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/lustre_param.h>
74 #include <lustre_fid.h>
75 #include <lustre_lfsck.h>
76 #include <lustre_dlm.h>
77 #include <lustre_quota.h>
78 #include <lustre_nodemap.h>
79 #include <lustre_log.h>
81 #include "ofd_internal.h"
83 /* Slab for OFD object allocation */
84 static struct kmem_cache *ofd_object_kmem;
86 static struct lu_kmem_descr ofd_caches[] = {
88 .ckd_cache = &ofd_object_kmem,
89 .ckd_name = "ofd_obj",
90 .ckd_size = sizeof(struct ofd_object)
98 * Connect OFD to the next device in the stack.
100 * This function is used for device stack configuration and links OFD
101 * device with bottom OSD device.
103 * \param[in] env execution environment
104 * \param[in] m OFD device
105 * \param[in] next name of next device in the stack
106 * \param[out] exp export to return
108 * \retval 0 and export in \a exp if successful
109 * \retval negative value on error
111 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
112 const char *next, struct obd_export **exp)
114 struct obd_connect_data *data = NULL;
115 struct obd_device *obd;
121 GOTO(out, rc = -ENOMEM);
123 obd = class_name2obd(next);
125 CERROR("%s: can't locate next device: %s\n",
127 GOTO(out, rc = -ENOTCONN);
130 data->ocd_connect_flags = OBD_CONNECT_VERSION;
131 data->ocd_version = LUSTRE_VERSION_CODE;
133 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
135 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
136 ofd_name(m), next, rc);
140 m->ofd_dt_dev.dd_lu_dev.ld_site =
141 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
142 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
143 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
144 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
153 * Initialize stack of devices.
155 * This function initializes OFD-OSD device stack to serve OST requests
157 * \param[in] env execution environment
158 * \param[in] m OFD device
159 * \param[in] cfg Lustre config for this server
161 * \retval 0 if successful
162 * \retval negative value on error
164 static int ofd_stack_init(const struct lu_env *env,
165 struct ofd_device *m, struct lustre_cfg *cfg)
167 const char *dev = lustre_cfg_string(cfg, 0);
169 struct ofd_thread_info *info = ofd_info(env);
170 struct lustre_mount_info *lmi;
171 struct lustre_mount_data *lmd;
177 lmi = server_get_mount(dev);
179 CERROR("Cannot get mount info for %s!\n", dev);
183 lmd = s2lsi(lmi->lmi_sb)->lsi_lmd;
184 if (lmd != NULL && lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
185 m->ofd_skip_lfsck = 1;
187 /* find bottom osd */
188 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
192 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
193 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
194 OBD_FREE(osdname, MTI_NAME_MAXLEN);
198 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
200 m->ofd_osd = lu2dt_dev(d);
202 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
203 "%s-osd", lustre_cfg_string(cfg, 0));
209 * Finalize the device stack OFD-OSD.
211 * This function cleans OFD-OSD device stack and
212 * disconnects OFD from the OSD.
214 * \param[in] env execution environment
215 * \param[in] m OFD device
216 * \param[in] top top device of stack
218 * \retval 0 if successful
219 * \retval negative value on error
221 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
222 struct lu_device *top)
224 struct obd_device *obd = ofd_obd(m);
225 struct lustre_cfg_bufs bufs;
226 struct lustre_cfg *lcfg;
231 lu_site_purge(env, top->ld_site, ~0);
232 /* process cleanup, pass mdt obd name to get obd umount flags */
233 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
238 lustre_cfg_bufs_set_string(&bufs, 1, flags);
239 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
242 lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs);
245 top->ld_ops->ldo_process_config(env, top, lcfg);
246 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
248 lu_site_purge(env, top->ld_site, ~0);
249 if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
250 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
251 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
254 LASSERT(m->ofd_osd_exp);
255 obd_disconnect(m->ofd_osd_exp);
260 /* For interoperability, see mdt_interop_param[]. */
261 static struct cfg_interop_param ofd_interop_param[] = {
262 { "ost.quota_type", NULL },
267 * Check if parameters are symlinks to the OSD.
269 * Some parameters were moved from ofd to osd and only their
270 * symlinks were kept in ofd by LU-3106. They are:
271 * -writehthrough_cache_enable
272 * -readcache_max_filesize
276 * Since they are not included by the static lprocfs var list, a pre-check
277 * is added for them to avoid "unknown param" errors. If they are matched
278 * in this check, they will be passed to the OSD directly.
280 * \param[in] param parameters to check
282 * \retval true if param is symlink to OSD param
285 static bool match_symlink_param(char *param)
290 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
291 sval = strchr(param, '=');
293 paramlen = sval - param;
294 if (strncmp(param, "writethrough_cache_enable",
296 strncmp(param, "readcache_max_filesize",
298 strncmp(param, "read_cache_enable",
300 strncmp(param, "brw_stats", paramlen) == 0)
309 * Process various configuration parameters.
311 * This function is used by MGS to process specific configurations and
312 * pass them through to the next device in server stack, i.e. the OSD.
314 * \param[in] env execution environment
315 * \param[in] d LU device of OFD
316 * \param[in] cfg parameters to process
318 * \retval 0 if successful
319 * \retval negative value on error
321 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
322 struct lustre_cfg *cfg)
324 struct ofd_device *m = ofd_dev(d);
325 struct dt_device *dt_next = m->ofd_osd;
326 struct lu_device *next = &dt_next->dd_lu_dev;
331 switch (cfg->lcfg_command) {
333 struct obd_device *obd = ofd_obd(m);
334 /* For interoperability */
335 struct cfg_interop_param *ptr = NULL;
336 struct lustre_cfg *old_cfg = NULL;
339 param = lustre_cfg_string(cfg, 1);
341 CERROR("param is empty\n");
346 ptr = class_find_old_param(param, ofd_interop_param);
348 if (ptr->new_param == NULL) {
350 CWARN("For interoperability, skip this %s."
351 " It is obsolete.\n", ptr->old_param);
355 CWARN("Found old param %s, changed it to %s.\n",
356 ptr->old_param, ptr->new_param);
359 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
366 if (match_symlink_param(param)) {
367 rc = next->ld_ops->ldo_process_config(env, next, cfg);
371 rc = class_process_proc_param(PARAM_OST, obd->obd_vars, cfg,
373 if (rc > 0 || rc == -ENOSYS) {
374 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
376 /* we don't understand; pass it on */
377 rc = next->ld_ops->ldo_process_config(env, next, cfg);
381 case LCFG_SPTLRPC_CONF: {
386 /* others are passed further */
387 rc = next->ld_ops->ldo_process_config(env, next, cfg);
394 * Implementation of lu_object_operations::loo_object_init for OFD
396 * Allocate just the next object (OSD) in stack.
398 * \param[in] env execution environment
399 * \param[in] o lu_object of OFD object
400 * \param[in] conf additional configuration parameters, not used here
402 * \retval 0 if successful
403 * \retval negative value on error
405 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
406 const struct lu_object_conf *conf)
408 struct ofd_device *d = ofd_dev(o->lo_dev);
409 struct lu_device *under;
410 struct lu_object *below;
415 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
416 PFID(lu_object_fid(o)));
418 under = &d->ofd_osd->dd_lu_dev;
419 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
421 lu_object_add(o, below);
429 * Implementation of lu_object_operations::loo_object_free.
431 * Finish OFD object lifecycle and free its memory.
433 * \param[in] env execution environment
434 * \param[in] o LU object of OFD object
436 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
438 struct ofd_object *of = ofd_obj(o);
439 struct lu_object_header *h;
444 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
445 PFID(lu_object_fid(o)));
448 lu_object_header_fini(h);
449 OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
454 * Implementation of lu_object_operations::loo_object_print.
456 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
457 * LU_OBJECT_DEBUG() for more details about the compound object printing.
459 * \param[in] env execution environment
460 * \param[in] cookie opaque data passed to the printer function
461 * \param[in] p printer function to use
462 * \param[in] o LU object of OFD object
464 * \retval 0 if successful
465 * \retval negative value on error
467 static int ofd_object_print(const struct lu_env *env, void *cookie,
468 lu_printer_t p, const struct lu_object *o)
470 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
473 static struct lu_object_operations ofd_obj_ops = {
474 .loo_object_init = ofd_object_init,
475 .loo_object_free = ofd_object_free,
476 .loo_object_print = ofd_object_print
480 * Implementation of lu_device_operations::lod_object_alloc.
482 * This function allocates OFD part of compound OFD-OSD object and
483 * initializes its header, because OFD is the top device in stack
485 * \param[in] env execution environment
486 * \param[in] hdr object header, NULL for OFD
487 * \param[in] d lu_device
489 * \retval allocated object if successful
490 * \retval NULL value on failed allocation
492 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
493 const struct lu_object_header *hdr,
496 struct ofd_object *of;
500 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
503 struct lu_object_header *h;
505 o = &of->ofo_obj.do_lu;
507 lu_object_header_init(h);
508 lu_object_init(o, h, d);
509 lu_object_add_top(h, o);
510 o->lo_ops = &ofd_obj_ops;
518 * Return the result of LFSCK run to the OFD.
520 * Notify OFD about result of LFSCK run. That may block the new object
521 * creation until problem is fixed by LFSCK.
523 * \param[in] env execution environment
524 * \param[in] data pointer to the OFD device
525 * \param[in] event LFSCK event type
527 * \retval 0 if successful
528 * \retval negative value on unknown event
530 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
531 enum lfsck_events event)
533 struct ofd_device *ofd = data;
534 struct obd_device *obd = ofd_obd(ofd);
537 case LE_LASTID_REBUILDING:
538 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
539 "on the device until the LAST_ID rebuilt successfully.\n",
541 down_write(&ofd->ofd_lastid_rwsem);
542 ofd->ofd_lastid_rebuilding = 1;
543 up_write(&ofd->ofd_lastid_rwsem);
545 case LE_LASTID_REBUILT: {
546 down_write(&ofd->ofd_lastid_rwsem);
547 ofd_seqs_free(env, ofd);
548 ofd->ofd_lastid_rebuilding = 0;
549 ofd->ofd_lastid_gen++;
550 up_write(&ofd->ofd_lastid_rwsem);
551 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
556 CERROR("%s: unknown lfsck event: rc = %d\n",
557 ofd_name(ofd), event);
565 * Implementation of lu_device_operations::ldo_prepare.
567 * This method is called after layer has been initialized and before it starts
568 * serving user requests. In OFD it starts lfsk check routines and initializes
571 * \param[in] env execution environment
572 * \param[in] pdev higher device in stack, NULL for OFD
573 * \param[in] dev lu_device of OFD device
575 * \retval 0 if successful
576 * \retval negative value on error
578 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
579 struct lu_device *dev)
581 struct ofd_thread_info *info;
582 struct ofd_device *ofd = ofd_dev(dev);
583 struct obd_device *obd = ofd_obd(ofd);
584 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
589 info = ofd_info_init(env, NULL);
593 /* initialize lower device */
594 rc = next->ld_ops->ldo_prepare(env, dev, next);
598 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
599 ofd_lfsck_out_notify, ofd, false);
601 CERROR("%s: failed to initialize lfsck: rc = %d\n",
606 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
607 /* The LFSCK instance is registered just now, so it must be there when
608 * register the namespace to such instance. */
609 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
611 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
612 LASSERT(obd->obd_no_conn);
613 spin_lock(&obd->obd_dev_lock);
614 obd->obd_no_conn = 0;
615 spin_unlock(&obd->obd_dev_lock);
617 if (obd->obd_recovering == 0)
618 ofd_postrecov(env, ofd);
624 * Implementation of lu_device_operations::ldo_recovery_complete.
626 * This method notifies all layers about 'recovery complete' event. That means
627 * device is in full state and consistent. An OFD calculates available grant
628 * space upon this event.
630 * \param[in] env execution environment
631 * \param[in] dev lu_device of OFD device
633 * \retval 0 if successful
634 * \retval negative value on error
636 static int ofd_recovery_complete(const struct lu_env *env,
637 struct lu_device *dev)
639 struct ofd_thread_info *oti = ofd_info(env);
640 struct ofd_device *ofd = ofd_dev(dev);
641 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
647 * Grant space for object precreation on the self export.
648 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
649 * is enough to create 10k objects. More space is then acquired for
650 * precreation in tgt_grant_create().
652 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
653 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
654 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
655 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
656 OBD_CONNECT_GRANT_PARAM;
657 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
659 rc = next->ld_ops->ldo_recovery_complete(env, next);
664 * lu_device_operations matrix for OFD device.
666 static struct lu_device_operations ofd_lu_ops = {
667 .ldo_object_alloc = ofd_object_alloc,
668 .ldo_process_config = ofd_process_config,
669 .ldo_recovery_complete = ofd_recovery_complete,
670 .ldo_prepare = ofd_prepare,
673 LPROC_SEQ_FOPS(lprocfs_nid_stats_clear);
676 * Initialize all needed procfs entries for OFD device.
678 * \param[in] ofd OFD device
680 * \retval 0 if successful
681 * \retval negative value on error
683 static int ofd_procfs_init(struct ofd_device *ofd)
685 struct obd_device *obd = ofd_obd(ofd);
686 struct proc_dir_entry *entry;
691 /* lprocfs must be setup before the ofd so state can be safely added
692 * to /proc incrementally as the ofd is setup */
693 obd->obd_vars = lprocfs_ofd_obd_vars;
694 rc = lprocfs_obd_setup(obd, false);
696 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
701 rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
703 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
705 GOTO(obd_cleanup, rc);
708 obd->obd_uses_nid_stats = 1;
710 entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
713 CERROR("%s: error %d setting up lprocfs for %s\n",
714 obd->obd_name, rc, "exports");
715 GOTO(obd_cleanup, rc);
717 obd->obd_proc_exports_entry = entry;
719 entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
720 obd, &lprocfs_nid_stats_clear_fops);
723 CERROR("%s: add proc entry 'clear' failed: %d.\n",
725 GOTO(obd_cleanup, rc);
728 ofd_stats_counter_init(obd->obd_stats);
730 rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
731 ofd_stats_counter_init);
733 GOTO(obd_cleanup, rc);
736 lprocfs_obd_cleanup(obd);
737 lprocfs_free_obd_stats(obd);
743 * Expose OSD statistics to OFD layer.
745 * The osd interfaces to the backend file system exposes useful data
746 * such as brw_stats and read or write cache states. This same data
747 * needs to be exposed into the obdfilter (ofd) layer to maintain
748 * backwards compatibility. This function creates the symlinks in the
749 * proc layer to enable this.
751 * \param[in] ofd OFD device
753 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
755 struct obd_device *obd = ofd_obd(ofd);
756 struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
758 if (obd->obd_proc_entry == NULL)
761 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
762 "../../%s/%s/brw_stats",
763 osd_obd->obd_type->typ_name, obd->obd_name);
765 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
766 "../../%s/%s/read_cache_enable",
767 osd_obd->obd_type->typ_name, obd->obd_name);
769 lprocfs_add_symlink("readcache_max_filesize",
771 "../../%s/%s/readcache_max_filesize",
772 osd_obd->obd_type->typ_name, obd->obd_name);
774 lprocfs_add_symlink("writethrough_cache_enable",
776 "../../%s/%s/writethrough_cache_enable",
777 osd_obd->obd_type->typ_name, obd->obd_name);
781 * Cleanup all procfs entries in OFD.
783 * \param[in] ofd OFD device
785 static void ofd_procfs_fini(struct ofd_device *ofd)
787 struct obd_device *obd = ofd_obd(ofd);
789 lprocfs_free_per_client_stats(obd);
790 lprocfs_obd_cleanup(obd);
791 lprocfs_free_obd_stats(obd);
792 lprocfs_job_stats_fini(obd);
796 * Stop SEQ/FID server on OFD.
798 * \param[in] env execution environment
799 * \param[in] ofd OFD device
801 * \retval 0 if successful
802 * \retval negative value on error
804 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
806 return seq_site_fini(env, &ofd->ofd_seq_site);
810 * Start SEQ/FID server on OFD.
812 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
813 * It also connects to the master server to get own FID sequence (SEQ) range
814 * to this particular OFD. Typically that happens when the OST is first
815 * formatted or in the rare case that it exhausts the local sequence range.
817 * The sequence range is allocated out to the MDTs for OST object allocations,
818 * and not directly to the clients.
820 * \param[in] env execution environment
821 * \param[in] ofd OFD device
823 * \retval 0 if successful
824 * \retval negative value on error
826 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
828 struct seq_server_site *ss = &ofd->ofd_seq_site;
829 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
830 char *obd_name = ofd_name(ofd);
834 ss = &ofd->ofd_seq_site;
835 lu->ld_site->ld_seq_site = ss;
836 ss->ss_lu = lu->ld_site;
837 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
839 OBD_ALLOC(name, sizeof(obd_name) * 2 + 10);
843 OBD_ALLOC_PTR(ss->ss_server_seq);
844 if (ss->ss_server_seq == NULL)
845 GOTO(out_name, rc = -ENOMEM);
847 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
848 LUSTRE_SEQ_SERVER, ss);
850 CERROR("%s : seq server init error %d\n", obd_name, rc);
851 GOTO(out_server, rc);
853 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
855 OBD_ALLOC_PTR(ss->ss_client_seq);
856 if (ss->ss_client_seq == NULL)
857 GOTO(out_server, rc = -ENOMEM);
860 * It always printed as "%p", so that the name is unique in the kernel,
861 * even if the filesystem is mounted twice. So sizeof(.) * 2 is enough.
863 snprintf(name, sizeof(obd_name) * 2 + 7, "%p-super", obd_name);
864 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
867 CERROR("%s : seq client init error %d\n", obd_name, rc);
868 GOTO(out_client, rc);
871 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
875 seq_client_fini(ss->ss_client_seq);
876 OBD_FREE_PTR(ss->ss_client_seq);
877 ss->ss_client_seq = NULL;
879 seq_server_fini(ss->ss_server_seq, env);
880 OBD_FREE_PTR(ss->ss_server_seq);
881 ss->ss_server_seq = NULL;
884 OBD_FREE(name, sizeof(obd_name) * 2 + 10);
890 * OFD request handler for OST_SET_INFO RPC.
892 * This is OFD-specific part of request handling
894 * \param[in] tsi target session environment for this request
896 * \retval 0 if successful
897 * \retval negative value on error
899 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
901 struct ptlrpc_request *req = tgt_ses_req(tsi);
902 struct ost_body *body = NULL, *repbody;
903 void *key, *val = NULL;
904 int keylen, vallen, rc = 0;
905 bool is_grant_shrink;
909 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
911 DEBUG_REQ(D_HA, req, "no set_info key");
912 RETURN(err_serious(-EFAULT));
914 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
917 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
919 DEBUG_REQ(D_HA, req, "no set_info val");
920 RETURN(err_serious(-EFAULT));
922 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
925 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
927 /* In this case the value is actually an RMF_OST_BODY, so we
928 * transmutate the type of this PTLRPC */
929 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
931 rc = req_capsule_server_pack(tsi->tsi_pill);
935 if (is_grant_shrink) {
936 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
938 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
941 /** handle grant shrink, similar to a read request */
942 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
944 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
946 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
949 CERROR("%s: Unsupported key %s\n",
950 tgt_name(tsi->tsi_tgt), (char *)key);
953 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
960 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
962 * This function returns a list of extents which describes how a file's
963 * blocks are laid out on the disk.
965 * \param[in] env execution environment
966 * \param[in] ofd OFD device
967 * \param[in] fid FID of object
968 * \param[in] fiemap fiemap structure to fill with data
970 * \retval 0 if \a fiemap is filled with data successfully
971 * \retval negative value on error
973 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
974 struct lu_fid *fid, struct fiemap *fiemap)
976 struct ofd_object *fo;
979 fo = ofd_object_find(env, ofd, fid);
981 CERROR("%s: error finding object "DFID"\n",
982 ofd_name(ofd), PFID(fid));
986 ofd_read_lock(env, fo);
987 if (ofd_object_exists(fo))
988 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
991 ofd_read_unlock(env, fo);
992 ofd_object_put(env, fo);
996 struct locked_region {
997 struct list_head list;
998 struct lustre_handle lh;
1002 * Lock single extent and save lock handle in the list.
1004 * This is supplemental function for lock_zero_regions(). It allocates
1005 * new locked_region structure and locks it with extent lock, then adds
1006 * it to the list of all such regions.
1008 * \param[in] ns LDLM namespace
1009 * \param[in] res_id resource ID
1010 * \param[in] begin start of region
1011 * \param[in] end end of region
1012 * \param[in] locked list head of regions list
1014 * \retval 0 if successful locking
1015 * \retval negative value on error
1017 static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1018 unsigned long long begin, unsigned long long end,
1019 struct list_head *locked)
1021 struct locked_region *region = NULL;
1025 LASSERT(begin <= end);
1026 OBD_ALLOC_PTR(region);
1030 rc = tgt_extent_lock(ns, res_id, begin, end, ®ion->lh,
1035 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
1037 list_add(®ion->list, locked);
1043 * Lock the sparse areas of given resource.
1045 * The locking of sparse areas will cause dirty data to be flushed back from
1046 * clients. This is used when getting the FIEMAP of an object to make sure
1047 * there is no unaccounted cached data on clients.
1049 * This function goes through \a fiemap list of extents and locks only sparse
1050 * areas between extents.
1052 * \param[in] ns LDLM namespace
1053 * \param[in] res_id resource ID
1054 * \param[in] fiemap file extents mapping on disk
1055 * \param[in] locked list head of regions list
1057 * \retval 0 if successful
1058 * \retval negative value on error
1060 static int lock_zero_regions(struct ldlm_namespace *ns,
1061 struct ldlm_res_id *res_id,
1062 struct fiemap *fiemap,
1063 struct list_head *locked)
1065 __u64 begin = fiemap->fm_start;
1068 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
1072 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1073 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1074 if (fiemap_start[i].fe_logical > begin) {
1075 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1076 begin, fiemap_start[i].fe_logical);
1077 rc = lock_region(ns, res_id, begin,
1078 fiemap_start[i].fe_logical, locked);
1083 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1086 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1087 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1088 begin, fiemap->fm_start + fiemap->fm_length);
1089 rc = lock_region(ns, res_id, begin,
1090 fiemap->fm_start + fiemap->fm_length, locked);
1097 * Unlock all previously locked sparse areas for given resource.
1099 * This function goes through list of locked regions, unlocking and freeing
1102 * \param[in] ns LDLM namespace
1103 * \param[in] locked list head of regions list
1106 unlock_zero_regions(struct ldlm_namespace *ns, struct list_head *locked)
1108 struct locked_region *entry, *temp;
1110 list_for_each_entry_safe(entry, temp, locked, list) {
1111 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
1112 tgt_extent_unlock(&entry->lh, LCK_PR);
1113 list_del(&entry->list);
1114 OBD_FREE_PTR(entry);
1119 * OFD request handler for OST_GET_INFO RPC.
1121 * This is OFD-specific part of request handling. The OFD-specific keys are:
1122 * - KEY_LAST_ID (obsolete)
1126 * This function reads needed data from storage and fills reply with it.
1128 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1129 * and is kept for compatibility.
1131 * \param[in] tsi target session environment for this request
1133 * \retval 0 if successful
1134 * \retval negative value on error
1136 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1138 struct obd_export *exp = tsi->tsi_exp;
1139 struct ofd_device *ofd = ofd_exp(exp);
1140 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1143 int replylen, rc = 0;
1147 /* this common part for get_info rpc */
1148 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1150 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1151 RETURN(err_serious(-EPROTO));
1153 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1156 if (KEY_IS(KEY_LAST_ID)) {
1158 struct ofd_seq *oseq;
1160 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1161 rc = req_capsule_server_pack(tsi->tsi_pill);
1163 RETURN(err_serious(rc));
1165 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1167 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1168 (u64)exp->exp_filter_data.fed_group);
1172 *last_id = ofd_seq_last_oid(oseq);
1173 ofd_seq_put(tsi->tsi_env, oseq);
1174 } else if (KEY_IS(KEY_FIEMAP)) {
1175 struct ll_fiemap_info_key *fm_key;
1176 struct fiemap *fiemap;
1179 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1181 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1182 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1184 RETURN(err_serious(rc));
1186 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1188 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1190 replylen = fiemap_count_to_size(
1191 fm_key->lfik_fiemap.fm_extent_count);
1192 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1193 RCL_SERVER, replylen);
1195 rc = req_capsule_server_pack(tsi->tsi_pill);
1197 RETURN(err_serious(rc));
1199 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1203 *fiemap = fm_key->lfik_fiemap;
1204 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1206 /* LU-3219: Lock the sparse areas to make sure dirty
1207 * flushed back from client, then call fiemap again. */
1208 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1209 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1210 struct list_head locked;
1212 INIT_LIST_HEAD(&locked);
1213 ost_fid_build_resid(fid, &fti->fti_resid);
1214 rc = lock_zero_regions(ofd->ofd_namespace,
1215 &fti->fti_resid, fiemap,
1217 if (rc == 0 && !list_empty(&locked)) {
1218 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1220 unlock_zero_regions(ofd->ofd_namespace,
1224 } else if (KEY_IS(KEY_LAST_FID)) {
1225 struct ofd_device *ofd = ofd_exp(exp);
1226 struct ofd_seq *oseq;
1230 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1231 rc = req_capsule_server_pack(tsi->tsi_pill);
1233 RETURN(err_serious(rc));
1235 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1237 RETURN(err_serious(-EPROTO));
1239 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1241 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1245 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1246 ostid_seq(&fti->fti_ostid));
1248 RETURN(PTR_ERR(oseq));
1250 rc = ostid_to_fid(fid, &oseq->os_oi,
1251 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1255 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1258 ofd_seq_put(tsi->tsi_env, oseq);
1260 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1264 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1271 * OFD request handler for OST_GETATTR RPC.
1273 * This is OFD-specific part of request handling. It finds the OFD object
1274 * by its FID, gets attributes from storage and packs result to the reply.
1276 * \param[in] tsi target session environment for this request
1278 * \retval 0 if successful
1279 * \retval negative value on error
1281 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1283 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1284 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1285 struct ost_body *repbody;
1286 struct lustre_handle lh = { 0 };
1287 struct ofd_object *fo;
1289 enum ldlm_mode lock_mode = LCK_PR;
1294 LASSERT(tsi->tsi_ost_body != NULL);
1296 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1297 if (repbody == NULL)
1300 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1301 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1303 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1304 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1307 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1310 rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace,
1311 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1317 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1319 GOTO(out, rc = PTR_ERR(fo));
1321 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1325 obdo_from_la(&repbody->oa, &fti->fti_attr,
1326 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1328 /* Store object version in reply */
1329 curr_version = dt_version_get(tsi->tsi_env,
1330 ofd_object_child(fo));
1331 if ((__s64)curr_version != -EOPNOTSUPP) {
1332 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1333 repbody->oa.o_data_version = curr_version;
1337 ofd_object_put(tsi->tsi_env, fo);
1340 tgt_extent_unlock(&lh, lock_mode);
1342 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1345 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1346 repbody->oa.o_flags = OBD_FL_FLUSH;
1352 * OFD request handler for OST_SETATTR RPC.
1354 * This is OFD-specific part of request handling. It finds the OFD object
1355 * by its FID, sets attributes from request and packs result to the reply.
1357 * \param[in] tsi target session environment for this request
1359 * \retval 0 if successful
1360 * \retval negative value on error
1362 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1364 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1365 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1366 struct ost_body *body = tsi->tsi_ost_body;
1367 struct ost_body *repbody;
1368 struct ldlm_resource *res;
1369 struct ofd_object *fo;
1370 struct filter_fid *ff = NULL;
1375 LASSERT(body != NULL);
1377 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1378 if (repbody == NULL)
1381 repbody->oa.o_oi = body->oa.o_oi;
1382 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1384 /* This would be very bad - accidentally truncating a file when
1385 * changing the time or similar - bug 12203. */
1386 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1387 body->oa.o_size != OBD_OBJECT_EOF) {
1388 static char mdsinum[48];
1390 if (body->oa.o_valid & OBD_MD_FLFID)
1391 snprintf(mdsinum, sizeof(mdsinum) - 1,
1392 "of parent "DFID, body->oa.o_parent_seq,
1393 body->oa.o_parent_oid, 0);
1397 CERROR("%s: setattr from %s is trying to truncate object "DFID
1398 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1399 PFID(&tsi->tsi_fid), mdsinum);
1403 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1405 GOTO(out, rc = PTR_ERR(fo));
1407 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1408 fti->fti_attr.la_valid &= ~LA_TYPE;
1410 if (body->oa.o_valid & OBD_MD_FLFID) {
1411 ff = &fti->fti_mds_fid;
1412 ofd_prepare_fidea(ff, &body->oa);
1415 /* setting objects attributes (including owner/group) */
1416 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff);
1420 obdo_from_la(&repbody->oa, &fti->fti_attr,
1421 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1423 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1427 ofd_object_put(tsi->tsi_env, fo);
1430 /* we do not call this before to avoid lu_object_find() in
1431 * ->lvbo_update() holding another reference on the object.
1432 * otherwise concurrent destroy can make the object unavailable
1433 * for 2nd lu_object_find() waiting for the first reference
1434 * to go... deadlock! */
1435 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1436 &tsi->tsi_resid, LDLM_EXTENT, 0);
1438 ldlm_res_lvbo_update(res, NULL, 0);
1439 ldlm_resource_putref(res);
1446 * Destroy OST orphans.
1448 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1449 * set then we must destroy possible orphaned objects.
1451 * \param[in] env execution environment
1452 * \param[in] exp OBD export
1453 * \param[in] ofd OFD device
1454 * \param[in] oa obdo structure for reply
1456 * \retval 0 if successful
1457 * \retval negative value on error
1459 static int ofd_orphans_destroy(const struct lu_env *env,
1460 struct obd_export *exp,
1461 struct ofd_device *ofd, struct obdo *oa)
1463 struct ofd_thread_info *info = ofd_info(env);
1464 struct lu_fid *fid = &info->fti_fid;
1465 struct ost_id *oi = &oa->o_oi;
1466 struct ofd_seq *oseq;
1467 u64 seq = ostid_seq(oi);
1468 u64 end_id = ostid_id(oi);
1476 oseq = ofd_seq_get(ofd, seq);
1478 CERROR("%s: Can not find seq for "DOSTID"\n",
1479 ofd_name(ofd), POSTID(oi));
1484 last = ofd_seq_last_oid(oseq);
1487 LASSERT(exp != NULL);
1488 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1490 if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1493 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1494 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1496 while (oid > end_id) {
1497 rc = fid_set_id(fid, oid);
1498 if (unlikely(rc != 0))
1501 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1502 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1503 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1504 /* this is pretty fatal... */
1505 CEMERG("%s: error destroying precreated id "
1507 ofd_name(ofd), PFID(fid), rc);
1511 ofd_seq_last_oid_set(oseq, oid);
1512 /* update last_id on disk periodically so that if we
1513 * restart * we don't need to re-scan all of the just
1514 * deleted objects. */
1515 if ((oid & 511) == 0)
1516 ofd_seq_last_oid_write(env, ofd, oseq);
1520 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1521 ofd_name(ofd), seq, oid);
1525 ofd_seq_last_oid_set(oseq, oid);
1526 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1528 /* don't reuse orphan object, return last used objid */
1529 rc = 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;
1568 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1571 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1572 if (repbody == NULL)
1575 down_read(&ofd->ofd_lastid_rwsem);
1576 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1577 * we may do that in the future.
1578 * Return -ENOSPC until the LAST_ID rebuilt. */
1579 if (unlikely(ofd->ofd_lastid_rebuilding))
1580 GOTO(out_sem, rc = -ENOSPC);
1582 rep_oa = &repbody->oa;
1583 rep_oa->o_oi = oa->o_oi;
1585 LASSERT(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 rc = ostid_set_id(&rep_oa->o_oi,
1628 ofd_seq_last_oid(oseq));
1631 diff = oid - ofd_seq_last_oid(oseq);
1632 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1633 ofd_seq_last_oid(oseq), diff);
1634 if (-diff > OST_MAX_PRECREATE) {
1635 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1636 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1637 "(%llu), trust the OST\n",
1638 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1639 PFID(&oseq->os_oi.oi_fid),
1640 ofd_seq_last_oid(oseq));
1642 /* Let MDS know that we are so far ahead. */
1643 rc = ostid_set_id(&rep_oa->o_oi,
1644 ofd_seq_last_oid(oseq) + 1);
1645 } else if (diff < 0) {
1646 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1648 oseq->os_destroys_in_progress = 0;
1650 /* XXX: Used by MDS for the first time! */
1651 oseq->os_destroys_in_progress = 0;
1654 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1655 ofd->ofd_lastid_gen)) {
1656 /* Keep the export ref so we can send the reply. */
1657 ofd_obd_disconnect(class_export_get(exp));
1658 GOTO(out_nolock, rc = -ENOTCONN);
1661 mutex_lock(&oseq->os_create_lock);
1662 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1663 exp->exp_conn_cnt) {
1664 CERROR("%s: dropping old precreate request\n",
1668 /* only precreate if seq is 0, IDIF or normal and also o_id
1669 * must be specfied */
1670 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1671 !fid_seq_is_idif(seq)) || oid == 0) {
1672 diff = 1; /* shouldn't we create this right now? */
1674 diff = oid - ofd_seq_last_oid(oseq);
1675 /* Do sync create if the seq is about to used up */
1676 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1677 if (unlikely(oid >= IDIF_MAX_OID - 1))
1679 } else if (fid_seq_is_norm(seq)) {
1681 LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1684 CERROR("%s : invalid o_seq "DOSTID"\n",
1685 ofd_name(ofd), POSTID(&oa->o_oi));
1686 GOTO(out, rc = -EINVAL);
1691 CERROR("%s: invalid precreate request for "
1692 DOSTID", last_id %llu. "
1693 "Likely MDS last_id corruption\n",
1694 ofd_name(ofd), POSTID(&oa->o_oi),
1695 ofd_seq_last_oid(oseq));
1696 GOTO(out, rc = -EINVAL);
1701 cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
1706 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1707 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1708 /* don't enforce grant during orphan recovery */
1709 granted = tgt_grant_create(tsi->tsi_env,
1710 ofd_obd(ofd)->obd_self_export,
1715 CDEBUG(D_HA, "%s: failed to acquire grant "
1716 "space for precreate (%lld): rc = %d\n",
1717 ofd_name(ofd), diff, rc);
1722 /* This can happen if a new OST is formatted and installed
1723 * in place of an old one at the same index. Instead of
1724 * precreating potentially millions of deleted old objects
1725 * (possibly filling the OST), only precreate the last batch.
1726 * LFSCK will eventually clean up any orphans. LU-14 */
1727 if (diff > 5 * OST_MAX_PRECREATE) {
1728 diff = OST_MAX_PRECREATE / 2;
1729 LCONSOLE_WARN("%s: Too many FIDs to precreate "
1730 "OST replaced or reformatted: "
1731 "LFSCK will clean up",
1734 CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over "
1735 "%u larger than the LAST_ID "DOSTID", only "
1736 "precreating the last %lld objects.\n",
1737 ofd_name(ofd), POSTID(&oa->o_oi),
1738 5 * OST_MAX_PRECREATE,
1739 POSTID(&oseq->os_oi), diff);
1740 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1744 next_id = ofd_seq_last_oid(oseq) + 1;
1745 count = ofd_precreate_batch(ofd, (int)diff);
1747 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1748 " at %llu\n", ofd_name(ofd),
1749 count, seq, next_id);
1751 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1752 && cfs_time_after(jiffies, enough_time)) {
1753 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1754 " created at a rate of %d/s\n",
1755 ofd_name(ofd), created, diff + created,
1756 created / DISK_TIMEOUT);
1760 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1761 oseq, count, sync_trans);
1765 } else if (rc < 0) {
1771 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1772 LCONSOLE_WARN("%s: can't create the same count of"
1773 " objects when replaying the request"
1774 " (diff is %lld). see LU-4621\n",
1775 ofd_name(ofd), diff);
1778 /* some objects got created, we can return
1779 * them, even if last creation failed */
1782 CERROR("%s: unable to precreate: rc = %d\n",
1785 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1786 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1787 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1792 rc = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1795 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1798 mutex_unlock(&oseq->os_create_lock);
1801 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1802 struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
1803 struct lu_fid *fid = &info->fti_fid;
1805 /* For compatible purpose, it needs to convert back to
1806 * OST ID before put it on wire. */
1807 *fid = rep_oa->o_oi.oi_fid;
1808 fid_to_ostid(fid, &rep_oa->o_oi);
1810 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1812 ofd_seq_put(tsi->tsi_env, oseq);
1815 up_read(&ofd->ofd_lastid_rwsem);
1820 * OFD request handler for OST_DESTROY RPC.
1822 * This is OFD-specific part of request handling. It destroys data objects
1823 * related to destroyed object on MDT.
1825 * \param[in] tsi target session environment for this request
1827 * \retval 0 if successful
1828 * \retval negative value on error
1830 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1832 const struct ost_body *body = tsi->tsi_ost_body;
1833 struct ost_body *repbody;
1834 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1835 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1836 struct lu_fid *fid = &fti->fti_fid;
1843 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1846 /* This is old case for clients before Lustre 2.4 */
1847 /* If there's a DLM request, cancel the locks mentioned in it */
1848 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1850 struct ldlm_request *dlm;
1852 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1855 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1858 *fid = body->oa.o_oi.oi_fid;
1859 oid = ostid_id(&body->oa.o_oi);
1862 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1864 /* check that o_misc makes sense */
1865 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1866 count = body->oa.o_misc;
1868 count = 1; /* default case - single destroy */
1870 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1871 POSTID(&body->oa.o_oi), count);
1876 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1877 if (lrc == -ENOENT) {
1879 "%s: destroying non-existent object "DFID"\n",
1880 ofd_name(ofd), PFID(fid));
1881 /* rewrite rc with -ENOENT only if it is 0 */
1884 } else if (lrc != 0) {
1885 CERROR("%s: error destroying object "DFID": %d\n",
1886 ofd_name(ofd), PFID(fid), lrc);
1892 lrc = fid_set_id(fid, oid);
1893 if (unlikely(lrc != 0 && count > 0))
1894 GOTO(out, rc = lrc);
1897 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1903 fid_to_ostid(fid, &repbody->oa.o_oi);
1908 * OFD request handler for OST_STATFS RPC.
1910 * This function gets statfs data from storage as part of request
1913 * \param[in] tsi target session environment for this request
1915 * \retval 0 if successful
1916 * \retval negative value on error
1918 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1920 struct obd_statfs *osfs;
1925 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1927 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1928 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
1930 CERROR("%s: statfs failed: rc = %d\n",
1931 tgt_name(tsi->tsi_tgt), rc);
1933 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1936 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1943 * OFD request handler for OST_SYNC RPC.
1945 * Sync object data or all filesystem data to the disk and pack the
1948 * \param[in] tsi target session environment for this request
1950 * \retval 0 if successful
1951 * \retval negative value on error
1953 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1955 struct ost_body *body = tsi->tsi_ost_body;
1956 struct ost_body *repbody;
1957 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1958 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1959 struct ofd_object *fo = NULL;
1964 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1966 /* if no objid is specified, it means "sync whole filesystem" */
1967 if (!fid_is_zero(&tsi->tsi_fid)) {
1968 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1970 RETURN(PTR_ERR(fo));
1973 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1974 fo != NULL ? ofd_object_child(fo) : NULL,
1975 repbody->oa.o_size, repbody->oa.o_blocks);
1979 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1984 repbody->oa.o_oi = body->oa.o_oi;
1985 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1987 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1989 obdo_from_la(&repbody->oa, &fti->fti_attr,
1992 /* don't return rc from getattr */
1997 ofd_object_put(tsi->tsi_env, fo);
2002 * OFD request handler for OST_PUNCH RPC.
2004 * This is part of request processing. Validate request fields,
2005 * punch (truncate) the given OFD object and pack reply.
2007 * \param[in] tsi target session environment for this request
2009 * \retval 0 if successful
2010 * \retval negative value on error
2012 static int ofd_punch_hdl(struct tgt_session_info *tsi)
2014 const struct obdo *oa = &tsi->tsi_ost_body->oa;
2015 struct ost_body *repbody;
2016 struct ofd_thread_info *info = tsi2ofd_info(tsi);
2017 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
2018 struct ldlm_resource *res;
2019 struct ofd_object *fo;
2020 struct filter_fid *ff = NULL;
2022 struct lustre_handle lh = { 0, };
2029 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
2031 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
2032 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
2034 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2035 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2036 RETURN(err_serious(-EPROTO));
2038 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2039 if (repbody == NULL)
2040 RETURN(err_serious(-ENOMEM));
2042 /* punch start,end are passed in o_size,o_blocks throught wire */
2046 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2049 /* standard truncate optimization: if file body is completely
2050 * destroyed, don't send data back to the server. */
2052 flags |= LDLM_FL_AST_DISCARD_DATA;
2054 repbody->oa.o_oi = oa->o_oi;
2055 repbody->oa.o_valid = OBD_MD_FLID;
2057 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2058 oa->o_flags & OBD_FL_SRVLOCK;
2061 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
2067 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2068 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2069 oa->o_valid, start, end);
2071 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2074 GOTO(out, rc = PTR_ERR(fo));
2076 la_from_obdo(&info->fti_attr, oa,
2077 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2078 info->fti_attr.la_size = start;
2079 info->fti_attr.la_valid |= LA_SIZE;
2081 if (oa->o_valid & OBD_MD_FLFID) {
2082 ff = &info->fti_mds_fid;
2083 ofd_prepare_fidea(ff, oa);
2086 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2087 ff, (struct obdo *)oa);
2091 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2095 ofd_object_put(tsi->tsi_env, fo);
2098 tgt_extent_unlock(&lh, LCK_PW);
2100 /* we do not call this before to avoid lu_object_find() in
2101 * ->lvbo_update() holding another reference on the object.
2102 * otherwise concurrent destroy can make the object unavailable
2103 * for 2nd lu_object_find() waiting for the first reference
2104 * to go... deadlock! */
2105 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
2108 ldlm_res_lvbo_update(res, NULL, 0);
2109 ldlm_resource_putref(res);
2115 static int ofd_ladvise_prefetch(const struct lu_env *env,
2116 struct ofd_object *fo,
2117 struct niobuf_local *lnb,
2118 __u64 start, __u64 end, enum dt_bufs_type dbt)
2120 struct ofd_thread_info *info = ofd_info(env);
2121 pgoff_t start_index, end_index, pages;
2122 struct niobuf_remote rnb;
2123 unsigned long nr_local;
2129 ofd_read_lock(env, fo);
2130 if (!ofd_object_exists(fo))
2131 GOTO(out_unlock, rc = -ENOENT);
2133 rc = ofd_attr_get(env, fo, &info->fti_attr);
2135 GOTO(out_unlock, rc);
2137 if (end > info->fti_attr.la_size)
2138 end = info->fti_attr.la_size;
2141 GOTO(out_unlock, rc);
2143 /* We need page aligned offset and length */
2144 start_index = start >> PAGE_SHIFT;
2145 end_index = (end - 1) >> PAGE_SHIFT;
2146 pages = end_index - start_index + 1;
2148 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2149 PTLRPC_MAX_BRW_PAGES;
2150 rnb.rnb_offset = start_index << PAGE_SHIFT;
2151 rnb.rnb_len = nr_local << PAGE_SHIFT;
2152 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, dbt);
2153 if (unlikely(rc < 0))
2156 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2157 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2160 start_index += nr_local;
2165 ofd_read_unlock(env, fo);
2170 * OFD request handler for OST_LADVISE RPC.
2172 * Tune cache or perfetch policies according to advices.
2174 * \param[in] tsi target session environment for this request
2176 * \retval 0 if successful
2177 * \retval negative errno on error
2179 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2181 struct ptlrpc_request *req = tgt_ses_req(tsi);
2182 struct obd_export *exp = tsi->tsi_exp;
2183 struct ofd_device *ofd = ofd_exp(exp);
2184 struct ost_body *body, *repbody;
2185 struct ofd_thread_info *info;
2186 struct ofd_object *fo;
2187 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2188 const struct lu_env *env = svc_thread->t_env;
2189 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2190 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2191 struct lu_ladvise *ladvise;
2193 struct ladvise_hdr *ladvise_hdr;
2194 struct obd_ioobj ioo;
2195 struct lustre_handle lockh = { 0 };
2198 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 if (ptlrpc_connection_is_local(exp->exp_connection))
2251 dbt |= DT_BUFS_TYPE_LOCAL;
2253 for (i = 0; i < num_advise; i++, ladvise++) {
2254 start = ladvise->lla_start;
2255 end = ladvise->lla_end;
2257 rc = err_serious(-EPROTO);
2261 /* Handle different advice types */
2262 switch (ladvise->lla_advice) {
2266 case LU_LADVISE_WILLREAD:
2270 ioo.ioo_oid = body->oa.o_oi;
2272 rc = tgt_extent_lock(exp->exp_obd->obd_namespace,
2273 &tsi->tsi_resid, start, end - 1,
2274 &lockh, LCK_PR, &flags);
2278 req->rq_status = ofd_ladvise_prefetch(env, fo,
2281 tgt_extent_unlock(&lockh, LCK_PR);
2283 case LU_LADVISE_DONTNEED:
2284 rc = dt_ladvise(env, dob, ladvise->lla_start,
2285 ladvise->lla_end, LU_LADVISE_DONTNEED);
2292 ofd_object_put(env, fo);
2293 req->rq_status = rc;
2298 * OFD request handler for OST_QUOTACTL RPC.
2300 * This is part of request processing to validate incoming request fields,
2301 * get the requested data from OSD and pack reply.
2303 * \param[in] tsi target session environment for this request
2305 * \retval 0 if successful
2306 * \retval negative value on error
2308 static int ofd_quotactl(struct tgt_session_info *tsi)
2310 struct obd_quotactl *oqctl, *repoqc;
2311 struct lu_nodemap *nodemap;
2317 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2319 RETURN(err_serious(-EPROTO));
2321 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2323 RETURN(err_serious(-ENOMEM));
2327 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2328 if (IS_ERR(nodemap))
2329 RETURN(PTR_ERR(nodemap));
2332 if (oqctl->qc_type == USRQUOTA)
2333 id = nodemap_map_id(nodemap, NODEMAP_UID,
2334 NODEMAP_CLIENT_TO_FS,
2336 else if (oqctl->qc_type == GRPQUOTA)
2337 id = nodemap_map_id(nodemap, NODEMAP_GID,
2338 NODEMAP_CLIENT_TO_FS,
2341 nodemap_putref(nodemap);
2343 if (repoqc->qc_id != id)
2344 swap(repoqc->qc_id, id);
2346 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
2348 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2351 if (repoqc->qc_id != id)
2352 swap(repoqc->qc_id, id);
2358 * Calculate the amount of time for lock prolongation.
2360 * This is helper for ofd_prolong_extent_locks() function to get
2361 * the timeout extra time.
2363 * \param[in] req current request
2365 * \retval amount of time to extend the timeout with
2367 static inline int prolong_timeout(struct ptlrpc_request *req)
2369 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2373 return obd_timeout / 2;
2375 req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec;
2376 return max_t(time_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
2381 * Prolong lock timeout for the given extent.
2383 * This function finds all locks related with incoming request and
2384 * prolongs their timeout.
2386 * If a client is holding a lock for a long time while it sends
2387 * read or write RPCs to the OST for the object under this lock,
2388 * then we don't want the OST to evict the client. Otherwise,
2389 * if the network or disk is very busy then the client may not
2390 * be able to make any progress to clear out dirty pages under
2391 * the lock and the application will fail.
2393 * Every time a Bulk Read/Write (BRW) request arrives for the object
2394 * covered by the lock, extend the timeout on that lock. The RPC should
2395 * contain a lock handle for the lock it is using, but this
2396 * isn't handled correctly by all client versions, and the
2397 * request may cover multiple locks.
2399 * \param[in] tsi target session environment for this request
2400 * \param[in] data struct of data to prolong locks
2403 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2404 struct ldlm_prolong_args *data)
2406 struct obdo *oa = &tsi->tsi_ost_body->oa;
2407 struct ldlm_lock *lock;
2411 data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi));
2412 data->lpa_export = tsi->tsi_exp;
2413 data->lpa_resid = tsi->tsi_resid;
2415 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2416 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2417 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2418 data->lpa_extent.end);
2420 if (oa->o_valid & OBD_MD_FLHANDLE) {
2421 /* mostly a request should be covered by only one lock, try
2423 lock = ldlm_handle2lock(&oa->o_handle);
2425 /* Fast path to check if the lock covers the whole IO
2426 * region exclusively. */
2427 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2428 &data->lpa_extent)) {
2430 LASSERT(lock->l_export == data->lpa_export);
2431 ldlm_lock_prolong_one(lock, data);
2432 LDLM_LOCK_PUT(lock);
2435 lock->l_last_used = ktime_get();
2436 LDLM_LOCK_PUT(lock);
2440 ldlm_resource_prolong(data);
2445 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2447 * Determine if \a lock and the lock from request \a req are equivalent
2448 * by comparing their resource names, modes, and extents.
2450 * It is used to give priority to read and write RPCs being done
2451 * under this lock so that the client can drop the contended
2452 * lock more quickly and let other clients use it. This improves
2453 * overall performance in the case where the first client gets a
2454 * very large lock extent that prevents other clients from
2455 * submitting their writes.
2457 * \param[in] req ptlrpc_request being processed
2458 * \param[in] lock contended lock to match
2460 * \retval 1 if lock is matched
2461 * \retval 0 otherwise
2463 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2464 struct ldlm_lock *lock)
2466 struct niobuf_remote *rnb;
2467 struct obd_ioobj *ioo;
2468 enum ldlm_mode mode;
2469 struct ldlm_extent ext;
2470 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2474 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2475 LASSERT(ioo != NULL);
2477 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2478 LASSERT(rnb != NULL);
2480 ext.start = rnb->rnb_offset;
2481 rnb += ioo->ioo_bufcnt - 1;
2482 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2484 LASSERT(lock->l_resource != NULL);
2485 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2488 /* a bulk write can only hold a reference on a PW extent lock
2491 mode = LCK_PW | LCK_GROUP;
2492 if (opc == OST_READ)
2493 /* whereas a bulk read can be protected by either a PR or PW
2497 if (!(lock->l_granted_mode & mode))
2500 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2504 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2506 * Check for whether the given PTLRPC request (\a req) is blocking
2507 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2510 * \param[in] req the incoming request
2512 * \retval 1 if \a req is blocking an LDLM lock cancel
2513 * \retval 0 if it is not
2514 * \retval -ESTALE if lock is not found
2516 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2518 struct tgt_session_info *tsi;
2519 struct obd_ioobj *ioo;
2520 struct niobuf_remote *rnb;
2522 struct ldlm_prolong_args pa = { 0 };
2526 /* Don't use tgt_ses_info() to get session info, because lock_match()
2527 * can be called while request has no processing thread yet. */
2528 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2531 * Use LASSERT below because malformed RPCs should have
2532 * been filtered out in tgt_hpreq_handler().
2534 opc = lustre_msg_get_opc(req->rq_reqmsg);
2535 LASSERT(opc == OST_READ || opc == OST_WRITE);
2537 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2538 LASSERT(ioo != NULL);
2540 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2541 LASSERT(rnb != NULL);
2542 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2544 pa.lpa_mode = LCK_PW | LCK_GROUP;
2545 if (opc == OST_READ)
2546 pa.lpa_mode |= LCK_PR;
2548 pa.lpa_extent.start = rnb->rnb_offset;
2549 rnb += ioo->ioo_bufcnt - 1;
2550 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2552 DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
2553 " (%llu->%llu)\n", tgt_name(tsi->tsi_tgt),
2554 current->comm, PFID(&tsi->tsi_fid), pa.lpa_extent.start,
2557 ofd_prolong_extent_locks(tsi, &pa);
2559 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2560 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2562 if (pa.lpa_blocks_cnt > 0)
2565 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2569 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2571 * Called after the request has been handled. It refreshes lock timeout again
2572 * so that client has more time to send lock cancel RPC.
2574 * \param[in] req request which is being processed.
2576 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2578 ofd_rw_hpreq_check(req);
2582 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2584 * This function checks if the given lock is the same by its resname, mode
2585 * and extent as one taken from the request.
2586 * It is used to give priority to punch/truncate RPCs that might lead to
2587 * the fastest release of that lock when a lock is contended.
2589 * \param[in] req ptlrpc_request being processed
2590 * \param[in] lock contended lock to match
2592 * \retval 1 if lock is matched
2593 * \retval 0 otherwise
2595 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2596 struct ldlm_lock *lock)
2598 struct tgt_session_info *tsi;
2600 struct ldlm_extent ext;
2604 /* Don't use tgt_ses_info() to get session info, because lock_match()
2605 * can be called while request has no processing thread yet. */
2606 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2609 * Use LASSERT below because malformed RPCs should have
2610 * been filtered out in tgt_hpreq_handler().
2612 LASSERT(tsi->tsi_ost_body != NULL);
2613 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2614 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2617 oa = &tsi->tsi_ost_body->oa;
2618 ext.start = oa->o_size;
2619 ext.end = oa->o_blocks;
2621 LASSERT(lock->l_resource != NULL);
2622 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2625 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2628 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2632 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2634 * High-priority queue request check for whether the given punch request
2635 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2636 * covered by an LDLM lock.
2640 * \param[in] req the incoming request
2642 * \retval 1 if \a req is blocking an LDLM lock cancel
2643 * \retval 0 if it is not
2644 * \retval -ESTALE if lock is not found
2646 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2648 struct tgt_session_info *tsi;
2650 struct ldlm_prolong_args pa = { 0 };
2654 /* Don't use tgt_ses_info() to get session info, because lock_match()
2655 * can be called while request has no processing thread yet. */
2656 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2657 LASSERT(tsi != NULL);
2658 oa = &tsi->tsi_ost_body->oa;
2660 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2661 oa->o_flags & OBD_FL_SRVLOCK));
2663 pa.lpa_mode = LCK_PW | LCK_GROUP;
2664 pa.lpa_extent.start = oa->o_size;
2665 pa.lpa_extent.end = oa->o_blocks;
2668 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2669 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2670 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2672 ofd_prolong_extent_locks(tsi, &pa);
2674 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2675 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2677 if (pa.lpa_blocks_cnt > 0)
2680 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2684 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2686 * Called after the request has been handled. It refreshes lock timeout again
2687 * so that client has more time to send lock cancel RPC.
2689 * \param[in] req request which is being processed.
2691 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2693 ofd_punch_hpreq_check(req);
2696 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2697 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2698 .hpreq_check = ofd_rw_hpreq_check,
2699 .hpreq_fini = ofd_rw_hpreq_fini
2702 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2703 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2704 .hpreq_check = ofd_punch_hpreq_check,
2705 .hpreq_fini = ofd_punch_hpreq_fini
2709 * Assign high priority operations to an IO request.
2711 * Check if the incoming request is a candidate for
2712 * high-priority processing. If it is, assign it a high
2713 * priority operations table.
2715 * \param[in] tsi target session environment for this request
2717 static void ofd_hp_brw(struct tgt_session_info *tsi)
2719 struct niobuf_remote *rnb;
2720 struct obd_ioobj *ioo;
2724 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2725 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2726 if (ioo->ioo_bufcnt > 0) {
2727 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2728 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2730 /* no high priority if server lock is needed */
2731 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2732 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2736 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2740 * Assign high priority operations to an punch request.
2742 * Check if the incoming request is a candidate for
2743 * high-priority processing. If it is, assign it a high
2744 * priority operations table.
2746 * \param[in] tsi target session environment for this request
2748 static void ofd_hp_punch(struct tgt_session_info *tsi)
2750 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2751 /* no high-priority if server lock is needed */
2752 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2753 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2754 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2755 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2757 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2760 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2761 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2762 #define OST_BRW_READ OST_READ
2763 #define OST_BRW_WRITE OST_WRITE
2766 * Table of OFD-specific request handlers
2768 * This table contains all opcodes accepted by OFD and
2769 * specifies handlers for them. The tgt_request_handler()
2770 * uses such table from each target to process incoming
2773 static struct tgt_handler ofd_tgt_handlers[] = {
2774 TGT_RPC_HANDLER(OST_FIRST_OPC,
2775 0, OST_CONNECT, tgt_connect,
2776 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2777 TGT_RPC_HANDLER(OST_FIRST_OPC,
2778 0, OST_DISCONNECT, tgt_disconnect,
2779 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2780 TGT_RPC_HANDLER(OST_FIRST_OPC,
2781 0, OST_SET_INFO, ofd_set_info_hdl,
2782 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2783 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2784 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl),
2785 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2786 OST_SETATTR, ofd_setattr_hdl),
2787 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2788 OST_CREATE, ofd_create_hdl),
2789 TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR,
2790 OST_DESTROY, ofd_destroy_hdl),
2791 TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl),
2792 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2793 OST_BRW_READ, tgt_brw_read,
2795 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2796 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write,
2798 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2799 OST_PUNCH, ofd_punch_hdl,
2801 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl),
2802 TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl),
2803 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_LADVISE, ofd_ladvise_hdl),
2806 static struct tgt_opc_slice ofd_common_slice[] = {
2808 .tos_opc_start = OST_FIRST_OPC,
2809 .tos_opc_end = OST_LAST_OPC,
2810 .tos_hs = ofd_tgt_handlers
2813 .tos_opc_start = OBD_FIRST_OPC,
2814 .tos_opc_end = OBD_LAST_OPC,
2815 .tos_hs = tgt_obd_handlers
2818 .tos_opc_start = LDLM_FIRST_OPC,
2819 .tos_opc_end = LDLM_LAST_OPC,
2820 .tos_hs = tgt_dlm_handlers
2823 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2824 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2825 .tos_hs = tgt_out_handlers
2828 .tos_opc_start = SEQ_FIRST_OPC,
2829 .tos_opc_end = SEQ_LAST_OPC,
2830 .tos_hs = seq_handlers
2833 .tos_opc_start = LFSCK_FIRST_OPC,
2834 .tos_opc_end = LFSCK_LAST_OPC,
2835 .tos_hs = tgt_lfsck_handlers
2838 .tos_opc_start = SEC_FIRST_OPC,
2839 .tos_opc_end = SEC_LAST_OPC,
2840 .tos_hs = tgt_sec_ctx_handlers
2847 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2848 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2851 * Implementation of lu_context_key::lct_key_exit.
2853 * Optional method called on lu_context_exit() for all allocated
2855 * It is used in OFD to sanitize context values which may be re-used
2856 * during another request processing by the same thread.
2858 * \param[in] ctx execution context
2859 * \param[in] key context key
2860 * \param[in] data ofd_thread_info
2862 static void ofd_key_exit(const struct lu_context *ctx,
2863 struct lu_context_key *key, void *data)
2865 struct ofd_thread_info *info = data;
2867 info->fti_env = NULL;
2868 info->fti_exp = NULL;
2871 info->fti_pre_version = 0;
2873 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2876 struct lu_context_key ofd_thread_key = {
2877 .lct_tags = LCT_DT_THREAD,
2878 .lct_init = ofd_key_init,
2879 .lct_fini = ofd_key_fini,
2880 .lct_exit = ofd_key_exit
2884 * Initialize OFD device according to parameters in the config log \a cfg.
2886 * This is the main starting point of OFD initialization. It fills all OFD
2887 * parameters with their initial values and calls other initializing functions
2888 * to set up all OFD subsystems.
2890 * \param[in] env execution environment
2891 * \param[in] m OFD device
2892 * \param[in] ldt LU device type of OFD
2893 * \param[in] cfg configuration log
2895 * \retval 0 if successful
2896 * \retval negative value on error
2898 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2899 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2901 const char *dev = lustre_cfg_string(cfg, 0);
2902 struct ofd_thread_info *info = NULL;
2903 struct obd_device *obd;
2904 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2905 struct obd_statfs *osfs;
2907 struct nm_config_file *nodemap_config;
2908 struct obd_device_target *obt;
2913 obd = class_name2obd(dev);
2915 CERROR("Cannot find obd with name %s\n", dev);
2919 rc = lu_env_refill((struct lu_env *)env);
2924 obt->obt_magic = OBT_MAGIC;
2926 m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2927 m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2929 spin_lock_init(&m->ofd_flags_lock);
2930 m->ofd_raid_degraded = 0;
2931 m->ofd_syncjournal = 0;
2933 tgd->tgd_grant_compat_disable = 0;
2934 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2937 spin_lock_init(&tgd->tgd_osfs_lock);
2938 tgd->tgd_osfs_age = cfs_time_shift_64(-1000);
2939 tgd->tgd_osfs_unstable = 0;
2940 tgd->tgd_statfs_inflight = 0;
2941 tgd->tgd_osfs_inflight = 0;
2944 spin_lock_init(&tgd->tgd_grant_lock);
2945 tgd->tgd_tot_dirty = 0;
2946 tgd->tgd_tot_granted = 0;
2947 tgd->tgd_tot_pending = 0;
2949 m->ofd_seq_count = 0;
2950 init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2951 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2952 spin_lock_init(&m->ofd_inconsistency_lock);
2954 spin_lock_init(&m->ofd_batch_lock);
2955 init_rwsem(&m->ofd_lastid_rwsem);
2957 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2958 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2959 /* set this lu_device to obd, because error handling need it */
2960 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2962 rc = ofd_procfs_init(m);
2964 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2968 /* No connection accepted until configurations will finish */
2969 spin_lock(&obd->obd_dev_lock);
2970 obd->obd_no_conn = 1;
2971 spin_unlock(&obd->obd_dev_lock);
2972 obd->obd_replayable = 1;
2973 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2974 char *str = lustre_cfg_string(cfg, 4);
2976 if (strchr(str, 'n')) {
2977 CWARN("%s: recovery disabled\n", obd->obd_name);
2978 obd->obd_replayable = 0;
2982 info = ofd_info_init(env, NULL);
2984 GOTO(err_fini_proc, rc = -EFAULT);
2986 rc = ofd_stack_init(env, m, cfg);
2988 CERROR("Can't init device stack, rc %d\n", rc);
2989 GOTO(err_fini_proc, rc);
2992 ofd_procfs_add_brw_stats_symlink(m);
2994 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2995 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2996 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2997 LDLM_NAMESPACE_SERVER,
2998 LDLM_NAMESPACE_GREEDY,
3000 if (m->ofd_namespace == NULL)
3001 GOTO(err_fini_stack, rc = -ENOMEM);
3002 /* set obd_namespace for compatibility with old code */
3003 obd->obd_namespace = m->ofd_namespace;
3004 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
3005 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
3006 m->ofd_namespace->ns_lvbp = m;
3008 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3009 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
3011 dt_conf_get(env, m->ofd_osd, &m->ofd_lut.lut_dt_conf);
3013 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
3014 OBD_FAIL_OST_ALL_REQUEST_NET,
3015 OBD_FAIL_OST_ALL_REPLY_NET);
3017 GOTO(err_free_ns, rc);
3019 /* populate cached statfs data */
3020 osfs = &ofd_info(env)->fti_u.osfs;
3021 rc = tgt_statfs_internal(env, &m->ofd_lut, osfs, 0, NULL);
3023 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
3024 GOTO(err_fini_lut, rc);
3026 if (!is_power_of_2(osfs->os_bsize)) {
3027 CERROR("%s: blocksize (%d) is not a power of 2\n",
3028 obd->obd_name, osfs->os_bsize);
3029 GOTO(err_fini_lut, rc = -EPROTO);
3031 tgd->tgd_blockbits = fls(osfs->os_bsize) - 1;
3033 if (DT_DEF_BRW_SIZE < (1U << tgd->tgd_blockbits))
3034 m->ofd_brw_size = 1U << tgd->tgd_blockbits;
3036 m->ofd_brw_size = DT_DEF_BRW_SIZE;
3038 m->ofd_cksum_types_supported = cksum_types_supported_server();
3039 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
3040 if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
3041 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
3043 rc = ofd_fs_setup(env, m, obd);
3045 GOTO(err_fini_lut, rc);
3047 fid.f_seq = FID_SEQ_LOCAL_NAME;
3050 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
3053 GOTO(err_fini_fs, rc);
3055 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
3057 if (IS_ERR(nodemap_config)) {
3058 rc = PTR_ERR(nodemap_config);
3060 GOTO(err_fini_los, rc);
3062 obt->obt_nodemap_config_file = nodemap_config;
3065 rc = ofd_start_inconsistency_verification_thread(m);
3067 GOTO(err_fini_nm, rc);
3069 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
3074 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
3075 obt->obt_nodemap_config_file = NULL;
3077 local_oid_storage_fini(env, m->ofd_los);
3080 ofd_fs_cleanup(env, m);
3082 tgt_fini(env, &m->ofd_lut);
3084 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3085 obd->obd_namespace = m->ofd_namespace = NULL;
3087 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3094 * Stop the OFD device
3096 * This function stops the OFD device and all its subsystems.
3097 * This is the end of OFD lifecycle.
3099 * \param[in] env execution environment
3100 * \param[in] m OFD device
3102 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3104 struct obd_device *obd = ofd_obd(m);
3105 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3106 struct lfsck_stop stop;
3108 stop.ls_status = LS_PAUSED;
3110 lfsck_stop(env, m->ofd_osd, &stop);
3111 target_recovery_fini(obd);
3112 if (m->ofd_namespace != NULL)
3113 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3114 d->ld_obd->obd_force);
3116 obd_exports_barrier(obd);
3117 obd_zombie_barrier();
3119 tgt_fini(env, &m->ofd_lut);
3120 ofd_stop_inconsistency_verification_thread(m);
3121 lfsck_degister(env, m->ofd_osd);
3122 ofd_fs_cleanup(env, m);
3123 nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
3124 obd->u.obt.obt_nodemap_config_file = NULL;
3126 if (m->ofd_los != NULL) {
3127 local_oid_storage_fini(env, m->ofd_los);
3131 if (m->ofd_namespace != NULL) {
3132 ldlm_namespace_free_post(m->ofd_namespace);
3133 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3136 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3138 LASSERT(atomic_read(&d->ld_ref) == 0);
3139 server_put_mount(obd->obd_name, true);
3144 * Implementation of lu_device_type_operations::ldto_device_fini.
3146 * Finalize device. Dual to ofd_device_init(). It is called from
3147 * obd_precleanup() and stops the current device.
3149 * \param[in] env execution environment
3150 * \param[in] d LU device of OFD
3154 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3155 struct lu_device *d)
3158 ofd_fini(env, ofd_dev(d));
3163 * Implementation of lu_device_type_operations::ldto_device_free.
3165 * Free OFD device. Dual to ofd_device_alloc().
3167 * \param[in] env execution environment
3168 * \param[in] d LU device of OFD
3172 static struct lu_device *ofd_device_free(const struct lu_env *env,
3173 struct lu_device *d)
3175 struct ofd_device *m = ofd_dev(d);
3177 dt_device_fini(&m->ofd_dt_dev);
3183 * Implementation of lu_device_type_operations::ldto_device_alloc.
3185 * This function allocates the new OFD device. It is called from
3186 * obd_setup() if OBD device had lu_device_type defined.
3188 * \param[in] env execution environment
3189 * \param[in] t lu_device_type of OFD device
3190 * \param[in] cfg configuration log
3192 * \retval pointer to the lu_device of just allocated OFD
3193 * \retval ERR_PTR of return value on error
3195 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3196 struct lu_device_type *t,
3197 struct lustre_cfg *cfg)
3199 struct ofd_device *m;
3200 struct lu_device *l;
3205 return ERR_PTR(-ENOMEM);
3207 l = &m->ofd_dt_dev.dd_lu_dev;
3208 dt_device_init(&m->ofd_dt_dev, t);
3209 rc = ofd_init0(env, m, t, cfg);
3211 ofd_device_free(env, l);
3218 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3219 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3221 static struct lu_device_type_operations ofd_device_type_ops = {
3222 .ldto_init = ofd_type_init,
3223 .ldto_fini = ofd_type_fini,
3225 .ldto_start = ofd_type_start,
3226 .ldto_stop = ofd_type_stop,
3228 .ldto_device_alloc = ofd_device_alloc,
3229 .ldto_device_free = ofd_device_free,
3230 .ldto_device_fini = ofd_device_fini
3233 static struct lu_device_type ofd_device_type = {
3234 .ldt_tags = LU_DEVICE_DT,
3235 .ldt_name = LUSTRE_OST_NAME,
3236 .ldt_ops = &ofd_device_type_ops,
3237 .ldt_ctx_tags = LCT_DT_THREAD
3241 * Initialize OFD module.
3243 * This function is called upon module loading. It registers OFD device type
3244 * and prepares all in-memory structures used by all OFD devices.
3246 * \retval 0 if successful
3247 * \retval negative value on error
3249 static int __init ofd_init(void)
3253 rc = lu_kmem_init(ofd_caches);
3257 rc = ofd_fmd_init();
3259 lu_kmem_fini(ofd_caches);
3263 rc = ofd_dlm_init();
3265 lu_kmem_fini(ofd_caches);
3270 rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
3271 LUSTRE_OST_NAME, &ofd_device_type);
3278 * This function is called upon OFD module unloading.
3279 * It frees all related structures and unregisters OFD device type.
3281 static void __exit ofd_exit(void)
3285 lu_kmem_fini(ofd_caches);
3286 class_unregister_type(LUSTRE_OST_NAME);
3289 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3290 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3291 MODULE_VERSION(LUSTRE_VERSION_STRING);
3292 MODULE_LICENSE("GPL");
3294 module_init(ofd_init);
3295 module_exit(ofd_exit);