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 <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>
79 #include <lustre_log.h>
81 #include "ofd_internal.h"
83 /* Slab for OFD object allocation */
84 static struct kmem_cache *ofd_object_kmem;
86 static struct lu_kmem_descr ofd_caches[] = {
88 .ckd_cache = &ofd_object_kmem,
89 .ckd_name = "ofd_obj",
90 .ckd_size = sizeof(struct ofd_object)
98 * Connect OFD to the next device in the stack.
100 * This function is used for device stack configuration and links OFD
101 * device with bottom OSD device.
103 * \param[in] env execution environment
104 * \param[in] m OFD device
105 * \param[in] next name of next device in the stack
106 * \param[out] exp export to return
108 * \retval 0 and export in \a exp if successful
109 * \retval negative value on error
111 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
112 const char *next, struct obd_export **exp)
114 struct obd_connect_data *data = NULL;
115 struct obd_device *obd;
121 GOTO(out, rc = -ENOMEM);
123 obd = class_name2obd(next);
125 CERROR("%s: can't locate next device: %s\n",
127 GOTO(out, rc = -ENOTCONN);
130 data->ocd_connect_flags = OBD_CONNECT_VERSION;
131 data->ocd_version = LUSTRE_VERSION_CODE;
133 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
135 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
136 ofd_name(m), next, rc);
140 m->ofd_dt_dev.dd_lu_dev.ld_site =
141 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
142 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
143 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
144 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
153 * Initialize stack of devices.
155 * This function initializes OFD-OSD device stack to serve OST requests
157 * \param[in] env execution environment
158 * \param[in] m OFD device
159 * \param[in] cfg Lustre config for this server
161 * \retval 0 if successful
162 * \retval negative value on error
164 static int ofd_stack_init(const struct lu_env *env,
165 struct ofd_device *m, struct lustre_cfg *cfg)
167 const char *dev = lustre_cfg_string(cfg, 0);
169 struct ofd_thread_info *info = ofd_info(env);
170 struct lustre_mount_info *lmi;
171 struct lustre_mount_data *lmd;
177 lmi = server_get_mount(dev);
179 CERROR("Cannot get mount info for %s!\n", dev);
183 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
184 if (lmd != NULL && lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
185 m->ofd_skip_lfsck = 1;
187 /* find bottom osd */
188 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
192 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
193 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
194 OBD_FREE(osdname, MTI_NAME_MAXLEN);
198 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
200 m->ofd_osd = lu2dt_dev(d);
202 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
203 "%s-osd", lustre_cfg_string(cfg, 0));
209 * Finalize the device stack OFD-OSD.
211 * This function cleans OFD-OSD device stack and
212 * disconnects OFD from the OSD.
214 * \param[in] env execution environment
215 * \param[in] m OFD device
216 * \param[in] top top device of stack
218 * \retval 0 if successful
219 * \retval negative value on error
221 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
222 struct lu_device *top)
224 struct obd_device *obd = ofd_obd(m);
225 struct lustre_cfg_bufs bufs;
226 struct lustre_cfg *lcfg;
231 lu_site_purge(env, top->ld_site, ~0);
232 /* process cleanup, pass mdt obd name to get obd umount flags */
233 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
238 lustre_cfg_bufs_set_string(&bufs, 1, flags);
239 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
244 top->ld_ops->ldo_process_config(env, top, lcfg);
245 lustre_cfg_free(lcfg);
247 lu_site_purge(env, top->ld_site, ~0);
248 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
249 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
250 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
253 LASSERT(m->ofd_osd_exp);
254 obd_disconnect(m->ofd_osd_exp);
259 /* For interoperability, see mdt_interop_param[]. */
260 static struct cfg_interop_param ofd_interop_param[] = {
261 { "ost.quota_type", NULL },
266 * Check if parameters are symlinks to the OSD.
268 * Some parameters were moved from ofd to osd and only their
269 * symlinks were kept in ofd by LU-3106. They are:
270 * -writehthrough_cache_enable
271 * -readcache_max_filesize
275 * Since they are not included by the static lprocfs var list, a pre-check
276 * is added for them to avoid "unknown param" errors. If they are matched
277 * in this check, they will be passed to the OSD directly.
279 * \param[in] param parameters to check
281 * \retval true if param is symlink to OSD param
284 static bool match_symlink_param(char *param)
289 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
290 sval = strchr(param, '=');
292 paramlen = sval - param;
293 if (strncmp(param, "writethrough_cache_enable",
295 strncmp(param, "readcache_max_filesize",
297 strncmp(param, "read_cache_enable",
299 strncmp(param, "brw_stats", paramlen) == 0)
308 * Process various configuration parameters.
310 * This function is used by MGS to process specific configurations and
311 * pass them through to the next device in server stack, i.e. the OSD.
313 * \param[in] env execution environment
314 * \param[in] d LU device of OFD
315 * \param[in] cfg parameters to process
317 * \retval 0 if successful
318 * \retval negative value on error
320 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
321 struct lustre_cfg *cfg)
323 struct ofd_device *m = ofd_dev(d);
324 struct dt_device *dt_next = m->ofd_osd;
325 struct lu_device *next = &dt_next->dd_lu_dev;
330 switch (cfg->lcfg_command) {
332 struct obd_device *obd = ofd_obd(m);
333 /* For interoperability */
334 struct cfg_interop_param *ptr = NULL;
335 struct lustre_cfg *old_cfg = NULL;
338 param = lustre_cfg_string(cfg, 1);
340 CERROR("param is empty\n");
345 ptr = class_find_old_param(param, ofd_interop_param);
347 if (ptr->new_param == NULL) {
349 CWARN("For interoperability, skip this %s."
350 " It is obsolete.\n", ptr->old_param);
354 CWARN("Found old param %s, changed it to %s.\n",
355 ptr->old_param, ptr->new_param);
358 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
365 if (match_symlink_param(param)) {
366 rc = next->ld_ops->ldo_process_config(env, next, cfg);
370 rc = class_process_proc_param(PARAM_OST, obd->obd_vars, cfg,
372 if (rc > 0 || rc == -ENOSYS) {
373 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
375 /* we don't understand; pass it on */
376 rc = next->ld_ops->ldo_process_config(env, next, cfg);
380 case LCFG_SPTLRPC_CONF: {
385 /* others are passed further */
386 rc = next->ld_ops->ldo_process_config(env, next, cfg);
393 * Implementation of lu_object_operations::loo_object_init for OFD
395 * Allocate just the next object (OSD) in stack.
397 * \param[in] env execution environment
398 * \param[in] o lu_object of OFD object
399 * \param[in] conf additional configuration parameters, not used here
401 * \retval 0 if successful
402 * \retval negative value on error
404 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
405 const struct lu_object_conf *conf)
407 struct ofd_device *d = ofd_dev(o->lo_dev);
408 struct lu_device *under;
409 struct lu_object *below;
414 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
415 PFID(lu_object_fid(o)));
417 under = &d->ofd_osd->dd_lu_dev;
418 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
420 lu_object_add(o, below);
428 * Implementation of lu_object_operations::loo_object_free.
430 * Finish OFD object lifecycle and free its memory.
432 * \param[in] env execution environment
433 * \param[in] o LU object of OFD object
435 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
437 struct ofd_object *of = ofd_obj(o);
438 struct lu_object_header *h;
443 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
444 PFID(lu_object_fid(o)));
447 lu_object_header_fini(h);
448 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
453 * Implementation of lu_object_operations::loo_object_print.
455 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
456 * LU_OBJECT_DEBUG() for more details about the compound object printing.
458 * \param[in] env execution environment
459 * \param[in] cookie opaque data passed to the printer function
460 * \param[in] p printer function to use
461 * \param[in] o LU object of OFD object
463 * \retval 0 if successful
464 * \retval negative value on error
466 static int ofd_object_print(const struct lu_env *env, void *cookie,
467 lu_printer_t p, const struct lu_object *o)
469 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
472 static struct lu_object_operations ofd_obj_ops = {
473 .loo_object_init = ofd_object_init,
474 .loo_object_free = ofd_object_free,
475 .loo_object_print = ofd_object_print
479 * Implementation of lu_device_operations::lod_object_alloc.
481 * This function allocates OFD part of compound OFD-OSD object and
482 * initializes its header, because OFD is the top device in stack
484 * \param[in] env execution environment
485 * \param[in] hdr object header, NULL for OFD
486 * \param[in] d lu_device
488 * \retval allocated object if successful
489 * \retval NULL value on failed allocation
491 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
492 const struct lu_object_header *hdr,
495 struct ofd_object *of;
499 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
502 struct lu_object_header *h;
504 o = &of->ofo_obj.do_lu;
506 lu_object_header_init(h);
507 lu_object_init(o, h, d);
508 lu_object_add_top(h, o);
509 o->lo_ops = &ofd_obj_ops;
517 * Return the result of LFSCK run to the OFD.
519 * Notify OFD about result of LFSCK run. That may block the new object
520 * creation until problem is fixed by LFSCK.
522 * \param[in] env execution environment
523 * \param[in] data pointer to the OFD device
524 * \param[in] event LFSCK event type
526 * \retval 0 if successful
527 * \retval negative value on unknown event
529 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
530 enum lfsck_events event)
532 struct ofd_device *ofd = data;
533 struct obd_device *obd = ofd_obd(ofd);
536 case LE_LASTID_REBUILDING:
537 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
538 "on the device until the LAST_ID rebuilt successfully.\n",
540 down_write(&ofd->ofd_lastid_rwsem);
541 ofd->ofd_lastid_rebuilding = 1;
542 up_write(&ofd->ofd_lastid_rwsem);
544 case LE_LASTID_REBUILT: {
545 down_write(&ofd->ofd_lastid_rwsem);
546 ofd_seqs_free(env, ofd);
547 ofd->ofd_lastid_rebuilding = 0;
548 ofd->ofd_lastid_gen++;
549 up_write(&ofd->ofd_lastid_rwsem);
550 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
555 CERROR("%s: unknown lfsck event: rc = %d\n",
556 ofd_name(ofd), event);
564 * Implementation of lu_device_operations::ldo_prepare.
566 * This method is called after layer has been initialized and before it starts
567 * serving user requests. In OFD it starts lfsk check routines and initializes
570 * \param[in] env execution environment
571 * \param[in] pdev higher device in stack, NULL for OFD
572 * \param[in] dev lu_device of OFD device
574 * \retval 0 if successful
575 * \retval negative value on error
577 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
578 struct lu_device *dev)
580 struct ofd_thread_info *info;
581 struct ofd_device *ofd = ofd_dev(dev);
582 struct obd_device *obd = ofd_obd(ofd);
583 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
588 info = ofd_info_init(env, NULL);
592 /* initialize lower device */
593 rc = next->ld_ops->ldo_prepare(env, dev, next);
597 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
598 ofd_lfsck_out_notify, ofd, false);
600 CERROR("%s: failed to initialize lfsck: rc = %d\n",
605 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
606 /* The LFSCK instance is registered just now, so it must be there when
607 * register the namespace to such instance. */
608 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
610 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
611 LASSERT(obd->obd_no_conn);
612 spin_lock(&obd->obd_dev_lock);
613 obd->obd_no_conn = 0;
614 spin_unlock(&obd->obd_dev_lock);
616 if (obd->obd_recovering == 0)
617 ofd_postrecov(env, ofd);
623 * Implementation of lu_device_operations::ldo_recovery_complete.
625 * This method notifies all layers about 'recovery complete' event. That means
626 * device is in full state and consistent. An OFD calculates available grant
627 * space upon this event.
629 * \param[in] env execution environment
630 * \param[in] dev lu_device of OFD device
632 * \retval 0 if successful
633 * \retval negative value on error
635 static int ofd_recovery_complete(const struct lu_env *env,
636 struct lu_device *dev)
638 struct ofd_thread_info *oti = ofd_info(env);
639 struct ofd_device *ofd = ofd_dev(dev);
640 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
646 * Grant space for object precreation on the self export.
647 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
648 * is enough to create 10k objects. More space is then acquired for
649 * precreation in ofd_grant_create().
651 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
652 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
653 oti->fti_ocd.ocd_grant *= ofd->ofd_dt_conf.ddp_inodespace;
654 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
655 OBD_CONNECT_GRANT_PARAM;
656 ofd_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
658 rc = next->ld_ops->ldo_recovery_complete(env, next);
663 * lu_device_operations matrix for OFD device.
665 static struct lu_device_operations ofd_lu_ops = {
666 .ldo_object_alloc = ofd_object_alloc,
667 .ldo_process_config = ofd_process_config,
668 .ldo_recovery_complete = ofd_recovery_complete,
669 .ldo_prepare = ofd_prepare,
672 LPROC_SEQ_FOPS(lprocfs_nid_stats_clear);
675 * Initialize all needed procfs entries for OFD device.
677 * \param[in] ofd OFD device
679 * \retval 0 if successful
680 * \retval negative value on error
682 static int ofd_procfs_init(struct ofd_device *ofd)
684 struct obd_device *obd = ofd_obd(ofd);
685 struct proc_dir_entry *entry;
690 /* lprocfs must be setup before the ofd so state can be safely added
691 * to /proc incrementally as the ofd is setup */
692 obd->obd_vars = lprocfs_ofd_obd_vars;
693 rc = lprocfs_obd_setup(obd);
695 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
700 rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
702 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
704 GOTO(obd_cleanup, rc);
707 obd->obd_uses_nid_stats = 1;
709 entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
712 CERROR("%s: error %d setting up lprocfs for %s\n",
713 obd->obd_name, rc, "exports");
714 GOTO(obd_cleanup, rc);
716 obd->obd_proc_exports_entry = entry;
718 entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
719 obd, &lprocfs_nid_stats_clear_fops);
722 CERROR("%s: add proc entry 'clear' failed: %d.\n",
724 GOTO(obd_cleanup, rc);
727 ofd_stats_counter_init(obd->obd_stats);
729 rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
730 ofd_stats_counter_init);
732 GOTO(obd_cleanup, rc);
735 lprocfs_obd_cleanup(obd);
736 lprocfs_free_obd_stats(obd);
742 * Expose OSD statistics to OFD layer.
744 * The osd interfaces to the backend file system exposes useful data
745 * such as brw_stats and read or write cache states. This same data
746 * needs to be exposed into the obdfilter (ofd) layer to maintain
747 * backwards compatibility. This function creates the symlinks in the
748 * proc layer to enable this.
750 * \param[in] ofd OFD device
752 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
754 struct obd_device *obd = ofd_obd(ofd);
755 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
757 if (obd->obd_proc_entry == NULL)
760 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
761 "../../%s/%s/brw_stats",
762 osd_obd->obd_type->typ_name, obd->obd_name);
764 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
765 "../../%s/%s/read_cache_enable",
766 osd_obd->obd_type->typ_name, obd->obd_name);
768 lprocfs_add_symlink("readcache_max_filesize",
770 "../../%s/%s/readcache_max_filesize",
771 osd_obd->obd_type->typ_name, obd->obd_name);
773 lprocfs_add_symlink("writethrough_cache_enable",
775 "../../%s/%s/writethrough_cache_enable",
776 osd_obd->obd_type->typ_name, obd->obd_name);
780 * Cleanup all procfs entries in OFD.
782 * \param[in] ofd OFD device
784 static void ofd_procfs_fini(struct ofd_device *ofd)
786 struct obd_device *obd = ofd_obd(ofd);
788 lprocfs_free_per_client_stats(obd);
789 lprocfs_obd_cleanup(obd);
790 lprocfs_free_obd_stats(obd);
791 lprocfs_job_stats_fini(obd);
795 * Stop SEQ/FID server on OFD.
797 * \param[in] env execution environment
798 * \param[in] ofd OFD device
800 * \retval 0 if successful
801 * \retval negative value on error
803 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
805 return seq_site_fini(env, &ofd->ofd_seq_site);
809 * Start SEQ/FID server on OFD.
811 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
812 * It also connects to the master server to get own FID sequence (SEQ) range
813 * to this particular OFD. Typically that happens when the OST is first
814 * formatted or in the rare case that it exhausts the local sequence range.
816 * The sequence range is allocated out to the MDTs for OST object allocations,
817 * and not directly to the clients.
819 * \param[in] env execution environment
820 * \param[in] ofd OFD device
822 * \retval 0 if successful
823 * \retval negative value on error
825 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
827 struct seq_server_site *ss = &ofd->ofd_seq_site;
828 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
829 char *obd_name = ofd_name(ofd);
833 ss = &ofd->ofd_seq_site;
834 lu->ld_site->ld_seq_site = ss;
835 ss->ss_lu = lu->ld_site;
836 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
838 OBD_ALLOC(name, sizeof(obd_name) * 2 + 10);
842 OBD_ALLOC_PTR(ss->ss_server_seq);
843 if (ss->ss_server_seq == NULL)
844 GOTO(out_name, rc = -ENOMEM);
846 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
847 LUSTRE_SEQ_SERVER, ss);
849 CERROR("%s : seq server init error %d\n", obd_name, rc);
850 GOTO(out_server, rc);
852 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
854 OBD_ALLOC_PTR(ss->ss_client_seq);
855 if (ss->ss_client_seq == NULL)
856 GOTO(out_server, rc = -ENOMEM);
859 * It always printed as "%p", so that the name is unique in the kernel,
860 * even if the filesystem is mounted twice. So sizeof(.) * 2 is enough.
862 snprintf(name, sizeof(obd_name) * 2 + 7, "%p-super", obd_name);
863 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
866 CERROR("%s : seq client init error %d\n", obd_name, rc);
867 GOTO(out_client, rc);
870 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
874 seq_client_fini(ss->ss_client_seq);
875 OBD_FREE_PTR(ss->ss_client_seq);
876 ss->ss_client_seq = NULL;
878 seq_server_fini(ss->ss_server_seq, env);
879 OBD_FREE_PTR(ss->ss_server_seq);
880 ss->ss_server_seq = NULL;
883 OBD_FREE(name, sizeof(obd_name) * 2 + 10);
889 * OFD request handler for OST_SET_INFO RPC.
891 * This is OFD-specific part of request handling
893 * \param[in] tsi target session environment for this request
895 * \retval 0 if successful
896 * \retval negative value on error
898 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
900 struct ptlrpc_request *req = tgt_ses_req(tsi);
901 struct ost_body *body = NULL, *repbody;
902 void *key, *val = NULL;
903 int keylen, vallen, rc = 0;
904 bool is_grant_shrink;
908 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
910 DEBUG_REQ(D_HA, req, "no set_info key");
911 RETURN(err_serious(-EFAULT));
913 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
916 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
918 DEBUG_REQ(D_HA, req, "no set_info val");
919 RETURN(err_serious(-EFAULT));
921 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
924 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
926 /* In this case the value is actually an RMF_OST_BODY, so we
927 * transmutate the type of this PTLRPC */
928 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
930 rc = req_capsule_server_pack(tsi->tsi_pill);
934 if (is_grant_shrink) {
935 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
937 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
940 /** handle grant shrink, similar to a read request */
941 ofd_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
943 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
945 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
948 CERROR("%s: Unsupported key %s\n",
949 tgt_name(tsi->tsi_tgt), (char *)key);
952 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
959 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
961 * This function returns a list of extents which describes how a file's
962 * blocks are laid out on the disk.
964 * \param[in] env execution environment
965 * \param[in] ofd OFD device
966 * \param[in] fid FID of object
967 * \param[in] fiemap fiemap structure to fill with data
969 * \retval 0 if \a fiemap is filled with data successfully
970 * \retval negative value on error
972 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
973 struct lu_fid *fid, struct fiemap *fiemap)
975 struct ofd_object *fo;
978 fo = ofd_object_find(env, ofd, fid);
980 CERROR("%s: error finding object "DFID"\n",
981 ofd_name(ofd), PFID(fid));
985 ofd_read_lock(env, fo);
986 if (ofd_object_exists(fo))
987 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
990 ofd_read_unlock(env, fo);
991 ofd_object_put(env, fo);
995 struct locked_region {
996 struct list_head list;
997 struct lustre_handle lh;
1001 * Lock single extent and save lock handle in the list.
1003 * This is supplemental function for lock_zero_regions(). It allocates
1004 * new locked_region structure and locks it with extent lock, then adds
1005 * it to the list of all such regions.
1007 * \param[in] ns LDLM namespace
1008 * \param[in] res_id resource ID
1009 * \param[in] begin start of region
1010 * \param[in] end end of region
1011 * \param[in] locked list head of regions list
1013 * \retval 0 if successful locking
1014 * \retval negative value on error
1016 static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1017 unsigned long long begin, unsigned long long end,
1018 struct list_head *locked)
1020 struct locked_region *region = NULL;
1024 LASSERT(begin <= end);
1025 OBD_ALLOC_PTR(region);
1029 rc = tgt_extent_lock(ns, res_id, begin, end, ®ion->lh,
1034 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
1036 list_add(®ion->list, locked);
1042 * Lock the sparse areas of given resource.
1044 * The locking of sparse areas will cause dirty data to be flushed back from
1045 * clients. This is used when getting the FIEMAP of an object to make sure
1046 * there is no unaccounted cached data on clients.
1048 * This function goes through \a fiemap list of extents and locks only sparse
1049 * areas between extents.
1051 * \param[in] ns LDLM namespace
1052 * \param[in] res_id resource ID
1053 * \param[in] fiemap file extents mapping on disk
1054 * \param[in] locked list head of regions list
1056 * \retval 0 if successful
1057 * \retval negative value on error
1059 static int lock_zero_regions(struct ldlm_namespace *ns,
1060 struct ldlm_res_id *res_id,
1061 struct fiemap *fiemap,
1062 struct list_head *locked)
1064 __u64 begin = fiemap->fm_start;
1067 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1071 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1072 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1073 if (fiemap_start[i].fe_logical > begin) {
1074 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1075 begin, fiemap_start[i].fe_logical);
1076 rc = lock_region(ns, res_id, begin,
1077 fiemap_start[i].fe_logical, locked);
1082 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1085 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1086 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1087 begin, fiemap->fm_start + fiemap->fm_length);
1088 rc = lock_region(ns, res_id, begin,
1089 fiemap->fm_start + fiemap->fm_length, locked);
1096 * Unlock all previously locked sparse areas for given resource.
1098 * This function goes through list of locked regions, unlocking and freeing
1101 * \param[in] ns LDLM namespace
1102 * \param[in] locked list head of regions list
1105 unlock_zero_regions(struct ldlm_namespace *ns, struct list_head *locked)
1107 struct locked_region *entry, *temp;
1109 list_for_each_entry_safe(entry, temp, locked, list) {
1110 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
1111 tgt_extent_unlock(&entry->lh, LCK_PR);
1112 list_del(&entry->list);
1113 OBD_FREE_PTR(entry);
1118 * OFD request handler for OST_GET_INFO RPC.
1120 * This is OFD-specific part of request handling. The OFD-specific keys are:
1121 * - KEY_LAST_ID (obsolete)
1125 * This function reads needed data from storage and fills reply with it.
1127 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1128 * and is kept for compatibility.
1130 * \param[in] tsi target session environment for this request
1132 * \retval 0 if successful
1133 * \retval negative value on error
1135 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1137 struct obd_export *exp = tsi->tsi_exp;
1138 struct ofd_device *ofd = ofd_exp(exp);
1139 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1142 int replylen, rc = 0;
1146 /* this common part for get_info rpc */
1147 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1149 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1150 RETURN(err_serious(-EPROTO));
1152 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1155 if (KEY_IS(KEY_LAST_ID)) {
1157 struct ofd_seq *oseq;
1159 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1160 rc = req_capsule_server_pack(tsi->tsi_pill);
1162 RETURN(err_serious(rc));
1164 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1166 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1167 (u64)exp->exp_filter_data.fed_group);
1171 *last_id = ofd_seq_last_oid(oseq);
1172 ofd_seq_put(tsi->tsi_env, oseq);
1173 } else if (KEY_IS(KEY_FIEMAP)) {
1174 struct ll_fiemap_info_key *fm_key;
1175 struct fiemap *fiemap;
1178 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1180 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1181 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1183 RETURN(err_serious(rc));
1185 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1187 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1189 replylen = fiemap_count_to_size(
1190 fm_key->lfik_fiemap.fm_extent_count);
1191 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1192 RCL_SERVER, replylen);
1194 rc = req_capsule_server_pack(tsi->tsi_pill);
1196 RETURN(err_serious(rc));
1198 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1202 *fiemap = fm_key->lfik_fiemap;
1203 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1205 /* LU-3219: Lock the sparse areas to make sure dirty
1206 * flushed back from client, then call fiemap again. */
1207 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1208 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1209 struct list_head locked;
1211 INIT_LIST_HEAD(&locked);
1212 ost_fid_build_resid(fid, &fti->fti_resid);
1213 rc = lock_zero_regions(ofd->ofd_namespace,
1214 &fti->fti_resid, fiemap,
1216 if (rc == 0 && !list_empty(&locked)) {
1217 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1219 unlock_zero_regions(ofd->ofd_namespace,
1223 } else if (KEY_IS(KEY_LAST_FID)) {
1224 struct ofd_device *ofd = ofd_exp(exp);
1225 struct ofd_seq *oseq;
1229 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1230 rc = req_capsule_server_pack(tsi->tsi_pill);
1232 RETURN(err_serious(rc));
1234 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1236 RETURN(err_serious(-EPROTO));
1238 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1240 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1244 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1245 ostid_seq(&fti->fti_ostid));
1247 RETURN(PTR_ERR(oseq));
1249 rc = ostid_to_fid(fid, &oseq->os_oi,
1250 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1254 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1257 ofd_seq_put(tsi->tsi_env, oseq);
1259 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1263 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1270 * OFD request handler for OST_GETATTR RPC.
1272 * This is OFD-specific part of request handling. It finds the OFD object
1273 * by its FID, gets attributes from storage and packs result to the reply.
1275 * \param[in] tsi target session environment for this request
1277 * \retval 0 if successful
1278 * \retval negative value on error
1280 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1282 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1283 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1284 struct ost_body *repbody;
1285 struct lustre_handle lh = { 0 };
1286 struct ofd_object *fo;
1288 enum ldlm_mode lock_mode = LCK_PR;
1293 LASSERT(tsi->tsi_ost_body != NULL);
1295 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1296 if (repbody == NULL)
1299 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1300 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1302 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1303 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1306 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1309 rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace,
1310 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1316 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1318 GOTO(out, rc = PTR_ERR(fo));
1320 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1324 obdo_from_la(&repbody->oa, &fti->fti_attr,
1325 OFD_VALID_FLAGS | LA_UID | LA_GID);
1327 /* Store object version in reply */
1328 curr_version = dt_version_get(tsi->tsi_env,
1329 ofd_object_child(fo));
1330 if ((__s64)curr_version != -EOPNOTSUPP) {
1331 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1332 repbody->oa.o_data_version = curr_version;
1336 ofd_object_put(tsi->tsi_env, fo);
1339 tgt_extent_unlock(&lh, lock_mode);
1341 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1344 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1345 repbody->oa.o_flags = OBD_FL_FLUSH;
1351 * OFD request handler for OST_SETATTR RPC.
1353 * This is OFD-specific part of request handling. It finds the OFD object
1354 * by its FID, sets attributes from request and packs result to the reply.
1356 * \param[in] tsi target session environment for this request
1358 * \retval 0 if successful
1359 * \retval negative value on error
1361 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1363 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1364 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1365 struct ost_body *body = tsi->tsi_ost_body;
1366 struct ost_body *repbody;
1367 struct ldlm_resource *res;
1368 struct ofd_object *fo;
1369 struct filter_fid *ff = NULL;
1374 LASSERT(body != NULL);
1376 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1377 if (repbody == NULL)
1380 repbody->oa.o_oi = body->oa.o_oi;
1381 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1383 /* This would be very bad - accidentally truncating a file when
1384 * changing the time or similar - bug 12203. */
1385 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1386 body->oa.o_size != OBD_OBJECT_EOF) {
1387 static char mdsinum[48];
1389 if (body->oa.o_valid & OBD_MD_FLFID)
1390 snprintf(mdsinum, sizeof(mdsinum) - 1,
1391 "of parent "DFID, body->oa.o_parent_seq,
1392 body->oa.o_parent_oid, 0);
1396 CERROR("%s: setattr from %s is trying to truncate object "DFID
1397 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1398 PFID(&tsi->tsi_fid), mdsinum);
1402 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1404 GOTO(out, rc = PTR_ERR(fo));
1406 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1407 fti->fti_attr.la_valid &= ~LA_TYPE;
1409 if (body->oa.o_valid & OBD_MD_FLFID) {
1410 ff = &fti->fti_mds_fid;
1411 ofd_prepare_fidea(ff, &body->oa);
1414 /* setting objects attributes (including owner/group) */
1415 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff);
1419 obdo_from_la(&repbody->oa, &fti->fti_attr,
1420 OFD_VALID_FLAGS | LA_UID | LA_GID);
1422 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1426 ofd_object_put(tsi->tsi_env, fo);
1429 /* we do not call this before to avoid lu_object_find() in
1430 * ->lvbo_update() holding another reference on the object.
1431 * otherwise concurrent destroy can make the object unavailable
1432 * for 2nd lu_object_find() waiting for the first reference
1433 * to go... deadlock! */
1434 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1435 &tsi->tsi_resid, LDLM_EXTENT, 0);
1437 ldlm_res_lvbo_update(res, NULL, 0);
1438 ldlm_resource_putref(res);
1445 * Destroy OST orphans.
1447 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1448 * set then we must destroy possible orphaned objects.
1450 * \param[in] env execution environment
1451 * \param[in] exp OBD export
1452 * \param[in] ofd OFD device
1453 * \param[in] oa obdo structure for reply
1455 * \retval 0 if successful
1456 * \retval negative value on error
1458 static int ofd_orphans_destroy(const struct lu_env *env,
1459 struct obd_export *exp,
1460 struct ofd_device *ofd, struct obdo *oa)
1462 struct ofd_thread_info *info = ofd_info(env);
1463 struct lu_fid *fid = &info->fti_fid;
1464 struct ost_id *oi = &oa->o_oi;
1465 struct ofd_seq *oseq;
1466 u64 seq = ostid_seq(oi);
1467 u64 end_id = ostid_id(oi);
1475 oseq = ofd_seq_get(ofd, seq);
1477 CERROR("%s: Can not find seq for "DOSTID"\n",
1478 ofd_name(ofd), POSTID(oi));
1483 last = ofd_seq_last_oid(oseq);
1486 LASSERT(exp != NULL);
1487 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1489 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1492 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1493 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1495 while (oid > end_id) {
1496 rc = fid_set_id(fid, oid);
1497 if (unlikely(rc != 0))
1500 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1501 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1502 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1503 /* this is pretty fatal... */
1504 CEMERG("%s: error destroying precreated id "
1506 ofd_name(ofd), PFID(fid), rc);
1510 ofd_seq_last_oid_set(oseq, oid);
1511 /* update last_id on disk periodically so that if we
1512 * restart * we don't need to re-scan all of the just
1513 * deleted objects. */
1514 if ((oid & 511) == 0)
1515 ofd_seq_last_oid_write(env, ofd, oseq);
1519 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1520 ofd_name(ofd), seq, oid);
1524 ofd_seq_last_oid_set(oseq, oid);
1525 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1527 /* don't reuse orphan object, return last used objid */
1528 ostid_set_id(oi, last);
1535 ofd_seq_put(env, oseq);
1540 * OFD request handler for OST_CREATE RPC.
1542 * This is OFD-specific part of request handling. Its main purpose is to
1543 * create new data objects on OST, but it also used to destroy orphans.
1545 * \param[in] tsi target session environment for this request
1547 * \retval 0 if successful
1548 * \retval negative value on error
1550 static int ofd_create_hdl(struct tgt_session_info *tsi)
1552 struct ptlrpc_request *req = tgt_ses_req(tsi);
1553 struct ost_body *repbody;
1554 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1555 struct obdo *rep_oa;
1556 struct obd_export *exp = tsi->tsi_exp;
1557 struct ofd_device *ofd = ofd_exp(exp);
1558 u64 seq = ostid_seq(&oa->o_oi);
1559 u64 oid = ostid_id(&oa->o_oi);
1560 struct ofd_seq *oseq;
1567 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1570 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1571 if (repbody == NULL)
1574 down_read(&ofd->ofd_lastid_rwsem);
1575 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1576 * we may do that in the future.
1577 * Return -ENOSPC until the LAST_ID rebuilt. */
1578 if (unlikely(ofd->ofd_lastid_rebuilding))
1579 GOTO(out_sem, rc = -ENOSPC);
1581 rep_oa = &repbody->oa;
1582 rep_oa->o_oi = oa->o_oi;
1584 LASSERT(seq >= FID_SEQ_OST_MDT0);
1585 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1587 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1589 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1591 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1592 ofd_name(ofd), seq, PTR_ERR(oseq));
1593 GOTO(out_sem, rc = -EINVAL);
1596 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1597 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1598 if (!ofd_obd(ofd)->obd_recovering ||
1599 oid > ofd_seq_last_oid(oseq)) {
1600 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1601 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1602 ofd_seq_last_oid(oseq));
1603 GOTO(out_nolock, rc = -EINVAL);
1605 /* Do nothing here, we re-create objects during recovery
1606 * upon write replay, see ofd_preprw_write() */
1607 GOTO(out_nolock, rc = 0);
1609 /* former ofd_handle_precreate */
1610 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1611 (oa->o_flags & OBD_FL_DELORPHAN)) {
1612 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1614 /* destroy orphans */
1615 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1616 exp->exp_conn_cnt) {
1617 CERROR("%s: dropping old orphan cleanup request\n",
1619 GOTO(out_nolock, rc = 0);
1621 /* This causes inflight precreates to abort and drop lock */
1622 oseq->os_destroys_in_progress = 1;
1623 mutex_lock(&oseq->os_create_lock);
1624 if (!oseq->os_destroys_in_progress) {
1625 CERROR("%s:[%llu] destroys_in_progress already"
1626 " cleared\n", ofd_name(ofd), seq);
1627 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1630 diff = oid - ofd_seq_last_oid(oseq);
1631 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %d\n",
1632 ofd_seq_last_oid(oseq), diff);
1633 if (-diff > OST_MAX_PRECREATE) {
1634 /* Let MDS know that we are so far ahead. */
1635 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq) + 1);
1637 } else if (diff < 0) {
1638 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1640 oseq->os_destroys_in_progress = 0;
1642 /* XXX: Used by MDS for the first time! */
1643 oseq->os_destroys_in_progress = 0;
1646 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1647 ofd->ofd_lastid_gen)) {
1648 /* Keep the export ref so we can send the reply. */
1649 ofd_obd_disconnect(class_export_get(exp));
1650 GOTO(out_nolock, rc = -ENOTCONN);
1653 mutex_lock(&oseq->os_create_lock);
1654 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1655 exp->exp_conn_cnt) {
1656 CERROR("%s: dropping old precreate request\n",
1660 /* only precreate if seq is 0, IDIF or normal and also o_id
1661 * must be specfied */
1662 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1663 !fid_seq_is_idif(seq)) || oid == 0) {
1664 diff = 1; /* shouldn't we create this right now? */
1666 diff = oid - ofd_seq_last_oid(oseq);
1667 /* Do sync create if the seq is about to used up */
1668 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1669 if (unlikely(oid >= IDIF_MAX_OID - 1))
1671 } else if (fid_seq_is_norm(seq)) {
1673 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1676 CERROR("%s : invalid o_seq "DOSTID"\n",
1677 ofd_name(ofd), POSTID(&oa->o_oi));
1678 GOTO(out, rc = -EINVAL);
1683 CERROR("%s: invalid precreate request for "
1684 DOSTID", last_id %llu. "
1685 "Likely MDS last_id corruption\n",
1686 ofd_name(ofd), POSTID(&oa->o_oi),
1687 ofd_seq_last_oid(oseq));
1688 GOTO(out, rc = -EINVAL);
1693 cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
1698 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1699 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1700 /* don't enforce grant during orphan recovery */
1701 granted = ofd_grant_create(tsi->tsi_env,
1702 ofd_obd(ofd)->obd_self_export,
1707 CDEBUG(D_HA, "%s: failed to acquire grant "
1708 "space for precreate (%d): rc = %d\n",
1709 ofd_name(ofd), diff, rc);
1714 /* This can happen if a new OST is formatted and installed
1715 * in place of an old one at the same index. Instead of
1716 * precreating potentially millions of deleted old objects
1717 * (possibly filling the OST), only precreate the last batch.
1718 * LFSCK will eventually clean up any orphans. LU-14 */
1719 if (diff > 5 * OST_MAX_PRECREATE) {
1720 diff = OST_MAX_PRECREATE / 2;
1721 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1722 "OST replaced or reformatted: "
1723 "LFSCK will clean up",
1726 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1727 "%u larger than the LAST_ID "DOSTID", only "
1728 "precreating the last %u objects.\n",
1729 ofd_name(ofd), POSTID(&oa->o_oi),
1730 5 * OST_MAX_PRECREATE,
1731 POSTID(&oseq->os_oi), diff);
1732 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1736 next_id = ofd_seq_last_oid(oseq) + 1;
1737 count = ofd_precreate_batch(ofd, diff);
1739 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1740 " at %llu\n", ofd_name(ofd),
1741 count, seq, next_id);
1743 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1744 && cfs_time_after(jiffies, enough_time)) {
1745 CDEBUG(D_HA, "%s: Slow creates, %d/%d objects"
1746 " created at a rate of %d/s\n",
1747 ofd_name(ofd), created, diff + created,
1748 created / DISK_TIMEOUT);
1752 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1753 oseq, count, sync_trans);
1757 } else if (rc < 0) {
1763 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1764 LCONSOLE_WARN("%s: can't create the same count of"
1765 " objects when replaying the request"
1766 " (diff is %d). see LU-4621\n",
1767 ofd_name(ofd), diff);
1770 /* some objects got created, we can return
1771 * them, even if last creation failed */
1774 CERROR("%s: unable to precreate: rc = %d\n",
1777 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1778 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1779 ofd_grant_commit(ofd_obd(ofd)->obd_self_export, granted,
1784 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1787 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1790 mutex_unlock(&oseq->os_create_lock);
1793 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1794 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1795 struct lu_fid *fid = &info->fti_fid;
1797 /* For compatible purpose, it needs to convert back to
1798 * OST ID before put it on wire. */
1799 *fid = rep_oa->o_oi.oi_fid;
1800 fid_to_ostid(fid, &rep_oa->o_oi);
1802 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1804 ofd_seq_put(tsi->tsi_env, oseq);
1807 up_read(&ofd->ofd_lastid_rwsem);
1812 * OFD request handler for OST_DESTROY RPC.
1814 * This is OFD-specific part of request handling. It destroys data objects
1815 * related to destroyed object on MDT.
1817 * \param[in] tsi target session environment for this request
1819 * \retval 0 if successful
1820 * \retval negative value on error
1822 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1824 const struct ost_body *body = tsi->tsi_ost_body;
1825 struct ost_body *repbody;
1826 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1827 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1828 struct lu_fid *fid = &fti->fti_fid;
1835 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1838 /* This is old case for clients before Lustre 2.4 */
1839 /* If there's a DLM request, cancel the locks mentioned in it */
1840 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1842 struct ldlm_request *dlm;
1844 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1847 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1850 *fid = body->oa.o_oi.oi_fid;
1851 oid = ostid_id(&body->oa.o_oi);
1854 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1856 /* check that o_misc makes sense */
1857 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1858 count = body->oa.o_misc;
1860 count = 1; /* default case - single destroy */
1862 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1863 POSTID(&body->oa.o_oi), count);
1868 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1869 if (lrc == -ENOENT) {
1871 "%s: destroying non-existent object "DFID"\n",
1872 ofd_name(ofd), PFID(fid));
1873 /* rewrite rc with -ENOENT only if it is 0 */
1876 } else if (lrc != 0) {
1877 CERROR("%s: error destroying object "DFID": %d\n",
1878 ofd_name(ofd), PFID(fid), lrc);
1884 lrc = fid_set_id(fid, oid);
1885 if (unlikely(lrc != 0 && count > 0))
1886 GOTO(out, rc = lrc);
1889 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1895 fid_to_ostid(fid, &repbody->oa.o_oi);
1900 * OFD request handler for OST_STATFS RPC.
1902 * This function gets statfs data from storage as part of request
1905 * \param[in] tsi target session environment for this request
1907 * \retval 0 if successful
1908 * \retval negative value on error
1910 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1912 struct obd_statfs *osfs;
1917 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1919 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1920 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
1922 CERROR("%s: statfs failed: rc = %d\n",
1923 tgt_name(tsi->tsi_tgt), rc);
1925 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1928 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1935 * OFD request handler for OST_SYNC RPC.
1937 * Sync object data or all filesystem data to the disk and pack the
1940 * \param[in] tsi target session environment for this request
1942 * \retval 0 if successful
1943 * \retval negative value on error
1945 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1947 struct ost_body *body = tsi->tsi_ost_body;
1948 struct ost_body *repbody;
1949 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1950 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1951 struct ofd_object *fo = NULL;
1956 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1958 /* if no objid is specified, it means "sync whole filesystem" */
1959 if (!fid_is_zero(&tsi->tsi_fid)) {
1960 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1962 RETURN(PTR_ERR(fo));
1965 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1966 fo != NULL ? ofd_object_child(fo) : NULL,
1967 repbody->oa.o_size, repbody->oa.o_blocks);
1971 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1976 repbody->oa.o_oi = body->oa.o_oi;
1977 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1979 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1981 obdo_from_la(&repbody->oa, &fti->fti_attr,
1984 /* don't return rc from getattr */
1989 ofd_object_put(tsi->tsi_env, fo);
1994 * OFD request handler for OST_PUNCH RPC.
1996 * This is part of request processing. Validate request fields,
1997 * punch (truncate) the given OFD object and pack reply.
1999 * \param[in] tsi target session environment for this request
2001 * \retval 0 if successful
2002 * \retval negative value on error
2004 static int ofd_punch_hdl(struct tgt_session_info *tsi)
2006 const struct obdo *oa = &tsi->tsi_ost_body->oa;
2007 struct ost_body *repbody;
2008 struct ofd_thread_info *info = tsi2ofd_info(tsi);
2009 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
2010 struct ldlm_resource *res;
2011 struct ofd_object *fo;
2012 struct filter_fid *ff = NULL;
2014 struct lustre_handle lh = { 0, };
2021 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
2023 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
2024 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
2026 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2027 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2028 RETURN(err_serious(-EPROTO));
2030 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2031 if (repbody == NULL)
2032 RETURN(err_serious(-ENOMEM));
2034 /* punch start,end are passed in o_size,o_blocks throught wire */
2038 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2041 /* standard truncate optimization: if file body is completely
2042 * destroyed, don't send data back to the server. */
2044 flags |= LDLM_FL_AST_DISCARD_DATA;
2046 repbody->oa.o_oi = oa->o_oi;
2047 repbody->oa.o_valid = OBD_MD_FLID;
2049 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2050 oa->o_flags & OBD_FL_SRVLOCK;
2053 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
2059 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2060 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2061 oa->o_valid, start, end);
2063 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2066 GOTO(out, rc = PTR_ERR(fo));
2068 la_from_obdo(&info->fti_attr, oa,
2069 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2070 info->fti_attr.la_size = start;
2071 info->fti_attr.la_valid |= LA_SIZE;
2073 if (oa->o_valid & OBD_MD_FLFID) {
2074 ff = &info->fti_mds_fid;
2075 ofd_prepare_fidea(ff, oa);
2078 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2079 ff, (struct obdo *)oa);
2083 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2087 ofd_object_put(tsi->tsi_env, fo);
2090 tgt_extent_unlock(&lh, LCK_PW);
2092 /* we do not call this before to avoid lu_object_find() in
2093 * ->lvbo_update() holding another reference on the object.
2094 * otherwise concurrent destroy can make the object unavailable
2095 * for 2nd lu_object_find() waiting for the first reference
2096 * to go... deadlock! */
2097 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2100 ldlm_res_lvbo_update(res, NULL, 0);
2101 ldlm_resource_putref(res);
2107 static int ofd_ladvise_prefetch(const struct lu_env *env,
2108 struct ofd_object *fo,
2109 struct niobuf_local *lnb,
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;
2121 ofd_read_lock(env, fo);
2122 if (!ofd_object_exists(fo))
2123 GOTO(out_unlock, rc = -ENOENT);
2125 rc = ofd_attr_get(env, fo, &info->fti_attr);
2127 GOTO(out_unlock, rc);
2129 if (end > info->fti_attr.la_size)
2130 end = info->fti_attr.la_size;
2133 GOTO(out_unlock, rc);
2135 /* We need page aligned offset and length */
2136 start_index = start >> PAGE_SHIFT;
2137 end_index = (end - 1) >> PAGE_SHIFT;
2138 pages = end_index - start_index + 1;
2140 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2141 PTLRPC_MAX_BRW_PAGES;
2142 rnb.rnb_offset = start_index << PAGE_SHIFT;
2143 rnb.rnb_len = nr_local << PAGE_SHIFT;
2144 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, 0);
2145 if (unlikely(rc < 0))
2148 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2149 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2152 start_index += nr_local;
2157 ofd_read_unlock(env, fo);
2162 * OFD request handler for OST_LADVISE RPC.
2164 * Tune cache or perfetch policies according to advices.
2166 * \param[in] tsi target session environment for this request
2168 * \retval 0 if successful
2169 * \retval negative errno on error
2171 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2173 struct ptlrpc_request *req = tgt_ses_req(tsi);
2174 struct obd_export *exp = tsi->tsi_exp;
2175 struct ofd_device *ofd = ofd_exp(exp);
2176 struct ost_body *body, *repbody;
2177 struct ofd_thread_info *info;
2178 struct ofd_object *fo;
2179 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2180 const struct lu_env *env = svc_thread->t_env;
2181 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2183 struct lu_ladvise *ladvise;
2185 struct ladvise_hdr *ladvise_hdr;
2186 struct obd_ioobj ioo;
2187 struct lustre_handle lockh = { 0 };
2190 struct dt_object *dob;
2195 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2196 body = tsi->tsi_ost_body;
2198 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2199 RETURN(err_serious(-EPROTO));
2201 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2202 &RMF_OST_LADVISE_HDR);
2203 if (ladvise_hdr == NULL)
2204 RETURN(err_serious(-EPROTO));
2206 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2207 ladvise_hdr->lah_count < 1)
2208 RETURN(err_serious(-EPROTO));
2210 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2211 RETURN(err_serious(-EPROTO));
2213 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2214 if (ladvise == NULL)
2215 RETURN(err_serious(-EPROTO));
2217 num_advise = req_capsule_get_size(&req->rq_pill,
2218 &RMF_OST_LADVISE, RCL_CLIENT) /
2220 if (num_advise < ladvise_hdr->lah_count)
2221 RETURN(err_serious(-EPROTO));
2223 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2224 repbody->oa = body->oa;
2226 info = ofd_info_init(env, exp);
2228 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2229 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2233 fo = ofd_object_find(env, ofd, &info->fti_fid);
2238 LASSERT(fo != NULL);
2239 dob = ofd_object_child(fo);
2241 for (i = 0; i < num_advise; i++, ladvise++) {
2242 start = ladvise->lla_start;
2243 end = ladvise->lla_end;
2245 rc = err_serious(-EPROTO);
2249 /* Handle different advice types */
2250 switch (ladvise->lla_advice) {
2254 case LU_LADVISE_WILLREAD:
2258 ioo.ioo_oid = body->oa.o_oi;
2260 rc = tgt_extent_lock(exp->exp_obd->obd_namespace,
2261 &tsi->tsi_resid, start, end - 1,
2262 &lockh, LCK_PR, &flags);
2266 req->rq_status = ofd_ladvise_prefetch(env, fo,
2269 tgt_extent_unlock(&lockh, LCK_PR);
2271 case LU_LADVISE_DONTNEED:
2272 rc = dt_ladvise(env, dob, ladvise->lla_start,
2273 ladvise->lla_end, LU_LADVISE_DONTNEED);
2280 ofd_object_put(env, fo);
2281 req->rq_status = rc;
2286 * OFD request handler for OST_QUOTACTL RPC.
2288 * This is part of request processing to validate incoming request fields,
2289 * get the requested data from OSD and pack reply.
2291 * \param[in] tsi target session environment for this request
2293 * \retval 0 if successful
2294 * \retval negative value on error
2296 static int ofd_quotactl(struct tgt_session_info *tsi)
2298 struct obd_quotactl *oqctl, *repoqc;
2299 struct lu_nodemap *nodemap;
2305 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2307 RETURN(err_serious(-EPROTO));
2309 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2311 RETURN(err_serious(-ENOMEM));
2315 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2316 if (IS_ERR(nodemap))
2317 RETURN(PTR_ERR(nodemap));
2320 if (oqctl->qc_type == USRQUOTA)
2321 id = nodemap_map_id(nodemap, NODEMAP_UID,
2322 NODEMAP_CLIENT_TO_FS,
2324 else if (oqctl->qc_type == GRPQUOTA)
2325 id = nodemap_map_id(nodemap, NODEMAP_GID,
2326 NODEMAP_CLIENT_TO_FS,
2329 nodemap_putref(nodemap);
2331 if (repoqc->qc_id != id)
2332 swap(repoqc->qc_id, id);
2334 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2336 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2339 if (repoqc->qc_id != id)
2340 swap(repoqc->qc_id, id);
2346 * Calculate the amount of time for lock prolongation.
2348 * This is helper for ofd_prolong_extent_locks() function to get
2349 * the timeout extra time.
2351 * \param[in] req current request
2353 * \retval amount of time to extend the timeout with
2355 static inline int prolong_timeout(struct ptlrpc_request *req)
2357 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2361 return obd_timeout / 2;
2363 req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec;
2364 return max_t(time_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
2369 * Prolong lock timeout for the given extent.
2371 * This function finds all locks related with incoming request and
2372 * prolongs their timeout.
2374 * If a client is holding a lock for a long time while it sends
2375 * read or write RPCs to the OST for the object under this lock,
2376 * then we don't want the OST to evict the client. Otherwise,
2377 * if the network or disk is very busy then the client may not
2378 * be able to make any progress to clear out dirty pages under
2379 * the lock and the application will fail.
2381 * Every time a Bulk Read/Write (BRW) request arrives for the object
2382 * covered by the lock, extend the timeout on that lock. The RPC should
2383 * contain a lock handle for the lock it is using, but this
2384 * isn't handled correctly by all client versions, and the
2385 * request may cover multiple locks.
2387 * \param[in] tsi target session environment for this request
2388 * \param[in] data struct of data to prolong locks
2391 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2392 struct ldlm_prolong_args *data)
2394 struct obdo *oa = &tsi->tsi_ost_body->oa;
2395 struct ldlm_lock *lock;
2399 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2400 data->lpa_export = tsi->tsi_exp;
2401 data->lpa_resid = tsi->tsi_resid;
2403 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2404 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2405 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2406 data->lpa_extent.end);
2408 if (oa->o_valid & OBD_MD_FLHANDLE) {
2409 /* mostly a request should be covered by only one lock, try
2411 lock = ldlm_handle2lock(&oa->o_handle);
2413 /* Fast path to check if the lock covers the whole IO
2414 * region exclusively. */
2415 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2416 &data->lpa_extent)) {
2418 LASSERT(lock->l_export == data->lpa_export);
2419 ldlm_lock_prolong_one(lock, data);
2420 LDLM_LOCK_PUT(lock);
2423 lock->l_last_used = cfs_time_current();
2424 LDLM_LOCK_PUT(lock);
2428 ldlm_resource_prolong(data);
2433 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2435 * Determine if \a lock and the lock from request \a req are equivalent
2436 * by comparing their resource names, modes, and extents.
2438 * It is used to give priority to read and write RPCs being done
2439 * under this lock so that the client can drop the contended
2440 * lock more quickly and let other clients use it. This improves
2441 * overall performance in the case where the first client gets a
2442 * very large lock extent that prevents other clients from
2443 * submitting their writes.
2445 * \param[in] req ptlrpc_request being processed
2446 * \param[in] lock contended lock to match
2448 * \retval 1 if lock is matched
2449 * \retval 0 otherwise
2451 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2452 struct ldlm_lock *lock)
2454 struct niobuf_remote *rnb;
2455 struct obd_ioobj *ioo;
2456 enum ldlm_mode mode;
2457 struct ldlm_extent ext;
2458 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2462 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2463 LASSERT(ioo != NULL);
2465 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2466 LASSERT(rnb != NULL);
2468 ext.start = rnb->rnb_offset;
2469 rnb += ioo->ioo_bufcnt - 1;
2470 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2472 LASSERT(lock->l_resource != NULL);
2473 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2476 /* a bulk write can only hold a reference on a PW extent lock
2479 mode = LCK_PW | LCK_GROUP;
2480 if (opc == OST_READ)
2481 /* whereas a bulk read can be protected by either a PR or PW
2485 if (!(lock->l_granted_mode & mode))
2488 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2492 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2494 * Check for whether the given PTLRPC request (\a req) is blocking
2495 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2498 * \param[in] req the incoming request
2500 * \retval 1 if \a req is blocking an LDLM lock cancel
2501 * \retval 0 if it is not
2502 * \retval -ESTALE if lock is not found
2504 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2506 struct tgt_session_info *tsi;
2507 struct obd_ioobj *ioo;
2508 struct niobuf_remote *rnb;
2510 struct ldlm_prolong_args pa = { 0 };
2514 /* Don't use tgt_ses_info() to get session info, because lock_match()
2515 * can be called while request has no processing thread yet. */
2516 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2519 * Use LASSERT below because malformed RPCs should have
2520 * been filtered out in tgt_hpreq_handler().
2522 opc = lustre_msg_get_opc(req->rq_reqmsg);
2523 LASSERT(opc == OST_READ || opc == OST_WRITE);
2525 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2526 LASSERT(ioo != NULL);
2528 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2529 LASSERT(rnb != NULL);
2530 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2532 pa.lpa_mode = LCK_PW | LCK_GROUP;
2533 if (opc == OST_READ)
2534 pa.lpa_mode |= LCK_PR;
2536 pa.lpa_extent.start = rnb->rnb_offset;
2537 rnb += ioo->ioo_bufcnt - 1;
2538 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2540 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
2541 " (%llu->%llu)\n", tgt_name(tsi->tsi_tgt),
2542 current->comm, PFID(&tsi->tsi_fid), pa.lpa_extent.start,
2545 ofd_prolong_extent_locks(tsi, &pa);
2547 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2548 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2550 if (pa.lpa_blocks_cnt > 0)
2553 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2557 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2559 * Called after the request has been handled. It refreshes lock timeout again
2560 * so that client has more time to send lock cancel RPC.
2562 * \param[in] req request which is being processed.
2564 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2566 ofd_rw_hpreq_check(req);
2570 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2572 * This function checks if the given lock is the same by its resname, mode
2573 * and extent as one taken from the request.
2574 * It is used to give priority to punch/truncate RPCs that might lead to
2575 * the fastest release of that lock when a lock is contended.
2577 * \param[in] req ptlrpc_request being processed
2578 * \param[in] lock contended lock to match
2580 * \retval 1 if lock is matched
2581 * \retval 0 otherwise
2583 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2584 struct ldlm_lock *lock)
2586 struct tgt_session_info *tsi;
2588 struct ldlm_extent ext;
2592 /* Don't use tgt_ses_info() to get session info, because lock_match()
2593 * can be called while request has no processing thread yet. */
2594 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2597 * Use LASSERT below because malformed RPCs should have
2598 * been filtered out in tgt_hpreq_handler().
2600 LASSERT(tsi->tsi_ost_body != NULL);
2601 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2602 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2605 oa = &tsi->tsi_ost_body->oa;
2606 ext.start = oa->o_size;
2607 ext.end = oa->o_blocks;
2609 LASSERT(lock->l_resource != NULL);
2610 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2613 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2616 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2620 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2622 * High-priority queue request check for whether the given punch request
2623 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2624 * covered by an LDLM lock.
2628 * \param[in] req the incoming request
2630 * \retval 1 if \a req is blocking an LDLM lock cancel
2631 * \retval 0 if it is not
2632 * \retval -ESTALE if lock is not found
2634 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2636 struct tgt_session_info *tsi;
2638 struct ldlm_prolong_args pa = { 0 };
2642 /* Don't use tgt_ses_info() to get session info, because lock_match()
2643 * can be called while request has no processing thread yet. */
2644 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2645 LASSERT(tsi != NULL);
2646 oa = &tsi->tsi_ost_body->oa;
2648 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2649 oa->o_flags & OBD_FL_SRVLOCK));
2651 pa.lpa_mode = LCK_PW | LCK_GROUP;
2652 pa.lpa_extent.start = oa->o_size;
2653 pa.lpa_extent.end = oa->o_blocks;
2656 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2657 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2658 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2660 ofd_prolong_extent_locks(tsi, &pa);
2662 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2663 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2665 if (pa.lpa_blocks_cnt > 0)
2668 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2672 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2674 * Called after the request has been handled. It refreshes lock timeout again
2675 * so that client has more time to send lock cancel RPC.
2677 * \param[in] req request which is being processed.
2679 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2681 ofd_punch_hpreq_check(req);
2684 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2685 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2686 .hpreq_check = ofd_rw_hpreq_check,
2687 .hpreq_fini = ofd_rw_hpreq_fini
2690 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2691 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2692 .hpreq_check = ofd_punch_hpreq_check,
2693 .hpreq_fini = ofd_punch_hpreq_fini
2697 * Assign high priority operations to an IO request.
2699 * Check if the incoming request is a candidate for
2700 * high-priority processing. If it is, assign it a high
2701 * priority operations table.
2703 * \param[in] tsi target session environment for this request
2705 static void ofd_hp_brw(struct tgt_session_info *tsi)
2707 struct niobuf_remote *rnb;
2708 struct obd_ioobj *ioo;
2712 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2713 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2714 if (ioo->ioo_bufcnt > 0) {
2715 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2716 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2718 /* no high priority if server lock is needed */
2719 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2720 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2724 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2728 * Assign high priority operations to an punch request.
2730 * Check if the incoming request is a candidate for
2731 * high-priority processing. If it is, assign it a high
2732 * priority operations table.
2734 * \param[in] tsi target session environment for this request
2736 static void ofd_hp_punch(struct tgt_session_info *tsi)
2738 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2739 /* no high-priority if server lock is needed */
2740 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2741 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2742 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2743 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2745 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2748 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2749 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2750 #define OST_BRW_READ OST_READ
2751 #define OST_BRW_WRITE OST_WRITE
2754 * Table of OFD-specific request handlers
2756 * This table contains all opcodes accepted by OFD and
2757 * specifies handlers for them. The tgt_request_handler()
2758 * uses such table from each target to process incoming
2761 static struct tgt_handler ofd_tgt_handlers[] = {
2762 TGT_RPC_HANDLER(OST_FIRST_OPC,
2763 0, OST_CONNECT, tgt_connect,
2764 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2765 TGT_RPC_HANDLER(OST_FIRST_OPC,
2766 0, OST_DISCONNECT, tgt_disconnect,
2767 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2768 TGT_RPC_HANDLER(OST_FIRST_OPC,
2769 0, OST_SET_INFO, ofd_set_info_hdl,
2770 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2771 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2772 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
2773 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2774 OST_SETATTR, ofd_setattr_hdl),
2775 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2776 OST_CREATE, ofd_create_hdl),
2777 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2778 OST_DESTROY, ofd_destroy_hdl),
2779 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
2780 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2781 OST_BRW_READ, tgt_brw_read,
2783 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2784 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
2786 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2787 OST_PUNCH, ofd_punch_hdl,
2789 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
2790 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
2791 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_LADVISE, ofd_ladvise_hdl),
2794 static struct tgt_opc_slice ofd_common_slice[] = {
2796 .tos_opc_start = OST_FIRST_OPC,
2797 .tos_opc_end = OST_LAST_OPC,
2798 .tos_hs = ofd_tgt_handlers
2801 .tos_opc_start = OBD_FIRST_OPC,
2802 .tos_opc_end = OBD_LAST_OPC,
2803 .tos_hs = tgt_obd_handlers
2806 .tos_opc_start = LDLM_FIRST_OPC,
2807 .tos_opc_end = LDLM_LAST_OPC,
2808 .tos_hs = tgt_dlm_handlers
2811 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2812 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2813 .tos_hs = tgt_out_handlers
2816 .tos_opc_start = SEQ_FIRST_OPC,
2817 .tos_opc_end = SEQ_LAST_OPC,
2818 .tos_hs = seq_handlers
2821 .tos_opc_start = LFSCK_FIRST_OPC,
2822 .tos_opc_end = LFSCK_LAST_OPC,
2823 .tos_hs = tgt_lfsck_handlers
2826 .tos_opc_start = SEC_FIRST_OPC,
2827 .tos_opc_end = SEC_LAST_OPC,
2828 .tos_hs = tgt_sec_ctx_handlers
2835 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2836 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2839 * Implementation of lu_context_key::lct_key_exit.
2841 * Optional method called on lu_context_exit() for all allocated
2843 * It is used in OFD to sanitize context values which may be re-used
2844 * during another request processing by the same thread.
2846 * \param[in] ctx execution context
2847 * \param[in] key context key
2848 * \param[in] data ofd_thread_info
2850 static void ofd_key_exit(const struct lu_context *ctx,
2851 struct lu_context_key *key, void *data)
2853 struct ofd_thread_info *info = data;
2855 info->fti_env = NULL;
2856 info->fti_exp = NULL;
2859 info->fti_pre_version = 0;
2862 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2865 struct lu_context_key ofd_thread_key = {
2866 .lct_tags = LCT_DT_THREAD,
2867 .lct_init = ofd_key_init,
2868 .lct_fini = ofd_key_fini,
2869 .lct_exit = ofd_key_exit
2873 * Initialize OFD device according to parameters in the config log \a cfg.
2875 * This is the main starting point of OFD initialization. It fills all OFD
2876 * parameters with their initial values and calls other initializing functions
2877 * to set up all OFD subsystems.
2879 * \param[in] env execution environment
2880 * \param[in] m OFD device
2881 * \param[in] ldt LU device type of OFD
2882 * \param[in] cfg configuration log
2884 * \retval 0 if successful
2885 * \retval negative value on error
2887 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2888 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2890 const char *dev = lustre_cfg_string(cfg, 0);
2891 struct ofd_thread_info *info = NULL;
2892 struct obd_device *obd;
2893 struct obd_statfs *osfs;
2895 struct nm_config_file *nodemap_config;
2900 obd = class_name2obd(dev);
2902 CERROR("Cannot find obd with name %s\n", dev);
2906 rc = lu_env_refill((struct lu_env *)env);
2910 obd->u.obt.obt_magic = OBT_MAGIC;
2912 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2913 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2915 spin_lock_init(&m->ofd_flags_lock);
2916 m->ofd_raid_degraded = 0;
2917 m->ofd_syncjournal = 0;
2919 m->ofd_grant_compat_disable = 0;
2920 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2923 spin_lock_init(&m->ofd_osfs_lock);
2924 m->ofd_osfs_age = cfs_time_shift_64(-1000);
2925 m->ofd_osfs_unstable = 0;
2926 m->ofd_statfs_inflight = 0;
2927 m->ofd_osfs_inflight = 0;
2930 spin_lock_init(&m->ofd_grant_lock);
2931 m->ofd_tot_dirty = 0;
2932 m->ofd_tot_granted = 0;
2933 m->ofd_tot_pending = 0;
2934 m->ofd_seq_count = 0;
2935 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2936 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2937 spin_lock_init(&m->ofd_inconsistency_lock);
2939 spin_lock_init(&m->ofd_batch_lock);
2940 init_rwsem(&m->ofd_lastid_rwsem);
2942 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2943 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2944 /* set this lu_device to obd, because error handling need it */
2945 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2947 rc = ofd_procfs_init(m);
2949 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2953 /* No connection accepted until configurations will finish */
2954 spin_lock(&obd->obd_dev_lock);
2955 obd->obd_no_conn = 1;
2956 spin_unlock(&obd->obd_dev_lock);
2957 obd->obd_replayable = 1;
2958 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2959 char *str = lustre_cfg_string(cfg, 4);
2961 if (strchr(str, 'n')) {
2962 CWARN("%s: recovery disabled\n", obd->obd_name);
2963 obd->obd_replayable = 0;
2967 info = ofd_info_init(env, NULL);
2969 GOTO(err_fini_proc, rc = -EFAULT);
2971 rc = ofd_stack_init(env, m, cfg);
2973 CERROR("Can't init device stack, rc %d\n", rc);
2974 GOTO(err_fini_proc, rc);
2977 ofd_procfs_add_brw_stats_symlink(m);
2979 /* populate cached statfs data */
2980 osfs = &ofd_info(env)->fti_u.osfs;
2981 rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
2983 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
2984 GOTO(err_fini_stack, rc);
2986 if (!is_power_of_2(osfs->os_bsize)) {
2987 CERROR("%s: blocksize (%d) is not a power of 2\n",
2988 obd->obd_name, osfs->os_bsize);
2989 GOTO(err_fini_stack, rc = -EPROTO);
2991 m->ofd_blockbits = fls(osfs->os_bsize) - 1;
2993 if (ONE_MB_BRW_SIZE < (1U << m->ofd_blockbits))
2994 m->ofd_brw_size = 1U << m->ofd_blockbits;
2996 m->ofd_brw_size = ONE_MB_BRW_SIZE;
2998 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2999 if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
3000 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
3002 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
3003 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
3004 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
3005 LDLM_NAMESPACE_SERVER,
3006 LDLM_NAMESPACE_GREEDY,
3008 if (m->ofd_namespace == NULL)
3009 GOTO(err_fini_stack, rc = -ENOMEM);
3010 /* set obd_namespace for compatibility with old code */
3011 obd->obd_namespace = m->ofd_namespace;
3012 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
3013 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
3014 m->ofd_namespace->ns_lvbp = m;
3016 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3017 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
3019 dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
3021 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
3022 OBD_FAIL_OST_ALL_REQUEST_NET,
3023 OBD_FAIL_OST_ALL_REPLY_NET);
3025 GOTO(err_free_ns, rc);
3027 rc = ofd_fs_setup(env, m, obd);
3029 GOTO(err_fini_lut, rc);
3031 fid.f_seq = FID_SEQ_LOCAL_NAME;
3034 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
3037 GOTO(err_fini_fs, rc);
3039 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
3041 if (IS_ERR(nodemap_config))
3042 GOTO(err_fini_los, rc = PTR_ERR(nodemap_config));
3044 obd->u.obt.obt_nodemap_config_file = nodemap_config;
3046 rc = ofd_start_inconsistency_verification_thread(m);
3048 GOTO(err_fini_nm, rc);
3050 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
3055 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3056 obd->u.obt.obt_nodemap_config_file = NULL;
3058 local_oid_storage_fini(env, m->ofd_los);
3061 ofd_fs_cleanup(env, m);
3063 tgt_fini(env, &m->ofd_lut);
3065 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3066 obd->obd_namespace = m->ofd_namespace = NULL;
3068 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3075 * Stop the OFD device
3077 * This function stops the OFD device and all its subsystems.
3078 * This is the end of OFD lifecycle.
3080 * \param[in] env execution environment
3081 * \param[in] m OFD device
3083 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3085 struct obd_device *obd = ofd_obd(m);
3086 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3087 struct lfsck_stop stop;
3089 stop.ls_status = LS_PAUSED;
3091 lfsck_stop(env, m->ofd_osd, &stop);
3092 target_recovery_fini(obd);
3093 if (m->ofd_namespace != NULL)
3094 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3095 d->ld_obd->obd_force);
3097 obd_exports_barrier(obd);
3098 obd_zombie_barrier();
3100 tgt_fini(env, &m->ofd_lut);
3101 ofd_stop_inconsistency_verification_thread(m);
3102 lfsck_degister(env, m->ofd_osd);
3103 ofd_fs_cleanup(env, m);
3104 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3105 obd->u.obt.obt_nodemap_config_file = NULL;
3107 if (m->ofd_los != NULL) {
3108 local_oid_storage_fini(env, m->ofd_los);
3112 if (m->ofd_namespace != NULL) {
3113 ldlm_namespace_free_post(m->ofd_namespace);
3114 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3117 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3119 LASSERT(atomic_read(&d->ld_ref) == 0);
3120 server_put_mount(obd->obd_name, true);
3125 * Implementation of lu_device_type_operations::ldto_device_fini.
3127 * Finalize device. Dual to ofd_device_init(). It is called from
3128 * obd_precleanup() and stops the current device.
3130 * \param[in] env execution environment
3131 * \param[in] d LU device of OFD
3135 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3136 struct lu_device *d)
3139 ofd_fini(env, ofd_dev(d));
3144 * Implementation of lu_device_type_operations::ldto_device_free.
3146 * Free OFD device. Dual to ofd_device_alloc().
3148 * \param[in] env execution environment
3149 * \param[in] d LU device of OFD
3153 static struct lu_device *ofd_device_free(const struct lu_env *env,
3154 struct lu_device *d)
3156 struct ofd_device *m = ofd_dev(d);
3158 dt_device_fini(&m->ofd_dt_dev);
3164 * Implementation of lu_device_type_operations::ldto_device_alloc.
3166 * This function allocates the new OFD device. It is called from
3167 * obd_setup() if OBD device had lu_device_type defined.
3169 * \param[in] env execution environment
3170 * \param[in] t lu_device_type of OFD device
3171 * \param[in] cfg configuration log
3173 * \retval pointer to the lu_device of just allocated OFD
3174 * \retval ERR_PTR of return value on error
3176 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3177 struct lu_device_type *t,
3178 struct lustre_cfg *cfg)
3180 struct ofd_device *m;
3181 struct lu_device *l;
3186 return ERR_PTR(-ENOMEM);
3188 l = &m->ofd_dt_dev.dd_lu_dev;
3189 dt_device_init(&m->ofd_dt_dev, t);
3190 rc = ofd_init0(env, m, t, cfg);
3192 ofd_device_free(env, l);
3199 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3200 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3202 static struct lu_device_type_operations ofd_device_type_ops = {
3203 .ldto_init = ofd_type_init,
3204 .ldto_fini = ofd_type_fini,
3206 .ldto_start = ofd_type_start,
3207 .ldto_stop = ofd_type_stop,
3209 .ldto_device_alloc = ofd_device_alloc,
3210 .ldto_device_free = ofd_device_free,
3211 .ldto_device_fini = ofd_device_fini
3214 static struct lu_device_type ofd_device_type = {
3215 .ldt_tags = LU_DEVICE_DT,
3216 .ldt_name = LUSTRE_OST_NAME,
3217 .ldt_ops = &ofd_device_type_ops,
3218 .ldt_ctx_tags = LCT_DT_THREAD
3222 * Initialize OFD module.
3224 * This function is called upon module loading. It registers OFD device type
3225 * and prepares all in-memory structures used by all OFD devices.
3227 * \retval 0 if successful
3228 * \retval negative value on error
3230 static int __init ofd_init(void)
3234 rc = lu_kmem_init(ofd_caches);
3238 rc = ofd_fmd_init();
3240 lu_kmem_fini(ofd_caches);
3244 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3245 LUSTRE_OST_NAME, &ofd_device_type);
3252 * This function is called upon OFD module unloading.
3253 * It frees all related structures and unregisters OFD device type.
3255 static void __exit ofd_exit(void)
3258 lu_kmem_fini(ofd_caches);
3259 class_unregister_type(LUSTRE_OST_NAME);
3262 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3263 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3264 MODULE_VERSION(LUSTRE_VERSION_STRING);
3265 MODULE_LICENSE("GPL");
3267 module_init(ofd_init);
3268 module_exit(ofd_exit);