209 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1210 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1211 struct list_head locked;
1213 INIT_LIST_HEAD(&locked);
1214 ost_fid_build_resid(fid, &fti->fti_resid);
1215 rc = lock_zero_regions(ofd->ofd_namespace,
1216 &fti->fti_resid, fiemap,
1218 if (rc == 0 && !list_empty(&locked)) {
1219 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1221 unlock_zero_regions(ofd->ofd_namespace,
1225 } else if (KEY_IS(KEY_LAST_FID)) {
1226 struct ofd_device *ofd = ofd_exp(exp);
1227 struct ofd_seq *oseq;
1231 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1232 rc = req_capsule_server_pack(tsi->tsi_pill);
1234 RETURN(err_serious(rc));
1236 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1238 RETURN(err_serious(-EPROTO));
1240 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1242 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1246 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1247 ostid_seq(&fti->fti_ostid));
1249 RETURN(PTR_ERR(oseq));
1251 rc = ostid_to_fid(fid, &oseq->os_oi,
1252 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1256 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1259 ofd_seq_put(tsi->tsi_env, oseq);
1261 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1265 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1272 * OFD request handler for OST_GETATTR RPC.
1274 * This is OFD-specific part of request handling. It finds the OFD object
1275 * by its FID, gets attributes from storage and packs result to the reply.
1277 * \param[in] tsi target session environment for this request
1279 * \retval 0 if successful
1280 * \retval negative value on error
1282 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1284 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1285 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1286 struct ost_body *repbody;
1287 struct lustre_handle lh = { 0 };
1288 struct ofd_object *fo;
1290 enum ldlm_mode lock_mode = LCK_PR;
1295 LASSERT(tsi->tsi_ost_body != NULL);
1297 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1298 if (repbody == NULL)
1301 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1302 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1304 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1305 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1308 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1311 rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace,
1312 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1318 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1320 GOTO(out, rc = PTR_ERR(fo));
1322 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1326 obdo_from_la(&repbody->oa, &fti->fti_attr,
1327 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1329 /* Store object version in reply */
1330 curr_version = dt_version_get(tsi->tsi_env,
1331 ofd_object_child(fo));
1332 if ((__s64)curr_version != -EOPNOTSUPP) {
1333 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1334 repbody->oa.o_data_version = curr_version;
1338 ofd_object_put(tsi->tsi_env, fo);
1341 tgt_extent_unlock(&lh, lock_mode);
1343 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1346 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1347 repbody->oa.o_flags = OBD_FL_FLUSH;
1353 * OFD request handler for OST_SETATTR RPC.
1355 * This is OFD-specific part of request handling. It finds the OFD object
1356 * by its FID, sets attributes from request and packs result to the reply.
1358 * \param[in] tsi target session environment for this request
1360 * \retval 0 if successful
1361 * \retval negative value on error
1363 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1365 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1366 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1367 struct ost_body *body = tsi->tsi_ost_body;
1368 struct ost_body *repbody;
1369 struct ldlm_resource *res;
1370 struct ofd_object *fo;
1371 struct filter_fid *ff = NULL;
1376 LASSERT(body != NULL);
1378 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1379 if (repbody == NULL)
1382 repbody->oa.o_oi = body->oa.o_oi;
1383 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1385 /* This would be very bad - accidentally truncating a file when
1386 * changing the time or similar - bug 12203. */
1387 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1388 body->oa.o_size != OBD_OBJECT_EOF) {
1389 static char mdsinum[48];
1391 if (body->oa.o_valid & OBD_MD_FLFID)
1392 snprintf(mdsinum, sizeof(mdsinum) - 1,
1393 "of parent "DFID, body->oa.o_parent_seq,
1394 body->oa.o_parent_oid, 0);
1398 CERROR("%s: setattr from %s is trying to truncate object "DFID
1399 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1400 PFID(&tsi->tsi_fid), mdsinum);
1404 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1406 GOTO(out, rc = PTR_ERR(fo));
1408 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1409 fti->fti_attr.la_valid &= ~LA_TYPE;
1411 if (body->oa.o_valid & OBD_MD_FLFID) {
1412 ff = &fti->fti_mds_fid;
1413 ofd_prepare_fidea(ff, &body->oa);
1416 /* setting objects attributes (including owner/group) */
1417 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff);
1421 obdo_from_la(&repbody->oa, &fti->fti_attr,
1422 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1424 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1428 ofd_object_put(tsi->tsi_env, fo);
1431 /* we do not call this before to avoid lu_object_find() in
1432 * ->lvbo_update() holding another reference on the object.
1433 * otherwise concurrent destroy can make the object unavailable
1434 * for 2nd lu_object_find() waiting for the first reference
1435 * to go... deadlock! */
1436 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1437 &tsi->tsi_resid, LDLM_EXTENT, 0);
1439 ldlm_res_lvbo_update(res, NULL, 0);
1440 ldlm_resource_putref(res);
1447 * Destroy OST orphans.
1449 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1450 * set then we must destroy possible orphaned objects.
1452 * \param[in] env execution environment
1453 * \param[in] exp OBD export
1454 * \param[in] ofd OFD device
1455 * \param[in] oa obdo structure for reply
1457 * \retval 0 if successful
1458 * \retval negative value on error
1460 static int ofd_orphans_destroy(const struct lu_env *env,
1461 struct obd_export *exp,
1462 struct ofd_device *ofd, struct obdo *oa)
1464 struct ofd_thread_info *info = ofd_info(env);
1465 struct lu_fid *fid = &info->fti_fid;
1466 struct ost_id *oi = &oa->o_oi;
1467 struct ofd_seq *oseq;
1468 u64 seq = ostid_seq(oi);
1469 u64 end_id = ostid_id(oi);
1477 oseq = ofd_seq_get(ofd, seq);
1479 CERROR("%s: Can not find seq for "DOSTID"\n",
1480 ofd_name(ofd), POSTID(oi));
1485 last = ofd_seq_last_oid(oseq);
1488 LASSERT(exp != NULL);
1489 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1491 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1494 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1495 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1497 while (oid > end_id) {
1498 rc = fid_set_id(fid, oid);
1499 if (unlikely(rc != 0))
1502 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1503 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1504 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1505 /* this is pretty fatal... */
1506 CEMERG("%s: error destroying precreated id "
1508 ofd_name(ofd), PFID(fid), rc);
1512 ofd_seq_last_oid_set(oseq, oid);
1513 /* update last_id on disk periodically so that if we
1514 * restart * we don't need to re-scan all of the just
1515 * deleted objects. */
1516 if ((oid & 511) == 0)
1517 ofd_seq_last_oid_write(env, ofd, oseq);
1521 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1522 ofd_name(ofd), seq, oid);
1526 ofd_seq_last_oid_set(oseq, oid);
1527 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1529 /* don't reuse orphan object, return last used objid */
1530 rc = 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;
1569 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1572 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1573 if (repbody == NULL)
1576 down_read(&ofd->ofd_lastid_rwsem);
1577 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1578 * we may do that in the future.
1579 * Return -ENOSPC until the LAST_ID rebuilt. */
1580 if (unlikely(ofd->ofd_lastid_rebuilding))
1581 GOTO(out_sem, rc = -ENOSPC);
1583 rep_oa = &repbody->oa;
1584 rep_oa->o_oi = oa->o_oi;
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 %#llx: 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 %llu"
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:[%llu] destroys_in_progress already"
1627 " cleared\n", ofd_name(ofd), seq);
1628 rc = ostid_set_id(&rep_oa->o_oi,
1629 ofd_seq_last_oid(oseq));
1632 diff = oid - ofd_seq_last_oid(oseq);
1633 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1634 ofd_seq_last_oid(oseq), diff);
1635 if (-diff > OST_MAX_PRECREATE) {
1636 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1637 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1638 "(%llu), trust the OST\n",
1639 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1640 PFID(&oseq->os_oi.oi_fid),
1641 ofd_seq_last_oid(oseq));
1643 /* Let MDS know that we are so far ahead. */
1644 rc = ostid_set_id(&rep_oa->o_oi,
1645 ofd_seq_last_oid(oseq) + 1);
1646 } else if (diff < 0) {
1647 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1649 oseq->os_destroys_in_progress = 0;
1651 /* XXX: Used by MDS for the first time! */
1652 oseq->os_destroys_in_progress = 0;
1655 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1656 ofd->ofd_lastid_gen)) {
1657 /* Keep the export ref so we can send the reply. */
1658 ofd_obd_disconnect(class_export_get(exp));
1659 GOTO(out_nolock, rc = -ENOTCONN);
1662 mutex_lock(&oseq->os_create_lock);
1663 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1664 exp->exp_conn_cnt) {
1665 CERROR("%s: dropping old precreate request\n",
1669 /* only precreate if seq is 0, IDIF or normal and also o_id
1670 * must be specfied */
1671 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1672 !fid_seq_is_idif(seq)) || oid == 0) {
1673 diff = 1; /* shouldn't we create this right now? */
1675 diff = oid - ofd_seq_last_oid(oseq);
1676 /* Do sync create if the seq is about to used up */
1677 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1678 if (unlikely(oid >= IDIF_MAX_OID - 1))
1680 } else if (fid_seq_is_norm(seq)) {
1682 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1685 CERROR("%s : invalid o_seq "DOSTID"\n",
1686 ofd_name(ofd), POSTID(&oa->o_oi));
1687 GOTO(out, rc = -EINVAL);
1692 CERROR("%s: invalid precreate request for "
1693 DOSTID", last_id %llu. "
1694 "Likely MDS last_id corruption\n",
1695 ofd_name(ofd), POSTID(&oa->o_oi),
1696 ofd_seq_last_oid(oseq));
1697 GOTO(out, rc = -EINVAL);
1702 cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
1707 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1708 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1709 /* don't enforce grant during orphan recovery */
1710 granted = tgt_grant_create(tsi->tsi_env,
1711 ofd_obd(ofd)->obd_self_export,
1716 CDEBUG(D_HA, "%s: failed to acquire grant "
1717 "space for precreate (%lld): rc = %d\n",
1718 ofd_name(ofd), diff, rc);
1723 /* This can happen if a new OST is formatted and installed
1724 * in place of an old one at the same index. Instead of
1725 * precreating potentially millions of deleted old objects
1726 * (possibly filling the OST), only precreate the last batch.
1727 * LFSCK will eventually clean up any orphans. LU-14 */
1728 if (diff > 5 * OST_MAX_PRECREATE) {
1729 diff = OST_MAX_PRECREATE / 2;
1730 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1731 "OST replaced or reformatted: "
1732 "LFSCK will clean up",
1735 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1736 "%u larger than the LAST_ID "DOSTID", only "
1737 "precreating the last %lld objects.\n",
1738 ofd_name(ofd), POSTID(&oa->o_oi),
1739 5 * OST_MAX_PRECREATE,
1740 POSTID(&oseq->os_oi), diff);
1741 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1745 next_id = ofd_seq_last_oid(oseq) + 1;
1746 count = ofd_precreate_batch(ofd, (int)diff);
1748 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1749 " at %llu\n", ofd_name(ofd),
1750 count, seq, next_id);
1752 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1753 && cfs_time_after(jiffies, enough_time)) {
1754 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1755 " created at a rate of %d/s\n",
1756 ofd_name(ofd), created, diff + created,
1757 created / DISK_TIMEOUT);
1761 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1762 oseq, count, sync_trans);
1766 } else if (rc < 0) {
1772 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1773 LCONSOLE_WARN("%s: can't create the same count of"
1774 " objects when replaying the request"
1775 " (diff is %lld). see LU-4621\n",
1776 ofd_name(ofd), diff);
1779 /* some objects got created, we can return
1780 * them, even if last creation failed */
1783 CERROR("%s: unable to precreate: rc = %d\n",
1786 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1787 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1788 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1793 rc = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1796 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1799 mutex_unlock(&oseq->os_create_lock);
1802 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1803 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1804 struct lu_fid *fid = &info->fti_fid;
1806 /* For compatible purpose, it needs to convert back to
1807 * OST ID before put it on wire. */
1808 *fid = rep_oa->o_oi.oi_fid;
1809 fid_to_ostid(fid, &rep_oa->o_oi);
1811 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1813 ofd_seq_put(tsi->tsi_env, oseq);
1816 up_read(&ofd->ofd_lastid_rwsem);
1821 * OFD request handler for OST_DESTROY RPC.
1823 * This is OFD-specific part of request handling. It destroys data objects
1824 * related to destroyed object on MDT.
1826 * \param[in] tsi target session environment for this request
1828 * \retval 0 if successful
1829 * \retval negative value on error
1831 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1833 const struct ost_body *body = tsi->tsi_ost_body;
1834 struct ost_body *repbody;
1835 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1836 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1837 struct lu_fid *fid = &fti->fti_fid;
1844 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1847 /* This is old case for clients before Lustre 2.4 */
1848 /* If there's a DLM request, cancel the locks mentioned in it */
1849 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1851 struct ldlm_request *dlm;
1853 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1856 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1859 *fid = body->oa.o_oi.oi_fid;
1860 oid = ostid_id(&body->oa.o_oi);
1863 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1865 /* check that o_misc makes sense */
1866 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1867 count = body->oa.o_misc;
1869 count = 1; /* default case - single destroy */
1871 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1872 POSTID(&body->oa.o_oi), count);
1877 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1878 if (lrc == -ENOENT) {
1880 "%s: destroying non-existent object "DFID"\n",
1881 ofd_name(ofd), PFID(fid));
1882 /* rewrite rc with -ENOENT only if it is 0 */
1885 } else if (lrc != 0) {
1886 CERROR("%s: error destroying object "DFID": %d\n",
1887 ofd_name(ofd), PFID(fid), lrc);
1893 lrc = fid_set_id(fid, oid);
1894 if (unlikely(lrc != 0 && count > 0))
1895 GOTO(out, rc = lrc);
1898 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1904 fid_to_ostid(fid, &repbody->oa.o_oi);
1909 * OFD request handler for OST_STATFS RPC.
1911 * This function gets statfs data from storage as part of request
1914 * \param[in] tsi target session environment for this request
1916 * \retval 0 if successful
1917 * \retval negative value on error
1919 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1921 struct obd_statfs *osfs;
1926 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1928 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1929 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
1931 CERROR("%s: statfs failed: rc = %d\n",
1932 tgt_name(tsi->tsi_tgt), rc);
1934 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1937 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1944 * OFD request handler for OST_SYNC RPC.
1946 * Sync object data or all filesystem data to the disk and pack the
1949 * \param[in] tsi target session environment for this request
1951 * \retval 0 if successful
1952 * \retval negative value on error
1954 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1956 struct ost_body *body = tsi->tsi_ost_body;
1957 struct ost_body *repbody;
1958 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1959 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1960 struct ofd_object *fo = NULL;
1965 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1967 /* if no objid is specified, it means "sync whole filesystem" */
1968 if (!fid_is_zero(&tsi->tsi_fid)) {
1969 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1971 RETURN(PTR_ERR(fo));
1974 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1975 fo != NULL ? ofd_object_child(fo) : NULL,
1976 repbody->oa.o_size, repbody->oa.o_blocks);
1980 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1985 repbody->oa.o_oi = body->oa.o_oi;
1986 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1988 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1990 obdo_from_la(&repbody->oa, &fti->fti_attr,
1993 /* don't return rc from getattr */
1998 ofd_object_put(tsi->tsi_env, fo);
2003 * OFD request handler for OST_PUNCH RPC.
2005 * This is part of request processing. Validate request fields,
2006 * punch (truncate) the given OFD object and pack reply.
2008 * \param[in] tsi target session environment for this request
2010 * \retval 0 if successful
2011 * \retval negative value on error
2013 static int ofd_punch_hdl(struct tgt_session_info *tsi)
2015 const struct obdo *oa = &tsi->tsi_ost_body->oa;
2016 struct ost_body *repbody;
2017 struct ofd_thread_info *info = tsi2ofd_info(tsi);
2018 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
2019 struct ldlm_resource *res;
2020 struct ofd_object *fo;
2021 struct filter_fid *ff = NULL;
2023 struct lustre_handle lh = { 0, };
2030 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
2032 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
2033 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
2035 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2036 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2037 RETURN(err_serious(-EPROTO));
2039 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2040 if (repbody == NULL)
2041 RETURN(err_serious(-ENOMEM));
2043 /* punch start,end are passed in o_size,o_blocks throught wire */
2047 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2050 /* standard truncate optimization: if file body is completely
2051 * destroyed, don't send data back to the server. */
2053 flags |= LDLM_FL_AST_DISCARD_DATA;
2055 repbody->oa.o_oi = oa->o_oi;
2056 repbody->oa.o_valid = OBD_MD_FLID;
2058 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2059 oa->o_flags & OBD_FL_SRVLOCK;
2062 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
2068 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2069 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2070 oa->o_valid, start, end);
2072 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2075 GOTO(out, rc = PTR_ERR(fo));
2077 la_from_obdo(&info->fti_attr, oa,
2078 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2079 info->fti_attr.la_size = start;
2080 info->fti_attr.la_valid |= LA_SIZE;
2082 if (oa->o_valid & OBD_MD_FLFID) {
2083 ff = &info->fti_mds_fid;
2084 ofd_prepare_fidea(ff, oa);
2087 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2088 ff, (struct obdo *)oa);
2092 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2096 ofd_object_put(tsi->tsi_env, fo);
2099 tgt_extent_unlock(&lh, LCK_PW);
2101 /* we do not call this before to avoid lu_object_find() in
2102 * ->lvbo_update() holding another reference on the object.
2103 * otherwise concurrent destroy can make the object unavailable
2104 * for 2nd lu_object_find() waiting for the first reference
2105 * to go... deadlock! */
2106 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2109 ldlm_res_lvbo_update(res, NULL, 0);
2110 ldlm_resource_putref(res);
2116 static int ofd_ladvise_prefetch(const struct lu_env *env,
2117 struct ofd_object *fo,
2118 struct niobuf_local *lnb,
2119 __u64 start, __u64 end)
2121 struct ofd_thread_info *info = ofd_info(env);
2122 pgoff_t start_index, end_index, pages;
2123 struct niobuf_remote rnb;
2124 unsigned long nr_local;
2130 ofd_read_lock(env, fo);
2131 if (!ofd_object_exists(fo))
2132 GOTO(out_unlock, rc = -ENOENT);
2134 rc = ofd_attr_get(env, fo, &info->fti_attr);
2136 GOTO(out_unlock, rc);
2138 if (end > info->fti_attr.la_size)
2139 end = info->fti_attr.la_size;
2142 GOTO(out_unlock, rc);
2144 /* We need page aligned offset and length */
2145 start_index = start >> PAGE_SHIFT;
2146 end_index = (end - 1) >> PAGE_SHIFT;
2147 pages = end_index - start_index + 1;
2149 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2150 PTLRPC_MAX_BRW_PAGES;
2151 rnb.rnb_offset = start_index << PAGE_SHIFT;
2152 rnb.rnb_len = nr_local << PAGE_SHIFT;
2153 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, 0);
2154 if (unlikely(rc < 0))
2157 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2158 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2161 start_index += nr_local;
2166 ofd_read_unlock(env, fo);
2171 * OFD request handler for OST_LADVISE RPC.
2173 * Tune cache or perfetch policies according to advices.
2175 * \param[in] tsi target session environment for this request
2177 * \retval 0 if successful
2178 * \retval negative errno on error
2180 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2182 struct ptlrpc_request *req = tgt_ses_req(tsi);
2183 struct obd_export *exp = tsi->tsi_exp;
2184 struct ofd_device *ofd = ofd_exp(exp);
2185 struct ost_body *body, *repbody;
2186 struct ofd_thread_info *info;
2187 struct ofd_object *fo;
2188 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2189 const struct lu_env *env = svc_thread->t_env;
2190 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2192 struct lu_ladvise *ladvise;
2194 struct ladvise_hdr *ladvise_hdr;
2195 struct obd_ioobj ioo;
2196 struct lustre_handle lockh = { 0 };
2199 struct dt_object *dob;
2204 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2205 body = tsi->tsi_ost_body;
2207 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2208 RETURN(err_serious(-EPROTO));
2210 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2211 &RMF_OST_LADVISE_HDR);
2212 if (ladvise_hdr == NULL)
2213 RETURN(err_serious(-EPROTO));
2215 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2216 ladvise_hdr->lah_count < 1)
2217 RETURN(err_serious(-EPROTO));
2219 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2220 RETURN(err_serious(-EPROTO));
2222 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2223 if (ladvise == NULL)
2224 RETURN(err_serious(-EPROTO));
2226 num_advise = req_capsule_get_size(&req->rq_pill,
2227 &RMF_OST_LADVISE, RCL_CLIENT) /
2229 if (num_advise < ladvise_hdr->lah_count)
2230 RETURN(err_serious(-EPROTO));
2232 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2233 repbody->oa = body->oa;
2235 info = ofd_info_init(env, exp);
2237 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2238 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2242 fo = ofd_object_find(env, ofd, &info->fti_fid);
2247 LASSERT(fo != NULL);
2248 dob = ofd_object_child(fo);
2250 for (i = 0; i < num_advise; i++, ladvise++) {
2251 start = ladvise->lla_start;
2252 end = ladvise->lla_end;
2254 rc = err_serious(-EPROTO);
2258 /* Handle different advice types */
2259 switch (ladvise->lla_advice) {
2263 case LU_LADVISE_WILLREAD:
2267 ioo.ioo_oid = body->oa.o_oi;
2269 rc = tgt_extent_lock(exp->exp_obd->obd_namespace,
2270 &tsi->tsi_resid, start, end - 1,
2271 &lockh, LCK_PR, &flags);
2275 req->rq_status = ofd_ladvise_prefetch(env, fo,
2278 tgt_extent_unlock(&lockh, LCK_PR);
2280 case LU_LADVISE_DONTNEED:
2281 rc = dt_ladvise(env, dob, ladvise->lla_start,
2282 ladvise->lla_end, LU_LADVISE_DONTNEED);
2289 ofd_object_put(env, fo);
2290 req->rq_status = rc;
2295 * OFD request handler for OST_QUOTACTL RPC.
2297 * This is part of request processing to validate incoming request fields,
2298 * get the requested data from OSD and pack reply.
2300 * \param[in] tsi target session environment for this request
2302 * \retval 0 if successful
2303 * \retval negative value on error
2305 static int ofd_quotactl(struct tgt_session_info *tsi)
2307 struct obd_quotactl *oqctl, *repoqc;
2308 struct lu_nodemap *nodemap;
2314 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2316 RETURN(err_serious(-EPROTO));
2318 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2320 RETURN(err_serious(-ENOMEM));
2324 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2325 if (IS_ERR(nodemap))
2326 RETURN(PTR_ERR(nodemap));
2329 if (oqctl->qc_type == USRQUOTA)
2330 id = nodemap_map_id(nodemap, NODEMAP_UID,
2331 NODEMAP_CLIENT_TO_FS,
2333 else if (oqctl->qc_type == GRPQUOTA)
2334 id = nodemap_map_id(nodemap, NODEMAP_GID,
2335 NODEMAP_CLIENT_TO_FS,
2338 nodemap_putref(nodemap);
2340 if (repoqc->qc_id != id)
2341 swap(repoqc->qc_id, id);
2343 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2345 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2348 if (repoqc->qc_id != id)
2349 swap(repoqc->qc_id, id);
2355 * Calculate the amount of time for lock prolongation.
2357 * This is helper for ofd_prolong_extent_locks() function to get
2358 * the timeout extra time.
2360 * \param[in] req current request
2362 * \retval amount of time to extend the timeout with
2364 static inline int prolong_timeout(struct ptlrpc_request *req)
2366 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2370 return obd_timeout / 2;
2372 req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec;
2373 return max_t(time_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
2378 * Prolong lock timeout for the given extent.
2380 * This function finds all locks related with incoming request and
2381 * prolongs their timeout.
2383 * If a client is holding a lock for a long time while it sends
2384 * read or write RPCs to the OST for the object under this lock,
2385 * then we don't want the OST to evict the client. Otherwise,
2386 * if the network or disk is very busy then the client may not
2387 * be able to make any progress to clear out dirty pages under
2388 * the lock and the application will fail.
2390 * Every time a Bulk Read/Write (BRW) request arrives for the object
2391 * covered by the lock, extend the timeout on that lock. The RPC should
2392 * contain a lock handle for the lock it is using, but this
2393 * isn't handled correctly by all client versions, and the
2394 * request may cover multiple locks.
2396 * \param[in] tsi target session environment for this request
2397 * \param[in] data struct of data to prolong locks
2400 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2401 struct ldlm_prolong_args *data)
2403 struct obdo *oa = &tsi->tsi_ost_body->oa;
2404 struct ldlm_lock *lock;
2408 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2409 data->lpa_export = tsi->tsi_exp;
2410 data->lpa_resid = tsi->tsi_resid;
2412 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2413 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2414 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2415 data->lpa_extent.end);
2417 if (oa->o_valid & OBD_MD_FLHANDLE) {
2418 /* mostly a request should be covered by only one lock, try
2420 lock = ldlm_handle2lock(&oa->o_handle);
2422 /* Fast path to check if the lock covers the whole IO
2423 * region exclusively. */
2424 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2425 &data->lpa_extent)) {
2427 LASSERT(lock->l_export == data->lpa_export);
2428 ldlm_lock_prolong_one(lock, data);
2429 LDLM_LOCK_PUT(lock);
2432 lock->l_last_used = cfs_time_current();
2433 LDLM_LOCK_PUT(lock);
2437 ldlm_resource_prolong(data);
2442 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2444 * Determine if \a lock and the lock from request \a req are equivalent
2445 * by comparing their resource names, modes, and extents.
2447 * It is used to give priority to read and write RPCs being done
2448 * under this lock so that the client can drop the contended
2449 * lock more quickly and let other clients use it. This improves
2450 * overall performance in the case where the first client gets a
2451 * very large lock extent that prevents other clients from
2452 * submitting their writes.
2454 * \param[in] req ptlrpc_request being processed
2455 * \param[in] lock contended lock to match
2457 * \retval 1 if lock is matched
2458 * \retval 0 otherwise
2460 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2461 struct ldlm_lock *lock)
2463 struct niobuf_remote *rnb;
2464 struct obd_ioobj *ioo;
2465 enum ldlm_mode mode;
2466 struct ldlm_extent ext;
2467 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2471 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2472 LASSERT(ioo != NULL);
2474 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2475 LASSERT(rnb != NULL);
2477 ext.start = rnb->rnb_offset;
2478 rnb += ioo->ioo_bufcnt - 1;
2479 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2481 LASSERT(lock->l_resource != NULL);
2482 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2485 /* a bulk write can only hold a reference on a PW extent lock
2488 mode = LCK_PW | LCK_GROUP;
2489 if (opc == OST_READ)
2490 /* whereas a bulk read can be protected by either a PR or PW
2494 if (!(lock->l_granted_mode & mode))
2497 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2501 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2503 * Check for whether the given PTLRPC request (\a req) is blocking
2504 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2507 * \param[in] req the incoming request
2509 * \retval 1 if \a req is blocking an LDLM lock cancel
2510 * \retval 0 if it is not
2511 * \retval -ESTALE if lock is not found
2513 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2515 struct tgt_session_info *tsi;
2516 struct obd_ioobj *ioo;
2517 struct niobuf_remote *rnb;
2519 struct ldlm_prolong_args pa = { 0 };
2523 /* Don't use tgt_ses_info() to get session info, because lock_match()
2524 * can be called while request has no processing thread yet. */
2525 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2528 * Use LASSERT below because malformed RPCs should have
2529 * been filtered out in tgt_hpreq_handler().
2531 opc = lustre_msg_get_opc(req->rq_reqmsg);
2532 LASSERT(opc == OST_READ || opc == OST_WRITE);
2534 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2535 LASSERT(ioo != NULL);
2537 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2538 LASSERT(rnb != NULL);
2539 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2541 pa.lpa_mode = LCK_PW | LCK_GROUP;
2542 if (opc == OST_READ)
2543 pa.lpa_mode |= LCK_PR;
2545 pa.lpa_extent.start = rnb->rnb_offset;
2546 rnb += ioo->ioo_bufcnt - 1;
2547 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2549 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
2550 " (%llu->%llu)\n", tgt_name(tsi->tsi_tgt),
2551 current->comm, PFID(&tsi->tsi_fid), pa.lpa_extent.start,
2554 ofd_prolong_extent_locks(tsi, &pa);
2556 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2557 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2559 if (pa.lpa_blocks_cnt > 0)
2562 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2566 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2568 * Called after the request has been handled. It refreshes lock timeout again
2569 * so that client has more time to send lock cancel RPC.
2571 * \param[in] req request which is being processed.
2573 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2575 ofd_rw_hpreq_check(req);
2579 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2581 * This function checks if the given lock is the same by its resname, mode
2582 * and extent as one taken from the request.
2583 * It is used to give priority to punch/truncate RPCs that might lead to
2584 * the fastest release of that lock when a lock is contended.
2586 * \param[in] req ptlrpc_request being processed
2587 * \param[in] lock contended lock to match
2589 * \retval 1 if lock is matched
2590 * \retval 0 otherwise
2592 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2593 struct ldlm_lock *lock)
2595 struct tgt_session_info *tsi;
2597 struct ldlm_extent ext;
2601 /* Don't use tgt_ses_info() to get session info, because lock_match()
2602 * can be called while request has no processing thread yet. */
2603 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2606 * Use LASSERT below because malformed RPCs should have
2607 * been filtered out in tgt_hpreq_handler().
2609 LASSERT(tsi->tsi_ost_body != NULL);
2610 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2611 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2614 oa = &tsi->tsi_ost_body->oa;
2615 ext.start = oa->o_size;
2616 ext.end = oa->o_blocks;
2618 LASSERT(lock->l_resource != NULL);
2619 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2622 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2625 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2629 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2631 * High-priority queue request check for whether the given punch request
2632 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2633 * covered by an LDLM lock.
2637 * \param[in] req the incoming request
2639 * \retval 1 if \a req is blocking an LDLM lock cancel
2640 * \retval 0 if it is not
2641 * \retval -ESTALE if lock is not found
2643 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2645 struct tgt_session_info *tsi;
2647 struct ldlm_prolong_args pa = { 0 };
2651 /* Don't use tgt_ses_info() to get session info, because lock_match()
2652 * can be called while request has no processing thread yet. */
2653 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2654 LASSERT(tsi != NULL);
2655 oa = &tsi->tsi_ost_body->oa;
2657 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2658 oa->o_flags & OBD_FL_SRVLOCK));
2660 pa.lpa_mode = LCK_PW | LCK_GROUP;
2661 pa.lpa_extent.start = oa->o_size;
2662 pa.lpa_extent.end = oa->o_blocks;
2665 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2666 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2667 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2669 ofd_prolong_extent_locks(tsi, &pa);
2671 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2672 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2674 if (pa.lpa_blocks_cnt > 0)
2677 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2681 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2683 * Called after the request has been handled. It refreshes lock timeout again
2684 * so that client has more time to send lock cancel RPC.
2686 * \param[in] req request which is being processed.
2688 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2690 ofd_punch_hpreq_check(req);
2693 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2694 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2695 .hpreq_check = ofd_rw_hpreq_check,
2696 .hpreq_fini = ofd_rw_hpreq_fini
2699 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2700 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2701 .hpreq_check = ofd_punch_hpreq_check,
2702 .hpreq_fini = ofd_punch_hpreq_fini
2706 * Assign high priority operations to an IO request.
2708 * Check if the incoming request is a candidate for
2709 * high-priority processing. If it is, assign it a high
2710 * priority operations table.
2712 * \param[in] tsi target session environment for this request
2714 static void ofd_hp_brw(struct tgt_session_info *tsi)
2716 struct niobuf_remote *rnb;
2717 struct obd_ioobj *ioo;
2721 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2722 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2723 if (ioo->ioo_bufcnt > 0) {
2724 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2725 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2727 /* no high priority if server lock is needed */
2728 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2729 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2733 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2737 * Assign high priority operations to an punch request.
2739 * Check if the incoming request is a candidate for
2740 * high-priority processing. If it is, assign it a high
2741 * priority operations table.
2743 * \param[in] tsi target session environment for this request
2745 static void ofd_hp_punch(struct tgt_session_info *tsi)
2747 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2748 /* no high-priority if server lock is needed */
2749 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2750 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2751 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2752 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2754 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2757 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2758 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2759 #define OST_BRW_READ OST_READ
2760 #define OST_BRW_WRITE OST_WRITE
2763 * Table of OFD-specific request handlers
2765 * This table contains all opcodes accepted by OFD and
2766 * specifies handlers for them. The tgt_request_handler()
2767 * uses such table from each target to process incoming
2770 static struct tgt_handler ofd_tgt_handlers[] = {
2771 TGT_RPC_HANDLER(OST_FIRST_OPC,
2772 0, OST_CONNECT, tgt_connect,
2773 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2774 TGT_RPC_HANDLER(OST_FIRST_OPC,
2775 0, OST_DISCONNECT, tgt_disconnect,
2776 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2777 TGT_RPC_HANDLER(OST_FIRST_OPC,
2778 0, OST_SET_INFO, ofd_set_info_hdl,
2779 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2780 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2781 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
2782 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2783 OST_SETATTR, ofd_setattr_hdl),
2784 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2785 OST_CREATE, ofd_create_hdl),
2786 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2787 OST_DESTROY, ofd_destroy_hdl),
2788 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
2789 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2790 OST_BRW_READ, tgt_brw_read,
2792 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2793 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
2795 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2796 OST_PUNCH, ofd_punch_hdl,
2798 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
2799 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
2800 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_LADVISE, ofd_ladvise_hdl),
2803 static struct tgt_opc_slice ofd_common_slice[] = {
2805 .tos_opc_start = OST_FIRST_OPC,
2806 .tos_opc_end = OST_LAST_OPC,
2807 .tos_hs = ofd_tgt_handlers
2810 .tos_opc_start = OBD_FIRST_OPC,
2811 .tos_opc_end = OBD_LAST_OPC,
2812 .tos_hs = tgt_obd_handlers
2815 .tos_opc_start = LDLM_FIRST_OPC,
2816 .tos_opc_end = LDLM_LAST_OPC,
2817 .tos_hs = tgt_dlm_handlers
2820 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2821 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2822 .tos_hs = tgt_out_handlers
2825 .tos_opc_start = SEQ_FIRST_OPC,
2826 .tos_opc_end = SEQ_LAST_OPC,
2827 .tos_hs = seq_handlers
2830 .tos_opc_start = LFSCK_FIRST_OPC,
2831 .tos_opc_end = LFSCK_LAST_OPC,
2832 .tos_hs = tgt_lfsck_handlers
2835 .tos_opc_start = SEC_FIRST_OPC,
2836 .tos_opc_end = SEC_LAST_OPC,
2837 .tos_hs = tgt_sec_ctx_handlers
2844 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2845 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2848 * Implementation of lu_context_key::lct_key_exit.
2850 * Optional method called on lu_context_exit() for all allocated
2852 * It is used in OFD to sanitize context values which may be re-used
2853 * during another request processing by the same thread.
2855 * \param[in] ctx execution context
2856 * \param[in] key context key
2857 * \param[in] data ofd_thread_info
2859 static void ofd_key_exit(const struct lu_context *ctx,
2860 struct lu_context_key *key, void *data)
2862 struct ofd_thread_info *info = data;
2864 info->fti_env = NULL;
2865 info->fti_exp = NULL;
2868 info->fti_pre_version = 0;
2870 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2873 struct lu_context_key ofd_thread_key = {
2874 .lct_tags = LCT_DT_THREAD,
2875 .lct_init = ofd_key_init,
2876 .lct_fini = ofd_key_fini,
2877 .lct_exit = ofd_key_exit
2881 * Initialize OFD device according to parameters in the config log \a cfg.
2883 * This is the main starting point of OFD initialization. It fills all OFD
2884 * parameters with their initial values and calls other initializing functions
2885 * to set up all OFD subsystems.
2887 * \param[in] env execution environment
2888 * \param[in] m OFD device
2889 * \param[in] ldt LU device type of OFD
2890 * \param[in] cfg configuration log
2892 * \retval 0 if successful
2893 * \retval negative value on error
2895 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2896 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2898 const char *dev = lustre_cfg_string(cfg, 0);
2899 struct ofd_thread_info *info = NULL;
2900 struct obd_device *obd;
2901 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2902 struct obd_statfs *osfs;
2904 struct nm_config_file *nodemap_config;
2905 struct obd_device_target *obt;
2910 obd = class_name2obd(dev);
2912 CERROR("Cannot find obd with name %s\n", dev);
2916 rc = lu_env_refill((struct lu_env *)env);
2921 obt->obt_magic = OBT_MAGIC;
2923 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2924 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2926 spin_lock_init(&m->ofd_flags_lock);
2927 m->ofd_raid_degraded = 0;
2928 m->ofd_syncjournal = 0;
2930 tgd->tgd_grant_compat_disable = 0;
2931 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2934 spin_lock_init(&tgd->tgd_osfs_lock);
2935 tgd->tgd_osfs_age = cfs_time_shift_64(-1000);
2936 tgd->tgd_osfs_unstable = 0;
2937 tgd->tgd_statfs_inflight = 0;
2938 tgd->tgd_osfs_inflight = 0;
2941 spin_lock_init(&tgd->tgd_grant_lock);
2942 tgd->tgd_tot_dirty = 0;
2943 tgd->tgd_tot_granted = 0;
2944 tgd->tgd_tot_pending = 0;
2946 m->ofd_seq_count = 0;
2947 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2948 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2949 spin_lock_init(&m->ofd_inconsistency_lock);
2951 spin_lock_init(&m->ofd_batch_lock);
2952 init_rwsem(&m->ofd_lastid_rwsem);
2954 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2955 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2956 /* set this lu_device to obd, because error handling need it */
2957 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2959 rc = ofd_procfs_init(m);
2961 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2965 /* No connection accepted until configurations will finish */
2966 spin_lock(&obd->obd_dev_lock);
2967 obd->obd_no_conn = 1;
2968 spin_unlock(&obd->obd_dev_lock);
2969 obd->obd_replayable = 1;
2970 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2971 char *str = lustre_cfg_string(cfg, 4);
2973 if (strchr(str, 'n')) {
2974 CWARN("%s: recovery disabled\n", obd->obd_name);
2975 obd->obd_replayable = 0;
2979 info = ofd_info_init(env, NULL);
2981 GOTO(err_fini_proc, rc = -EFAULT);
2983 rc = ofd_stack_init(env, m, cfg);
2985 CERROR("Can't init device stack, rc %d\n", rc);
2986 GOTO(err_fini_proc, rc);
2989 ofd_procfs_add_brw_stats_symlink(m);
2991 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2992 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2993 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2994 LDLM_NAMESPACE_SERVER,
2995 LDLM_NAMESPACE_GREEDY,
2997 if (m->ofd_namespace == NULL)
2998 GOTO(err_fini_stack, rc = -ENOMEM);
2999 /* set obd_namespace for compatibility with old code */
3000 obd->obd_namespace = m->ofd_namespace;
3001 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
3002 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
3003 m->ofd_namespace->ns_lvbp = m;
3005 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3006 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
3008 dt_conf_get(env, m->ofd_osd, &m->ofd_lut.lut_dt_conf);
3010 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
3011 OBD_FAIL_OST_ALL_REQUEST_NET,
3012 OBD_FAIL_OST_ALL_REPLY_NET);
3014 GOTO(err_free_ns, rc);
3016 /* populate cached statfs data */
3017 osfs = &ofd_info(env)->fti_u.osfs;
3018 rc = tgt_statfs_internal(env, &m->ofd_lut, osfs, 0, NULL);
3020 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
3021 GOTO(err_fini_lut, rc);
3023 if (!is_power_of_2(osfs->os_bsize)) {
3024 CERROR("%s: blocksize (%d) is not a power of 2\n",
3025 obd->obd_name, osfs->os_bsize);
3026 GOTO(err_fini_lut, rc = -EPROTO);
3028 tgd->tgd_blockbits = fls(osfs->os_bsize) - 1;
3030 if (DT_DEF_BRW_SIZE < (1U << tgd->tgd_blockbits))
3031 m->ofd_brw_size = 1U << tgd->tgd_blockbits;
3033 m->ofd_brw_size = DT_DEF_BRW_SIZE;
3035 m->ofd_cksum_types_supported = cksum_types_supported_server();
3036 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
3037 if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
3038 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
3040 rc = ofd_fs_setup(env, m, obd);
3042 GOTO(err_fini_lut, rc);
3044 fid.f_seq = FID_SEQ_LOCAL_NAME;
3047 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
3050 GOTO(err_fini_fs, rc);
3052 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
3054 if (IS_ERR(nodemap_config)) {
3055 rc = PTR_ERR(nodemap_config);
3057 GOTO(err_fini_los, rc);
3059 obt->obt_nodemap_config_file = nodemap_config;
3062 rc = ofd_start_inconsistency_verification_thread(m);
3064 GOTO(err_fini_nm, rc);
3066 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
3071 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
3072 obt->obt_nodemap_config_file = NULL;
3074 local_oid_storage_fini(env, m->ofd_los);
3077 ofd_fs_cleanup(env, m);
3079 tgt_fini(env, &m->ofd_lut);
3081 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3082 obd->obd_namespace = m->ofd_namespace = NULL;
3084 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3091 * Stop the OFD device
3093 * This function stops the OFD device and all its subsystems.
3094 * This is the end of OFD lifecycle.
3096 * \param[in] env execution environment
3097 * \param[in] m OFD device
3099 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3101 struct obd_device *obd = ofd_obd(m);
3102 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3103 struct lfsck_stop stop;
3105 stop.ls_status = LS_PAUSED;
3107 lfsck_stop(env, m->ofd_osd, &stop);
3108 target_recovery_fini(obd);
3109 if (m->ofd_namespace != NULL)
3110 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3111 d->ld_obd->obd_force);
3113 obd_exports_barrier(obd);
3114 obd_zombie_barrier();
3116 tgt_fini(env, &m->ofd_lut);
3117 ofd_stop_inconsistency_verification_thread(m);
3118 lfsck_degister(env, m->ofd_osd);
3119 ofd_fs_cleanup(env, m);
3120 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3121 obd->u.obt.obt_nodemap_config_file = NULL;
3123 if (m->ofd_los != NULL) {
3124 local_oid_storage_fini(env, m->ofd_los);
3128 if (m->ofd_namespace != NULL) {
3129 ldlm_namespace_free_post(m->ofd_namespace);
3130 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3133 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3135 LASSERT(atomic_read(&d->ld_ref) == 0);
3136 server_put_mount(obd->obd_name, true);
3141 * Implementation of lu_device_type_operations::ldto_device_fini.
3143 * Finalize device. Dual to ofd_device_init(). It is called from
3144 * obd_precleanup() and stops the current device.
3146 * \param[in] env execution environment
3147 * \param[in] d LU device of OFD
3151 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3152 struct lu_device *d)
3155 ofd_fini(env, ofd_dev(d));
3160 * Implementation of lu_device_type_operations::ldto_device_free.
3162 * Free OFD device. Dual to ofd_device_alloc().
3164 * \param[in] env execution environment
3165 * \param[in] d LU device of OFD
3169 static struct lu_device *ofd_device_free(const struct lu_env *env,
3170 struct lu_device *d)
3172 struct ofd_device *m = ofd_dev(d);
3174 dt_device_fini(&m->ofd_dt_dev);
3180 * Implementation of lu_device_type_operations::ldto_device_alloc.
3182 * This function allocates the new OFD device. It is called from
3183 * obd_setup() if OBD device had lu_device_type defined.
3185 * \param[in] env execution environment
3186 * \param[in] t lu_device_type of OFD device
3187 * \param[in] cfg configuration log
3189 * \retval pointer to the lu_device of just allocated OFD
3190 * \retval ERR_PTR of return value on error
3192 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3193 struct lu_device_type *t,
3194 struct lustre_cfg *cfg)
3196 struct ofd_device *m;
3197 struct lu_device *l;
3202 return ERR_PTR(-ENOMEM);
3204 l = &m->ofd_dt_dev.dd_lu_dev;
3205 dt_device_init(&m->ofd_dt_dev, t);
3206 rc = ofd_init0(env, m, t, cfg);
3208 ofd_device_free(env, l);
3215 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3216 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3218 static struct lu_device_type_operations ofd_device_type_ops = {
3219 .ldto_init = ofd_type_init,
3220 .ldto_fini = ofd_type_fini,
3222 .ldto_start = ofd_type_start,
3223 .ldto_stop = ofd_type_stop,
3225 .ldto_device_alloc = ofd_device_alloc,
3226 .ldto_device_free = ofd_device_free,
3227 .ldto_device_fini = ofd_device_fini
3230 static struct lu_device_type ofd_device_type = {
3231 .ldt_tags = LU_DEVICE_DT,
3232 .ldt_name = LUSTRE_OST_NAME,
3233 .ldt_ops = &ofd_device_type_ops,
3234 .ldt_ctx_tags = LCT_DT_THREAD
3238 * Initialize OFD module.
3240 * This function is called upon module loading. It registers OFD device type
3241 * and prepares all in-memory structures used by all OFD devices.
3243 * \retval 0 if successful
3244 * \retval negative value on error
3246 static int __init ofd_init(void)
3250 rc = lu_kmem_init(ofd_caches);
3254 rc = ofd_fmd_init();
3256 lu_kmem_fini(ofd_caches);
3260 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3261 LUSTRE_OST_NAME, &ofd_device_type);
3268 * This function is called upon OFD module unloading.
3269 * It frees all related structures and unregisters OFD device type.
3271 static void __exit ofd_exit(void)
3274 lu_kmem_fini(ofd_caches);
3275 class_unregister_type(LUSTRE_OST_NAME);
3278 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3279 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3280 MODULE_VERSION(LUSTRE_VERSION_STRING);
3281 MODULE_LICENSE("GPL");
3283 module_init(ofd_init);
3284 module_exit(ofd_exit);