4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/ofd/ofd_dev.c
33 * This file contains OSD API methods for OBD Filter Device (OFD),
34 * request handlers and supplemental functions to set OFD up and clean it up.
36 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
37 * Author: Mike Pershin <mike.pershin@intel.com>
38 * Author: Johann Lombardi <johann.lombardi@intel.com>
41 * The OBD Filter Device (OFD) module belongs to the Object Storage
42 * Server stack and connects the RPC oriented Unified Target (TGT)
43 * layer (see lustre/include/lu_target.h) to the storage oriented OSD
44 * layer (see Documentation/osd-api.txt).
52 * OFD implements the LU and OBD device APIs and is responsible for:
54 * - Handling client requests (create, destroy, bulk IO, setattr,
55 * get_info, set_info, statfs) for the objects belonging to the OST
56 * (together with TGT).
58 * - Providing grant space management which allows clients to reserve
59 * disk space for data writeback. OFD tracks grants on global and
62 * - Handling object precreation requests from MDTs.
64 * - Operating the LDLM service that allows clients to maintain object
65 * data cache coherence.
68 #define DEBUG_SUBSYSTEM S_FILTER
70 #include <obd_class.h>
71 #include <obd_cksum.h>
72 #include <uapi/linux/lustre/lustre_param.h>
73 #include <lustre_fid.h>
74 #include <lustre_lfsck.h>
75 #include <lustre_dlm.h>
76 #include <lustre_quota.h>
77 #include <lustre_nodemap.h>
78 #include <lustre_log.h>
79 #include <linux/falloc.h>
81 #include "ofd_internal.h"
83 /* Slab for OFD object allocation */
84 static struct kmem_cache *ofd_object_kmem;
85 static struct lu_kmem_descr ofd_caches[] = {
87 .ckd_cache = &ofd_object_kmem,
88 .ckd_name = "ofd_obj",
89 .ckd_size = sizeof(struct ofd_object)
97 * Connect OFD to the next device in the stack.
99 * This function is used for device stack configuration and links OFD
100 * device with bottom OSD device.
102 * \param[in] env execution environment
103 * \param[in] m OFD device
104 * \param[in] next name of next device in the stack
105 * \param[out] exp export to return
107 * \retval 0 and export in \a exp if successful
108 * \retval negative value on error
110 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
111 const char *next, struct obd_export **exp)
113 struct obd_connect_data *data = NULL;
114 struct obd_device *obd;
120 GOTO(out, rc = -ENOMEM);
122 obd = class_name2obd(next);
124 CERROR("%s: can't locate next device: %s\n",
126 GOTO(out, rc = -ENOTCONN);
129 data->ocd_connect_flags = OBD_CONNECT_VERSION;
130 data->ocd_version = LUSTRE_VERSION_CODE;
132 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
134 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
135 ofd_name(m), next, rc);
139 m->ofd_dt_dev.dd_lu_dev.ld_site =
140 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
141 LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
142 m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
143 m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
152 * Initialize stack of devices.
154 * This function initializes OFD-OSD device stack to serve OST requests
156 * \param[in] env execution environment
157 * \param[in] m OFD device
158 * \param[in] cfg Lustre config for this server
160 * \retval 0 if successful
161 * \retval negative value on error
163 static int ofd_stack_init(const struct lu_env *env,
164 struct ofd_device *m, struct lustre_cfg *cfg,
165 unsigned long *lmd_flags)
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;
185 if (test_bit(LMD_FLG_SKIP_LFSCK, lmd->lmd_flags))
186 m->ofd_skip_lfsck = 1;
187 if (test_bit(LMD_FLG_NO_CREATE, lmd->lmd_flags))
188 m->ofd_lut.lut_no_create = 1;
189 bitmap_copy(lmd_flags, lmd->lmd_flags, LMD_FLG_NUM_FLAGS);
192 /* find bottom osd */
193 OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
197 snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
198 rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
199 OBD_FREE(osdname, MTI_NAME_MAXLEN);
203 d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
205 m->ofd_osd = lu2dt_dev(d);
206 if (m->ofd_osd->dd_rdonly)
207 ofd_obd(m)->obd_read_only = 1;
209 snprintf(info->fti_u.name, sizeof(info->fti_u.name),
210 "%s-osd", lustre_cfg_string(cfg, 0));
216 * Finalize the device stack OFD-OSD.
218 * This function cleans OFD-OSD device stack and
219 * disconnects OFD from the OSD.
221 * \param[in] env execution environment
222 * \param[in] m OFD device
223 * \param[in] top top device of stack
225 * \retval 0 if successful
226 * \retval negative value on error
228 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
229 struct lu_device *top)
231 struct obd_device *obd = ofd_obd(m);
232 struct lustre_cfg_bufs bufs;
233 struct lustre_cfg *lcfg;
238 lu_site_purge(env, top->ld_site, ~0);
239 /* process cleanup, pass mdt obd name to get obd umount flags */
240 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
245 lustre_cfg_bufs_set_string(&bufs, 1, flags);
246 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
249 lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs);
252 top->ld_ops->ldo_process_config(env, top, lcfg);
253 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
255 if (m->ofd_los != NULL) {
256 local_oid_storage_fini(env, m->ofd_los);
260 lu_site_purge(env, top->ld_site, ~0);
261 lu_site_print(env, top->ld_site, &top->ld_site->ls_obj_hash.nelems,
262 D_OTHER, lu_cdebug_printer);
263 LASSERT(m->ofd_osd_exp);
264 obd_disconnect(m->ofd_osd_exp);
269 static void ofd_stack_pre_fini(const struct lu_env *env, struct ofd_device *m,
270 struct lu_device *top)
272 struct lustre_cfg_bufs bufs;
273 struct lustre_cfg *lcfg;
278 lustre_cfg_bufs_reset(&bufs, ofd_name(m));
279 lustre_cfg_bufs_set_string(&bufs, 1, NULL);
280 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
282 CERROR("%s: failed to trigger LCFG_PRE_CLEANUP\n", ofd_name(m));
284 lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, &bufs);
285 top->ld_ops->ldo_process_config(env, top, lcfg);
286 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
287 lcfg->lcfg_buflens));
293 /* For interoperability, see mdt_interop_param[]. */
294 static struct cfg_interop_param ofd_interop_param[] = {
295 { "ost.quota_type", NULL },
300 * Check if parameters are symlinks to the OSD.
302 * Some parameters were moved from ofd to osd and only their
303 * symlinks were kept in ofd by LU-3106. They are:
304 * -writehthrough_cache_enable
305 * -readcache_max_filesize
309 * Since they are not included by the static lprocfs var list, a pre-check
310 * is added for them to avoid "unknown param" errors. If they are matched
311 * in this check, they will be passed to the OSD directly.
313 * \param[in] param parameters to check
315 * \retval true if param is symlink to OSD param
318 static bool match_symlink_param(char *param)
323 if (class_match_param(param, PARAM_OST, ¶m) == 0) {
324 sval = strchr(param, '=');
326 paramlen = sval - param;
327 if (strncmp(param, "brw_stats", paramlen) == 0)
336 * Process various configuration parameters.
338 * This function is used by MGS to process specific configurations and
339 * pass them through to the next device in server stack, i.e. the OSD.
341 * \param[in] env execution environment
342 * \param[in] d LU device of OFD
343 * \param[in] cfg parameters to process
345 * \retval 0 if successful
346 * \retval negative value on error
348 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
349 struct lustre_cfg *cfg)
351 struct ofd_device *m = ofd_dev(d);
352 struct dt_device *dt_next = m->ofd_osd;
353 struct lu_device *next = &dt_next->dd_lu_dev;
358 switch (cfg->lcfg_command) {
360 /* For interoperability */
361 struct cfg_interop_param *ptr = NULL;
362 struct lustre_cfg *old_cfg = NULL;
366 param = lustre_cfg_string(cfg, 1);
368 CERROR("param is empty\n");
373 ptr = class_find_old_param(param, ofd_interop_param);
375 if (ptr->new_param == NULL) {
377 CWARN("For interoperability, skip this %s."
378 " It is obsolete.\n", ptr->old_param);
382 CWARN("Found old param %s, changed it to %s.\n",
383 ptr->old_param, ptr->new_param);
386 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
393 if (match_symlink_param(param)) {
394 rc = next->ld_ops->ldo_process_config(env, next, cfg);
398 count = class_modify_config(cfg, PARAM_OST,
399 &d->ld_obd->obd_kset.kobj);
404 CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
406 /* we don't understand; pass it on */
407 rc = next->ld_ops->ldo_process_config(env, next, cfg);
410 case LCFG_SPTLRPC_CONF: {
415 /* others are passed further */
416 rc = next->ld_ops->ldo_process_config(env, next, cfg);
423 * Implementation of lu_object_operations::loo_object_init for OFD
425 * Allocate just the next object (OSD) in stack.
427 * \param[in] env execution environment
428 * \param[in] o lu_object of OFD object
429 * \param[in] conf additional configuration parameters, not used here
431 * \retval 0 if successful
432 * \retval negative value on error
434 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
435 const struct lu_object_conf *conf)
437 struct ofd_device *d = ofd_dev(o->lo_dev);
438 struct lu_device *under;
439 struct lu_object *below;
444 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
445 PFID(lu_object_fid(o)));
447 under = &d->ofd_osd->dd_lu_dev;
448 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
450 lu_object_add(o, below);
457 static void ofd_object_free_rcu(struct rcu_head *head)
459 struct ofd_object *of = container_of(head, struct ofd_object,
462 kmem_cache_free(ofd_object_kmem, of);
466 * Implementation of lu_object_operations::loo_object_free.
468 * Finish OFD object lifecycle and free its memory.
470 * \param[in] env execution environment
471 * \param[in] o LU object of OFD object
473 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
475 struct ofd_object *of = ofd_obj(o);
476 struct lu_object_header *h;
481 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
482 PFID(lu_object_fid(o)));
485 lu_object_header_fini(h);
486 OBD_FREE_PRE(of, sizeof(*of), "slab-freed");
487 call_rcu(&of->ofo_header.loh_rcu, ofd_object_free_rcu);
492 * Implementation of lu_object_operations::loo_object_print.
494 * Print OFD part of compound OFD-OSD object. See lu_object_print() and
495 * LU_OBJECT_DEBUG() for more details about the compound object printing.
497 * \param[in] env execution environment
498 * \param[in] cookie opaque data passed to the printer function
499 * \param[in] p printer function to use
500 * \param[in] o LU object of OFD object
502 * \retval 0 if successful
503 * \retval negative value on error
505 static int ofd_object_print(const struct lu_env *env, void *cookie,
506 lu_printer_t p, const struct lu_object *o)
508 return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
511 static const struct lu_object_operations ofd_obj_ops = {
512 .loo_object_init = ofd_object_init,
513 .loo_object_free = ofd_object_free,
514 .loo_object_print = ofd_object_print
518 * Implementation of lu_device_operations::lod_object_alloc.
520 * This function allocates OFD part of compound OFD-OSD object and
521 * initializes its header, because OFD is the top device in stack
523 * \param[in] env execution environment
524 * \param[in] hdr object header, NULL for OFD
525 * \param[in] d lu_device
527 * \retval allocated object if successful
528 * \retval NULL value on failed allocation
530 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
531 const struct lu_object_header *hdr,
534 struct ofd_object *of;
538 OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
541 struct lu_object_header *h;
543 o = &of->ofo_obj.do_lu;
545 lu_object_header_init(h);
546 lu_object_init(o, h, d);
547 lu_object_add_top(h, o);
548 o->lo_ops = &ofd_obj_ops;
556 * Return the result of LFSCK run to the OFD.
558 * Notify OFD about result of LFSCK run. That may block the new object
559 * creation until problem is fixed by LFSCK.
561 * \param[in] env execution environment
562 * \param[in] data pointer to the OFD device
563 * \param[in] event LFSCK event type
565 * \retval 0 if successful
566 * \retval negative value on unknown event
568 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
569 enum lfsck_events event)
571 struct ofd_device *ofd = data;
572 struct obd_device *obd = ofd_obd(ofd);
575 case LE_LASTID_REBUILDING:
576 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
577 "on the device until the LAST_ID rebuilt successfully.\n",
579 down_write(&ofd->ofd_lastid_rwsem);
580 ofd->ofd_lastid_rebuilding = 1;
581 up_write(&ofd->ofd_lastid_rwsem);
583 case LE_LASTID_REBUILT: {
584 down_write(&ofd->ofd_lastid_rwsem);
585 ofd_seqs_free(env, ofd);
586 ofd->ofd_lastid_rebuilding = 0;
587 ofd->ofd_lastid_gen++;
588 up_write(&ofd->ofd_lastid_rwsem);
589 CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
594 CERROR("%s: unknown lfsck event: rc = %d\n",
595 ofd_name(ofd), event);
603 * Implementation of lu_device_operations::ldo_prepare.
605 * This method is called after layer has been initialized and before it starts
606 * serving user requests. In OFD it starts lfsk check routines and initializes
609 * \param[in] env execution environment
610 * \param[in] pdev higher device in stack, NULL for OFD
611 * \param[in] dev lu_device of OFD device
613 * \retval 0 if successful
614 * \retval negative value on error
616 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
617 struct lu_device *dev)
619 struct ofd_thread_info *info;
620 struct ofd_device *ofd = ofd_dev(dev);
621 struct obd_device *obd = ofd_obd(ofd);
622 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
627 info = ofd_info_init(env, NULL);
631 /* initialize lower device */
632 rc = next->ld_ops->ldo_prepare(env, dev, next);
636 rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
637 ofd_lfsck_out_notify, ofd, false);
639 CERROR("%s: failed to initialize lfsck: rc = %d\n",
644 rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
645 /* The LFSCK instance is registered just now, so it must be there when
646 * register the namespace to such instance. */
647 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
649 target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
650 CFS_FAIL_TIMEOUT_ORSET(OBD_FAIL_OST_PREPARE_DELAY, CFS_FAIL_ONCE,
651 (OBD_TIMEOUT_DEFAULT + 1) / 4);
652 LASSERT(obd->obd_no_conn);
653 spin_lock(&obd->obd_dev_lock);
654 obd->obd_no_conn = 0;
655 spin_unlock(&obd->obd_dev_lock);
657 if (obd->obd_recovering == 0)
658 ofd_postrecov(env, ofd);
664 * Implementation of lu_device_operations::ldo_recovery_complete.
666 * This method notifies all layers about 'recovery complete' event. That means
667 * device is in full state and consistent. An OFD calculates available grant
668 * space upon this event.
670 * \param[in] env execution environment
671 * \param[in] dev lu_device of OFD device
673 * \retval 0 if successful
674 * \retval negative value on error
676 static int ofd_recovery_complete(const struct lu_env *env,
677 struct lu_device *dev)
679 struct ofd_thread_info *oti = ofd_info(env);
680 struct ofd_device *ofd = ofd_dev(dev);
681 struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
687 * Grant space for object precreation on the self export.
688 * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
689 * is enough to create 10k objects. More space is then acquired for
690 * precreation in tgt_grant_create().
692 memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd));
693 oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2;
694 oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace;
695 oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT |
696 OBD_CONNECT_GRANT_PARAM;
697 tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd,
699 rc = next->ld_ops->ldo_recovery_complete(env, next);
704 * lu_device_operations matrix for OFD device.
706 static const struct lu_device_operations ofd_lu_ops = {
707 .ldo_object_alloc = ofd_object_alloc,
708 .ldo_process_config = ofd_process_config,
709 .ldo_recovery_complete = ofd_recovery_complete,
710 .ldo_prepare = ofd_prepare,
714 * Cleanup all procfs entries in OFD.
716 * \param[in] ofd OFD device
718 static void ofd_procfs_fini(struct ofd_device *ofd)
720 struct obd_device *obd = ofd_obd(ofd);
722 tgt_tunables_fini(&ofd->ofd_lut);
723 lprocfs_free_per_client_stats(obd);
724 lprocfs_obd_cleanup(obd);
725 lprocfs_free_obd_stats(obd);
726 lprocfs_job_stats_fini(obd);
730 * Stop SEQ/FID server on OFD.
732 * \param[in] env execution environment
733 * \param[in] ofd OFD device
735 * \retval 0 if successful
736 * \retval negative value on error
738 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
740 return seq_site_fini(env, &ofd->ofd_seq_site);
744 * Start SEQ/FID server on OFD.
746 * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
747 * It also connects to the master server to get own FID sequence (SEQ) range
748 * to this particular OFD. Typically that happens when the OST is first
749 * formatted or in the rare case that it exhausts the local sequence range.
751 * The sequence range is allocated out to the MDTs for OST object allocations,
752 * and not directly to the clients.
754 * \param[in] env execution environment
755 * \param[in] ofd OFD device
757 * \retval 0 if successful
758 * \retval negative value on error
760 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
762 struct seq_server_site *ss = &ofd->ofd_seq_site;
763 struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev;
764 char *obd_name = ofd_name(ofd);
766 int len = strlen(obd_name) + 7;
769 ss = &ofd->ofd_seq_site;
770 lu->ld_site->ld_seq_site = ss;
771 ss->ss_lu = lu->ld_site;
772 ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
774 OBD_ALLOC(name, len);
778 OBD_ALLOC_PTR(ss->ss_server_seq);
779 if (ss->ss_server_seq == NULL)
780 GOTO(out_name, rc = -ENOMEM);
782 rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
783 LUSTRE_SEQ_SERVER, ss);
785 CERROR("%s: seq server init error: rc = %d\n", obd_name, rc);
786 GOTO(out_server, rc);
788 ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
790 OBD_ALLOC_PTR(ss->ss_client_seq);
791 if (ss->ss_client_seq == NULL)
792 GOTO(out_server, rc = -ENOMEM);
794 snprintf(name, len, "%s-super", obd_name);
795 seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
798 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
801 seq_client_fini(ss->ss_client_seq);
802 OBD_FREE_PTR(ss->ss_client_seq);
803 ss->ss_client_seq = NULL;
805 seq_server_fini(ss->ss_server_seq, env);
806 OBD_FREE_PTR(ss->ss_server_seq);
807 ss->ss_server_seq = NULL;
816 * OFD request handler for OST_SET_INFO RPC.
818 * This is OFD-specific part of request handling
820 * \param[in] tsi target session environment for this request
822 * \retval 0 if successful
823 * \retval negative value on error
825 static int ofd_set_info_hdl(struct tgt_session_info *tsi)
827 struct ptlrpc_request *req = tgt_ses_req(tsi);
828 struct ost_body *body = NULL, *repbody;
829 void *key, *val = NULL;
830 int keylen, vallen, rc = 0;
831 bool is_grant_shrink;
832 ktime_t kstart = ktime_get();
836 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
838 DEBUG_REQ(D_HA, req, "no set_info key");
839 RETURN(err_serious(-EFAULT));
841 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
844 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
846 DEBUG_REQ(D_HA, req, "no set_info val");
847 RETURN(err_serious(-EFAULT));
849 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
852 is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
854 /* In this case the value is actually an RMF_OST_BODY, so we
855 * transmutate the type of this PTLRPC */
856 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
858 rc = req_capsule_server_pack(tsi->tsi_pill);
862 if (is_grant_shrink) {
863 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
866 * Because we already sync grant info with client when
867 * reconnect, grant info will be cleared for resent
868 * req, otherwise, outdated grant count in the rpc
869 * would de-sync grant counters
871 if (lustre_msg_get_flags(req->rq_reqmsg) &
872 (MSG_RESENT | MSG_REPLAY)) {
873 DEBUG_REQ(D_CACHE, req,
874 "clear resent/replay req grant info");
875 body->oa.o_valid &= ~OBD_MD_FLGRANT;
878 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
881 /** handle grant shrink, similar to a read request */
882 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
884 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
886 obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
889 CERROR("%s: Unsupported key %s\n",
890 tgt_name(tsi->tsi_tgt), (char *)key);
893 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
894 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
900 * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
902 * This function returns a list of extents which describes how a file's
903 * blocks are laid out on the disk.
905 * \param[in] env execution environment
906 * \param[in] ofd OFD device
907 * \param[in] fid FID of object
908 * \param[in] fiemap fiemap structure to fill with data
910 * \retval 0 if \a fiemap is filled with data successfully
911 * \retval negative value on error
913 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
914 struct lu_fid *fid, struct fiemap *fiemap)
916 struct ofd_object *fo;
919 fo = ofd_object_find(env, ofd, fid);
921 CERROR("%s: error finding object "DFID"\n",
922 ofd_name(ofd), PFID(fid));
926 ofd_read_lock(env, fo);
927 if (ofd_object_exists(fo))
928 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
931 ofd_read_unlock(env, fo);
932 ofd_object_put(env, fo);
937 static int ofd_lock_unlock_region(const struct lu_env *env,
938 struct ldlm_namespace *ns,
939 struct ldlm_res_id *res_id,
940 unsigned long long begin,
941 unsigned long long end)
945 struct lustre_handle lh = { 0 };
947 LASSERT(begin <= end);
949 rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags);
953 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
954 tgt_data_unlock(&lh, LCK_PR);
960 * Lock the sparse areas of given resource.
962 * The locking of sparse areas will cause dirty data to be flushed back from
963 * clients. This is used when getting the FIEMAP of an object to make sure
964 * there is no unaccounted cached data on clients.
966 * This function goes through \a fiemap list of extents and locks only sparse
967 * areas between extents.
969 * \param[in] ns LDLM namespace
970 * \param[in] res_id resource ID
971 * \param[in] fiemap file extents mapping on disk
972 * \param[in] locked list head of regions list
974 * \retval 0 if successful
975 * \retval negative value on error
977 static int lock_zero_regions(const struct lu_env *env,
978 struct ldlm_namespace *ns,
979 struct ldlm_res_id *res_id,
980 struct fiemap *fiemap)
982 __u64 begin = fiemap->fm_start;
985 struct fiemap_extent *fiemap_start = fiemap->fm_extents;
989 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
990 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
991 if (fiemap_start[i].fe_logical > begin) {
992 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
993 begin, fiemap_start[i].fe_logical);
994 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
995 fiemap_start[i].fe_logical);
1000 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1003 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1004 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1005 begin, fiemap->fm_start + fiemap->fm_length);
1006 rc = ofd_lock_unlock_region(env, ns, res_id, begin,
1007 fiemap->fm_start + fiemap->fm_length);
1015 * OFD request handler for OST_GET_INFO RPC.
1017 * This is OFD-specific part of request handling. The OFD-specific keys are:
1018 * - KEY_LAST_ID (obsolete)
1022 * This function reads needed data from storage and fills reply with it.
1024 * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
1025 * and is kept for compatibility.
1027 * \param[in] tsi target session environment for this request
1029 * \retval 0 if successful
1030 * \retval negative value on error
1032 static int ofd_get_info_hdl(struct tgt_session_info *tsi)
1034 struct obd_export *exp = tsi->tsi_exp;
1035 struct ofd_device *ofd = ofd_exp(exp);
1036 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1039 int replylen, rc = 0;
1040 ktime_t kstart = ktime_get();
1044 /* this common part for get_info rpc */
1045 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
1047 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
1048 RETURN(err_serious(-EPROTO));
1050 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
1053 if (KEY_IS(KEY_LAST_ID)) {
1055 struct ofd_seq *oseq;
1057 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
1058 rc = req_capsule_server_pack(tsi->tsi_pill);
1060 RETURN(err_serious(rc));
1062 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
1064 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1065 (u64)exp->exp_filter_data.fed_group);
1069 *last_id = ofd_seq_last_oid(oseq);
1070 ofd_seq_put(tsi->tsi_env, oseq);
1071 } else if (KEY_IS(KEY_FIEMAP)) {
1072 struct ll_fiemap_info_key *fm_key;
1073 struct fiemap *fiemap;
1076 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
1078 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
1079 rc = tgt_validate_obdo(tsi, &fm_key->lfik_oa);
1081 RETURN(err_serious(rc));
1083 fid = &fm_key->lfik_oa.o_oi.oi_fid;
1085 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
1087 replylen = fiemap_count_to_size(
1088 fm_key->lfik_fiemap.fm_extent_count);
1089 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
1090 RCL_SERVER, replylen);
1092 rc = req_capsule_server_pack(tsi->tsi_pill);
1094 RETURN(err_serious(rc));
1096 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
1100 *fiemap = fm_key->lfik_fiemap;
1101 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
1103 /* LU-3219: Lock the sparse areas to make sure dirty
1104 * flushed back from client, then call fiemap again. */
1105 if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
1106 fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
1107 ost_fid_build_resid(fid, &fti->fti_resid);
1108 rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace,
1109 &fti->fti_resid, fiemap);
1111 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
1114 } else if (KEY_IS(KEY_LAST_FID)) {
1115 struct ofd_device *ofd = ofd_exp(exp);
1116 struct ofd_seq *oseq;
1120 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
1121 rc = req_capsule_server_pack(tsi->tsi_pill);
1123 RETURN(err_serious(rc));
1125 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
1127 RETURN(err_serious(-EPROTO));
1129 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
1131 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
1135 oseq = ofd_seq_load(tsi->tsi_env, ofd,
1136 ostid_seq(&fti->fti_ostid));
1140 rc = ostid_to_fid(fid, &oseq->os_oi,
1141 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1143 GOTO(out_put, rc = -EFAULT);
1145 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
1148 ofd_seq_put(tsi->tsi_env, oseq);
1150 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
1154 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
1155 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1161 * OFD request handler for OST_GETATTR RPC.
1163 * This is OFD-specific part of request handling. It finds the OFD object
1164 * by its FID, gets attributes from storage and packs result to the reply.
1166 * \param[in] tsi target session environment for this request
1168 * \retval 0 if successful
1169 * \retval negative value on error
1171 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
1173 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1174 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1175 struct ost_body *repbody;
1176 struct lustre_handle lh = { 0 };
1177 struct ofd_object *fo;
1179 enum ldlm_mode lock_mode = LCK_PR;
1180 ktime_t kstart = ktime_get();
1185 LASSERT(tsi->tsi_ost_body != NULL);
1187 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1188 if (repbody == NULL)
1191 repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1192 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1194 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1195 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1198 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1201 rc = tgt_extent_lock(tsi->tsi_env,
1202 tsi->tsi_tgt->lut_obd->obd_namespace,
1203 &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1209 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1211 GOTO(out, rc = PTR_ERR(fo));
1213 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1217 obdo_from_la(&repbody->oa, &fti->fti_attr,
1218 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1220 /* Store object version in reply */
1221 curr_version = dt_version_get(tsi->tsi_env,
1222 ofd_object_child(fo));
1223 if ((__s64)curr_version != -EOPNOTSUPP) {
1224 repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1225 repbody->oa.o_data_version = curr_version;
1228 if (fo->ofo_ff.ff_layout_version > 0) {
1229 repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
1230 repbody->oa.o_layout_version =
1231 fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
1233 CDEBUG(D_INODE, DFID": get layout version: %u\n",
1234 PFID(&tsi->tsi_fid),
1235 repbody->oa.o_layout_version);
1239 ofd_object_put(tsi->tsi_env, fo);
1242 tgt_data_unlock(&lh, lock_mode);
1244 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1245 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1247 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1248 repbody->oa.o_flags = OBD_FL_FLUSH;
1254 * OFD request handler for OST_SETATTR RPC.
1256 * This is OFD-specific part of request handling. It finds the OFD object
1257 * by its FID, sets attributes from request and packs result to the reply.
1259 * \param[in] tsi target session environment for this request
1261 * \retval 0 if successful
1262 * \retval negative value on error
1264 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1266 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1267 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1268 struct ost_body *body = tsi->tsi_ost_body;
1269 struct ost_body *repbody;
1270 struct ldlm_resource *res;
1271 struct ofd_object *fo;
1272 ktime_t kstart = ktime_get();
1277 LASSERT(body != NULL);
1279 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1280 if (repbody == NULL)
1283 repbody->oa.o_oi = body->oa.o_oi;
1284 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1286 /* This would be very bad - accidentally truncating a file when
1287 * changing the time or similar - bug 12203. */
1288 if (body->oa.o_valid & OBD_MD_FLSIZE &&
1289 body->oa.o_size != OBD_OBJECT_EOF) {
1290 static char mdsinum[48];
1292 if (body->oa.o_valid & OBD_MD_FLFID)
1293 snprintf(mdsinum, sizeof(mdsinum) - 1,
1294 "of parent "DFID, body->oa.o_parent_seq,
1295 body->oa.o_parent_oid, 0);
1299 CERROR("%s: setattr from %s is trying to truncate object "DFID
1300 " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1301 PFID(&tsi->tsi_fid), mdsinum);
1305 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1307 GOTO(out, rc = PTR_ERR(fo));
1309 la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1310 fti->fti_attr.la_valid &= ~LA_TYPE;
1312 /* setting objects attributes (including owner/group) */
1313 rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
1317 obdo_from_la(&repbody->oa, &fti->fti_attr,
1318 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1320 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1321 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1324 ofd_object_put(tsi->tsi_env, fo);
1327 /* we do not call this before to avoid lu_object_find() in
1328 * ->lvbo_update() holding another reference on the object.
1329 * otherwise concurrent destroy can make the object unavailable
1330 * for 2nd lu_object_find() waiting for the first reference
1331 * to go... deadlock! */
1332 res = ldlm_resource_get(ofd->ofd_namespace, &tsi->tsi_resid,
1335 ldlm_res_lvbo_update(res, NULL, 0);
1336 ldlm_resource_putref(res);
1343 * Destroy OST orphans.
1345 * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
1346 * set then we must destroy possible orphaned objects.
1348 * \param[in] env execution environment
1349 * \param[in] exp OBD export
1350 * \param[in] ofd OFD device
1351 * \param[in] oa obdo structure for reply
1353 * \retval 0 if successful
1354 * \retval negative value on error
1356 static int ofd_orphans_destroy(const struct lu_env *env,
1357 struct obd_export *exp,
1358 struct ofd_device *ofd, struct obdo *oa)
1360 struct ofd_thread_info *info = ofd_info(env);
1361 struct lu_fid *fid = &info->fti_fid;
1362 struct ost_id *oi = &oa->o_oi;
1363 struct ofd_seq *oseq;
1364 u64 seq = ostid_seq(oi);
1365 u64 end_id = ostid_id(oi);
1373 oseq = ofd_seq_get(ofd, seq);
1375 CERROR("%s: Can not find seq for "DOSTID"\n",
1376 ofd_name(ofd), POSTID(oi));
1381 last = ofd_seq_last_oid(oseq);
1384 LASSERT(exp != NULL);
1385 skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1387 if (CFS_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1390 LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1391 " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1393 while (oid > end_id) {
1394 rc = fid_set_id(fid, oid);
1395 if (unlikely(rc != 0))
1398 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1399 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1400 likely(rc != -EREMCHG && rc != -EINPROGRESS))
1401 /* this is pretty fatal... */
1402 CEMERG("%s: error destroying precreated id "
1404 ofd_name(ofd), PFID(fid), rc);
1408 ofd_seq_last_oid_set(oseq, oid);
1409 /* update last_id on disk periodically so that if we
1410 * restart * we don't need to re-scan all of the just
1411 * deleted objects. */
1412 if ((oid & 511) == 0)
1413 ofd_seq_last_oid_write(env, ofd, oseq);
1417 CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1418 ofd_name(ofd), seq, oid);
1422 ofd_seq_last_oid_set(oseq, oid);
1423 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1425 /* don't reuse orphan object, return last used objid */
1426 rc = ostid_set_id(oi, last);
1432 ofd_seq_put(env, oseq);
1437 * OFD request handler for OST_CREATE RPC.
1439 * This is OFD-specific part of request handling. Its main purpose is to
1440 * create new data objects on OST, but it also used to destroy orphans.
1442 * \param[in] tsi target session environment for this request
1444 * \retval 0 if successful
1445 * \retval negative value on error
1447 static int ofd_create_hdl(struct tgt_session_info *tsi)
1449 struct ptlrpc_request *req = tgt_ses_req(tsi);
1450 struct ost_body *repbody;
1451 const struct obdo *oa = &tsi->tsi_ost_body->oa;
1452 struct obdo *rep_oa;
1453 struct obd_export *exp = tsi->tsi_exp;
1454 struct ofd_device *ofd = ofd_exp(exp);
1455 struct seq_server_site *ss = &ofd->ofd_seq_site;
1456 __u64 seq_width = ss->ss_client_seq->lcs_width;
1457 u64 seq = ostid_seq(&oa->o_oi);
1458 u64 oid = ostid_id(&oa->o_oi);
1459 struct ofd_seq *oseq;
1462 ktime_t kstart = ktime_get();
1468 if (CFS_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1471 if (ofd->ofd_lut.lut_no_create)
1474 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1475 if (repbody == NULL)
1478 down_read(&ofd->ofd_lastid_rwsem);
1479 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1480 * we may do that in the future.
1481 * Return -ENOSPC until the LAST_ID rebuilt. */
1482 if (unlikely(ofd->ofd_lastid_rebuilding))
1483 GOTO(out_sem, rc = -ENOSPC);
1485 rep_oa = &repbody->oa;
1486 rep_oa->o_oi = oa->o_oi;
1487 rep_oa->o_valid |= OBD_MD_FLSIZE;
1488 rep_oa->o_size = seq_width;
1490 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1492 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1494 oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1496 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1497 ofd_name(ofd), seq, PTR_ERR(oseq));
1498 GOTO(out_sem, rc = -EINVAL);
1501 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1502 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1503 if (!ofd_obd(ofd)->obd_recovering ||
1504 oid > ofd_seq_last_oid(oseq)) {
1505 CERROR("%s: recreate objid "DOSTID" > last id %llu"
1506 "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1507 ofd_seq_last_oid(oseq));
1508 GOTO(out_nolock, rc = -EINVAL);
1510 /* Do nothing here, we re-create objects during recovery
1511 * upon write replay, see ofd_preprw_write() */
1512 GOTO(out_nolock, rc = 0);
1514 /* former ofd_handle_precreate */
1515 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1516 (oa->o_flags & OBD_FL_DELORPHAN)) {
1517 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1519 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_DELORPHAN_DELAY, cfs_fail_val);
1520 /* destroy orphans */
1521 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1522 exp->exp_conn_cnt) {
1523 CERROR("%s: dropping old orphan cleanup request\n",
1525 GOTO(out_nolock, rc = -ESTALE);
1527 /* This causes inflight precreates to abort and drop lock */
1528 oseq->os_destroys_in_progress = 1;
1529 mutex_lock(&oseq->os_create_lock);
1530 if (!oseq->os_destroys_in_progress) {
1532 "%s:[%llu] destroys_in_progress already cleared\n",
1533 ofd_name(ofd), seq);
1534 rc = ostid_set_id(&rep_oa->o_oi,
1535 ofd_seq_last_oid(oseq));
1538 diff = oid - ofd_seq_last_oid(oseq);
1539 CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n",
1540 ofd_seq_last_oid(oseq), diff);
1541 if (-diff > OST_MAX_PRECREATE) {
1542 LCONSOLE(D_INFO, "%s: too large difference between MDS "
1543 "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" "
1544 "(%llu), trust the OST\n",
1545 ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid,
1546 PFID(&oseq->os_oi.oi_fid),
1547 ofd_seq_last_oid(oseq));
1549 /* Let MDS know that we are so far ahead. */
1550 rc = ostid_set_id(&rep_oa->o_oi,
1551 ofd_seq_last_oid(oseq) + 1);
1552 } else if (diff < 0) {
1553 rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1555 oseq->os_destroys_in_progress = 0;
1557 /* XXX: Used by MDS for the first time! */
1558 oseq->os_destroys_in_progress = 0;
1561 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1562 ofd->ofd_lastid_gen)) {
1563 /* Keep the export ref so we can send the reply. */
1564 ofd_obd_disconnect(class_export_get(exp));
1565 GOTO(out_nolock, rc = -ENOTCONN);
1568 mutex_lock(&oseq->os_create_lock);
1569 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1570 exp->exp_conn_cnt) {
1571 CERROR("%s: dropping old precreate request\n",
1573 GOTO(out, rc = -ESTALE);
1575 /* only precreate if seq is 0, IDIF or normal and also o_id
1576 * must be specfied */
1577 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1578 !fid_seq_is_idif(seq)) || oid == 0) {
1579 diff = 1; /* shouldn't we create this right now? */
1581 diff = oid - ofd_seq_last_oid(oseq);
1582 /* Do sync create if the seq is about to used up */
1583 sync_trans = ofd_seq_is_exhausted(ofd, oa);
1585 GOTO(out, rc = sync_trans);
1587 if (diff <= -OST_MAX_PRECREATE) {
1589 CERROR("%s: invalid precreate request for "
1590 DOSTID", last_id %llu. "
1591 "Likely MDS last_id corruption\n",
1592 ofd_name(ofd), POSTID(&oa->o_oi),
1593 ofd_seq_last_oid(oseq));
1594 GOTO(out, rc = -EINVAL);
1595 } else if (diff < 0) {
1597 "%s: MDS LAST_ID "DFID" (%llu) is %lld behind OST LAST_ID "DFID" (%llu), trust the OST\n",
1598 ofd_name(ofd), PFID(&oa->o_oi.oi_fid),
1599 oid, -diff, PFID(&oseq->os_oi.oi_fid),
1600 ofd_seq_last_oid(oseq));
1601 /* Let MDS know that we are so far ahead. */
1602 rc = ostid_set_id(&rep_oa->o_oi,
1603 ofd_seq_last_oid(oseq) + 1);
1608 time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT;
1615 /* This can happen if a new OST is formatted and installed
1616 * in place of an old one at the same index. Instead of
1617 * precreating potentially millions of deleted old objects
1618 * (possibly filling the OST), only precreate the last batch.
1619 * LFSCK will eventually clean up any orphans. LU-14 */
1620 if (diff > 5 * OST_MAX_PRECREATE) {
1621 /* Message below is checked in conf-sanity test_122b */
1622 LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %lld higher than LAST_ID "DOSTID", only precreating the last %llu objects. OST replaced or reformatted?\n",
1623 ofd_name(ofd), POSTID(&oa->o_oi), diff,
1624 POSTID(&oseq->os_oi),
1625 min(seq_width, (__u64)OST_MAX_PRECREATE));
1626 /* From last created */
1627 diff = min(seq_width, (__u64)OST_MAX_PRECREATE);
1628 ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1629 /* no sync_trans when recreating last batch */
1633 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1634 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1635 /* don't enforce grant during orphan recovery */
1636 granted = tgt_grant_create(tsi->tsi_env,
1637 ofd_obd(ofd)->obd_self_export,
1642 CDEBUG(D_HA, "%s: failed to acquire grant "
1643 "space for precreate (%lld): rc = %d\n",
1644 ofd_name(ofd), diff, rc);
1649 trans_local = !exp_connect_replay_create(req->rq_export);
1651 next_id = ofd_seq_last_oid(oseq) + 1;
1652 count = ofd_precreate_batch(ofd, (int)diff);
1654 CDEBUG(D_HA, "%s: reserve %d objects in group %#llx"
1655 " at %llu\n", ofd_name(ofd),
1656 count, seq, next_id);
1658 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1659 && ktime_get_seconds() > enough_time) {
1660 CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects"
1661 " created at a rate of %d/s\n",
1662 ofd_name(ofd), created, diff + created,
1663 created / DISK_TIMEOUT);
1667 rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1668 oseq, count, sync_trans,
1673 } else if (rc < 0) {
1679 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1680 LCONSOLE_WARN("%s: can't create the same count of"
1681 " objects when replaying the request"
1682 " (diff is %lld). see LU-4621\n",
1683 ofd_name(ofd), diff);
1686 /* some objects got created, we can return
1687 * them, even if last creation failed */
1690 CERROR("%s: unable to precreate: rc = %d\n",
1693 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1694 !(oa->o_flags & OBD_FL_DELORPHAN)) {
1695 tgt_grant_commit(ofd_obd(ofd)->obd_self_export,
1700 rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1704 ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1705 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1706 if (unlikely(!oseq->os_last_id_synced))
1707 oseq->os_last_id_synced = 1;
1709 mutex_unlock(&oseq->os_create_lock);
1712 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1713 ofd_seq_put(tsi->tsi_env, oseq);
1716 up_read(&ofd->ofd_lastid_rwsem);
1721 * OFD request handler for OST_DESTROY RPC.
1723 * This is OFD-specific part of request handling. It destroys data objects
1724 * related to destroyed object on MDT.
1726 * \param[in] tsi target session environment for this request
1728 * \retval 0 if successful
1729 * \retval negative value on error
1731 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1733 const struct ost_body *body = tsi->tsi_ost_body;
1734 struct ost_body *repbody;
1735 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1736 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1737 struct lu_fid *fid = &fti->fti_fid;
1738 ktime_t kstart = ktime_get();
1745 if (CFS_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1748 /* This is old case for clients before Lustre 2.4 */
1749 /* If there's a DLM request, cancel the locks mentioned in it */
1750 if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1752 struct ldlm_request *dlm;
1754 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1757 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
1760 *fid = body->oa.o_oi.oi_fid;
1761 oid = ostid_id(&body->oa.o_oi);
1764 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1766 /* check that o_misc makes sense */
1767 if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1768 count = body->oa.o_misc;
1770 count = 1; /* default case - single destroy */
1772 CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1773 POSTID(&body->oa.o_oi), count);
1778 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1779 if (lrc == -ENOENT) {
1781 "%s: destroying non-existent object "DFID"\n",
1782 ofd_name(ofd), PFID(fid));
1783 /* rewrite rc with -ENOENT only if it is 0 */
1786 } else if (lrc != 0) {
1787 CERROR("%s: error destroying object "DFID": %d\n",
1788 ofd_name(ofd), PFID(fid), lrc);
1794 lrc = fid_set_id(fid, oid);
1795 if (unlikely(lrc != 0 && count > 0))
1796 GOTO(out, rc = lrc);
1799 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1800 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1805 fid_to_ostid(fid, &repbody->oa.o_oi);
1810 * OFD request handler for OST_STATFS RPC.
1812 * This function gets statfs data from storage as part of request
1815 * \param[in] tsi target session environment for this request
1817 * \retval 0 if successful
1818 * \retval negative value on error
1820 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1822 ktime_t kstart = ktime_get();
1823 struct obd_statfs *osfs;
1828 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
1830 osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1832 rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1833 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
1835 CERROR("%s: statfs failed: rc = %d\n",
1836 tgt_name(tsi->tsi_tgt), rc);
1838 if (CFS_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1841 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1842 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1848 * OFD request handler for OST_SYNC RPC.
1850 * Sync object data or all filesystem data to the disk and pack the
1853 * \param[in] tsi target session environment for this request
1855 * \retval 0 if successful
1856 * \retval negative value on error
1858 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1860 struct ost_body *body = tsi->tsi_ost_body;
1861 struct ost_body *repbody;
1862 struct ofd_thread_info *fti = tsi2ofd_info(tsi);
1863 struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
1864 struct ofd_object *fo = NULL;
1865 ktime_t kstart = ktime_get();
1870 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1872 /* if no objid is specified, it means "sync whole filesystem" */
1873 if (!fid_is_zero(&tsi->tsi_fid)) {
1874 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1876 RETURN(PTR_ERR(fo));
1879 rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1880 fo != NULL ? ofd_object_child(fo) : NULL,
1881 repbody->oa.o_size, repbody->oa.o_blocks);
1885 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1886 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
1890 repbody->oa.o_oi = body->oa.o_oi;
1891 repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1893 rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1895 obdo_from_la(&repbody->oa, &fti->fti_attr,
1898 /* don't return rc from getattr */
1903 ofd_object_put(tsi->tsi_env, fo);
1908 * OFD request handler for OST_FALLOCATE RPC.
1910 * This is part of request processing. Validate request fields,
1911 * preallocate the given OFD object and pack reply.
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_fallocate_hdl(struct tgt_session_info *tsi)
1920 struct obdo *oa = &tsi->tsi_ost_body->oa;
1921 struct ost_body *repbody;
1922 struct ofd_thread_info *info = tsi2ofd_info(tsi);
1923 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1924 struct ldlm_resource *res;
1925 struct ofd_object *fo;
1928 struct lustre_handle lh = { 0, };
1932 ktime_t kstart = ktime_get();
1934 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1935 if (repbody == NULL)
1936 RETURN(err_serious(-ENOMEM));
1939 * fallocate() start and end are passed in o_size and o_blocks
1940 * on the wire. Clients 2.15.0 and newer should always set
1941 * the OBD_MD_FLSIZE and OBD_MD_FLBLOCKS valid flags, but some
1942 * older client (exp_old_falloc is true) versions did not.
1943 * We permit older clients to not set these flags, checking their
1944 * version by proxy using the lack of OBD_CONNECT_TRUNCLOCK to
1945 * imply 2.14.0 and older.
1947 * Return -EOPNOTSUPP to also work with older clients not
1948 * supporting newer server modes.
1950 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1951 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)
1952 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 21, 53, 0)
1953 && !tsi->tsi_exp->exp_old_falloc
1956 RETURN(-EOPNOTSUPP);
1960 CDEBUG(D_INFO, "%s: start: %llu end: %llu\n", tgt_name(tsi->tsi_tgt),
1963 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 21, 53, 0)
1964 /* For inter-op case with older clients (where exp_old_falloc is true)
1965 * fallocate() start and end are passed in as 0 (For interior case
1966 * where end offset less than file size) This is fixed later.
1967 * For such cases we return -EOPNOTSUPP
1969 if (tsi->tsi_exp->exp_old_falloc && start >= end)
1970 RETURN(-EOPNOTSUPP);
1972 /* client should already limit len >= 0 */
1976 mode = oa->o_falloc_mode;
1978 * mode == 0 (which is standard prealloc) and PUNCH is supported
1979 * Rest of mode options are not supported yet.
1981 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1982 RETURN(-EOPNOTSUPP);
1984 /* PUNCH_HOLE mode should always be accompanied with KEEP_SIZE flag
1985 * Check that and add the missing flag for such invalid call with
1988 if (mode & FALLOC_FL_PUNCH_HOLE && !(mode & FALLOC_FL_KEEP_SIZE)) {
1989 CWARN("%s: PUNCH mode misses KEEP_SIZE flag, setting it\n",
1990 tsi->tsi_tgt->lut_obd->obd_name);
1991 mode |= FALLOC_FL_KEEP_SIZE;
1994 repbody->oa.o_oi = oa->o_oi;
1995 repbody->oa.o_valid = OBD_MD_FLID;
1997 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1998 oa->o_flags & OBD_FL_SRVLOCK;
2001 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid,
2002 start, end, &lh, LCK_PW, &flags);
2007 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2010 GOTO(out, rc = PTR_ERR(fo));
2012 valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID |
2013 OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME;
2014 la_from_obdo(&info->fti_attr, oa, valid);
2016 rc = ofd_object_fallocate(tsi->tsi_env, fo, start, end, mode,
2017 &info->fti_attr, oa);
2021 rc = ofd_attr_get(tsi->tsi_env, fo, &info->fti_attr);
2023 obdo_from_la(&repbody->oa, &info->fti_attr, OFD_VALID_FLAGS);
2027 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PREALLOC,
2028 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2032 ofd_object_put(tsi->tsi_env, fo);
2035 tgt_data_unlock(&lh, LCK_PW);
2037 res = ldlm_resource_get(ns, &tsi->tsi_resid, LDLM_EXTENT, 0);
2039 struct ost_lvb *res_lvb;
2041 ldlm_res_lvbo_update(res, NULL, 0);
2042 res_lvb = res->lr_lvb_data;
2044 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2045 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2047 repbody->oa.o_valid |= OBD_MD_FLSIZE;
2048 repbody->oa.o_size = res_lvb->lvb_size;
2050 ldlm_resource_putref(res);
2058 * OFD request handler for OST_PUNCH RPC.
2060 * This is part of request processing. Validate request fields,
2061 * punch (truncate) the given OFD object and pack reply.
2063 * \param[in] tsi target session environment for this request
2065 * \retval 0 if successful
2066 * \retval negative value on error
2068 static int ofd_punch_hdl(struct tgt_session_info *tsi)
2070 const struct obdo *oa = &tsi->tsi_ost_body->oa;
2071 struct ost_body *repbody;
2072 struct ofd_thread_info *info = tsi2ofd_info(tsi);
2073 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
2074 struct ldlm_resource *res;
2075 struct ofd_object *fo;
2077 struct lustre_handle lh = { 0, };
2080 ktime_t kstart = ktime_get();
2085 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val);
2087 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
2088 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
2089 RETURN(err_serious(-EPROTO));
2091 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2092 if (repbody == NULL)
2093 RETURN(err_serious(-ENOMEM));
2095 /* punch start,end are passed in o_size,o_blocks throught wire */
2099 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
2102 /* standard truncate optimization: if file body is completely
2103 * destroyed, don't send data back to the server. */
2105 flags |= LDLM_FL_AST_DISCARD_DATA;
2107 repbody->oa.o_oi = oa->o_oi;
2108 repbody->oa.o_valid = OBD_MD_FLID;
2110 srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
2111 oa->o_flags & OBD_FL_SRVLOCK;
2114 rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start,
2115 end, &lh, LCK_PW, &flags);
2120 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
2121 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
2122 oa->o_valid, start, end);
2124 fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
2127 GOTO(out, rc = PTR_ERR(fo));
2129 la_from_obdo(&info->fti_attr, oa,
2130 OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
2131 info->fti_attr.la_size = start;
2132 info->fti_attr.la_valid |= LA_SIZE;
2134 rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
2139 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
2140 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2143 ofd_object_put(tsi->tsi_env, fo);
2146 tgt_data_unlock(&lh, LCK_PW);
2148 /* we do not call this before to avoid lu_object_find() in
2149 * ->lvbo_update() holding another reference on the object.
2150 * otherwise concurrent destroy can make the object unavailable
2151 * for 2nd lu_object_find() waiting for the first reference
2152 * to go... deadlock! */
2153 res = ldlm_resource_get(ns, &tsi->tsi_resid, LDLM_EXTENT, 0);
2155 struct ost_lvb *res_lvb;
2157 ldlm_res_lvbo_update(res, NULL, 0);
2158 res_lvb = res->lr_lvb_data;
2159 repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
2160 repbody->oa.o_blocks = res_lvb->lvb_blocks;
2161 ldlm_resource_putref(res);
2167 static int ofd_ladvise_prefetch(const struct lu_env *env,
2168 struct ofd_object *fo,
2169 struct niobuf_local *lnb,
2170 __u64 start, __u64 end, enum dt_bufs_type dbt)
2172 struct ofd_thread_info *info = ofd_info(env);
2173 pgoff_t start_index, end_index, pages;
2174 struct niobuf_remote rnb;
2175 unsigned long nr_local;
2181 ofd_read_lock(env, fo);
2182 if (!ofd_object_exists(fo))
2183 GOTO(out_unlock, rc = -ENOENT);
2185 rc = ofd_attr_get(env, fo, &info->fti_attr);
2187 GOTO(out_unlock, rc);
2189 if (end > info->fti_attr.la_size)
2190 end = info->fti_attr.la_size;
2193 GOTO(out_unlock, rc);
2195 /* We need page aligned offset and length */
2196 start_index = start >> PAGE_SHIFT;
2197 end_index = (end - 1) >> PAGE_SHIFT;
2198 pages = end_index - start_index + 1;
2200 nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages :
2201 PTLRPC_MAX_BRW_PAGES;
2202 rnb.rnb_offset = start_index << PAGE_SHIFT;
2203 rnb.rnb_len = nr_local << PAGE_SHIFT;
2204 rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb,
2205 PTLRPC_MAX_BRW_PAGES, dbt);
2206 if (unlikely(rc < 0))
2209 rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local);
2210 dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local);
2213 start_index += nr_local;
2218 ofd_read_unlock(env, fo);
2223 * OFD request handler for OST_LADVISE RPC.
2225 * Tune cache or perfetch policies according to advices.
2227 * \param[in] tsi target session environment for this request
2229 * \retval 0 if successful
2230 * \retval negative errno on error
2232 static int ofd_ladvise_hdl(struct tgt_session_info *tsi)
2234 struct ptlrpc_request *req = tgt_ses_req(tsi);
2235 struct obd_export *exp = tsi->tsi_exp;
2236 struct ofd_device *ofd = ofd_exp(exp);
2237 struct ost_body *body, *repbody;
2238 struct ofd_thread_info *info;
2239 struct ofd_object *fo;
2240 struct ptlrpc_thread *svc_thread = req->rq_svc_thread;
2241 const struct lu_env *env = svc_thread->t_env;
2242 struct tgt_thread_big_cache *tbc = svc_thread->t_data;
2243 enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD;
2244 struct lu_ladvise *ladvise;
2246 struct ladvise_hdr *ladvise_hdr;
2247 struct obd_ioobj ioo;
2248 struct lustre_handle lockh = { 0 };
2251 struct dt_object *dob;
2257 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val);
2258 body = tsi->tsi_ost_body;
2260 if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID)
2261 RETURN(err_serious(-EPROTO));
2263 ladvise_hdr = req_capsule_client_get(tsi->tsi_pill,
2264 &RMF_OST_LADVISE_HDR);
2265 if (ladvise_hdr == NULL)
2266 RETURN(err_serious(-EPROTO));
2268 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2269 ladvise_hdr->lah_count < 1)
2270 RETURN(err_serious(-EPROTO));
2272 if ((ladvise_hdr->lah_flags & (~LF_MASK)) != 0)
2273 RETURN(err_serious(-EPROTO));
2275 ladvise = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_LADVISE);
2276 if (ladvise == NULL)
2277 RETURN(err_serious(-EPROTO));
2279 num_advise = req_capsule_get_size(&req->rq_pill,
2280 &RMF_OST_LADVISE, RCL_CLIENT) /
2282 if (num_advise < ladvise_hdr->lah_count)
2283 RETURN(err_serious(-EPROTO));
2285 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2286 repbody->oa = body->oa;
2288 info = ofd_info_init(env, exp);
2290 rc = ostid_to_fid(&info->fti_fid, &body->oa.o_oi,
2291 ofd->ofd_lut.lut_lsd.lsd_osd_index);
2295 fo = ofd_object_find(env, ofd, &info->fti_fid);
2300 LASSERT(fo != NULL);
2301 dob = ofd_object_child(fo);
2303 if (ptlrpc_connection_is_local(exp->exp_connection))
2304 dbt |= DT_BUFS_TYPE_LOCAL;
2306 for (i = 0; i < num_advise; i++, ladvise++) {
2307 start = ladvise->lla_start;
2308 end = ladvise->lla_end;
2310 rc = err_serious(-EPROTO);
2314 /* Handle different advice types */
2315 switch (ladvise->lla_advice) {
2319 case LU_LADVISE_WILLREAD:
2323 ioo.ioo_oid = body->oa.o_oi;
2325 rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace,
2326 &tsi->tsi_resid, start, end - 1,
2327 &lockh, LCK_PR, &flags);
2331 req->rq_status = ofd_ladvise_prefetch(env, fo,
2334 tgt_data_unlock(&lockh, LCK_PR);
2336 case LU_LADVISE_DONTNEED:
2337 rc = dt_ladvise(env, dob, ladvise->lla_start,
2338 ladvise->lla_end, LU_LADVISE_DONTNEED);
2345 ofd_object_put(env, fo);
2346 req->rq_status = rc;
2351 * OFD request handler for OST_QUOTACTL RPC.
2353 * This is part of request processing to validate incoming request fields,
2354 * get the requested data from OSD and pack reply.
2356 * \param[in] tsi target session environment for this request
2358 * \retval 0 if successful
2359 * \retval negative value on error
2361 static int ofd_quotactl(struct tgt_session_info *tsi)
2363 struct obd_quotactl *oqctl, *repoqc;
2364 struct lu_nodemap *nodemap;
2365 ktime_t kstart = ktime_get();
2366 char *buffer = NULL;
2371 oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2373 RETURN(err_serious(-EPROTO));
2375 if (oqctl->qc_cmd == LUSTRE_Q_ITEROQUOTA)
2376 req_capsule_set_size(tsi->tsi_pill, &RMF_OBD_QUOTA_ITER,
2377 RCL_SERVER, LQUOTA_ITER_BUFLEN);
2379 req_capsule_set_size(tsi->tsi_pill, &RMF_OBD_QUOTA_ITER,
2382 rc = req_capsule_server_pack(tsi->tsi_pill);
2384 RETURN(err_serious(rc));
2386 repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
2388 RETURN(err_serious(-ENOMEM));
2391 if (oqctl->qc_cmd == LUSTRE_Q_ITEROQUOTA) {
2392 buffer = req_capsule_server_get(tsi->tsi_pill,
2393 &RMF_OBD_QUOTA_ITER);
2395 RETURN(err_serious(-ENOMEM));
2398 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
2399 if (IS_ERR(nodemap))
2400 RETURN(PTR_ERR(nodemap));
2403 if (oqctl->qc_type == USRQUOTA)
2404 id = nodemap_map_id(nodemap, NODEMAP_UID,
2405 NODEMAP_CLIENT_TO_FS,
2407 else if (oqctl->qc_type == GRPQUOTA)
2408 id = nodemap_map_id(nodemap, NODEMAP_GID,
2409 NODEMAP_CLIENT_TO_FS,
2411 else if (oqctl->qc_type == PRJQUOTA)
2412 id = nodemap_map_id(nodemap, NODEMAP_PROJID,
2413 NODEMAP_CLIENT_TO_FS,
2416 nodemap_putref(nodemap);
2418 if (repoqc->qc_id != id)
2419 swap(repoqc->qc_id, id);
2421 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc,
2422 buffer, buffer == NULL ? 0 : LQUOTA_ITER_BUFLEN);
2424 ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
2425 tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart));
2427 if (repoqc->qc_id != id)
2428 swap(repoqc->qc_id, id);
2434 * Prolong lock timeout for the given extent.
2436 * This function finds all locks related with incoming request and
2437 * prolongs their timeout.
2439 * If a client is holding a lock for a long time while it sends
2440 * read or write RPCs to the OST for the object under this lock,
2441 * then we don't want the OST to evict the client. Otherwise,
2442 * if the network or disk is very busy then the client may not
2443 * be able to make any progress to clear out dirty pages under
2444 * the lock and the application will fail.
2446 * Every time a Bulk Read/Write (BRW) request arrives for the object
2447 * covered by the lock, extend the timeout on that lock. The RPC should
2448 * contain a lock handle for the lock it is using, but this
2449 * isn't handled correctly by all client versions, and the
2450 * request may cover multiple locks.
2452 * \param[in] tsi target session environment for this request
2453 * \param[in] data struct of data to prolong locks
2456 static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
2457 struct ldlm_prolong_args *data)
2459 struct obdo *oa = &tsi->tsi_ost_body->oa;
2460 struct ldlm_lock *lock;
2464 data->lpa_req = tgt_ses_req(tsi);
2465 data->lpa_export = tsi->tsi_exp;
2466 data->lpa_resid = tsi->tsi_resid;
2468 CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu"
2469 " ext(%llu->%llu)\n", tgt_ses_req(tsi),
2470 tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start,
2471 data->lpa_extent.end);
2473 if (oa->o_valid & OBD_MD_FLHANDLE) {
2474 /* mostly a request should be covered by only one lock, try
2476 lock = ldlm_handle2lock(&oa->o_handle);
2478 /* Fast path to check if the lock covers the whole IO
2479 * region exclusively. */
2480 if (ldlm_extent_contain(&lock->l_policy_data.l_extent,
2481 &data->lpa_extent)) {
2483 LASSERT(lock->l_export == data->lpa_export);
2484 ldlm_lock_prolong_one(lock, data);
2485 LDLM_LOCK_PUT(lock);
2486 if (data->lpa_locks_cnt > 0)
2488 /* The lock was destroyed probably lets try
2491 lock->l_last_used = ktime_get();
2492 LDLM_LOCK_PUT(lock);
2497 ldlm_resource_prolong(data);
2502 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
2504 * Determine if \a lock and the lock from request \a req are equivalent
2505 * by comparing their resource names, modes, and extents.
2507 * It is used to give priority to read and write RPCs being done
2508 * under this lock so that the client can drop the contended
2509 * lock more quickly and let other clients use it. This improves
2510 * overall performance in the case where the first client gets a
2511 * very large lock extent that prevents other clients from
2512 * submitting their writes.
2514 * \param[in] req ptlrpc_request being processed
2515 * \param[in] lock contended lock to match
2517 * \retval 1 if lock is matched
2518 * \retval 0 otherwise
2520 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
2521 struct ldlm_lock *lock)
2523 struct niobuf_remote *rnb;
2524 struct obd_ioobj *ioo;
2525 enum ldlm_mode mode;
2526 struct ldlm_extent ext;
2527 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2531 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2532 LASSERT(ioo != NULL);
2534 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2535 LASSERT(rnb != NULL);
2537 ext.start = rnb->rnb_offset;
2538 rnb += ioo->ioo_bufcnt - 1;
2539 ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
2541 LASSERT(lock->l_resource != NULL);
2542 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2545 /* a bulk write can only hold a reference on a PW extent lock
2548 mode = LCK_PW | LCK_GROUP;
2549 if (opc == OST_READ)
2550 /* whereas a bulk read can be protected by either a PR or PW
2554 if (!(lock->l_granted_mode & mode))
2557 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2561 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
2563 * Check for whether the given PTLRPC request (\a req) is blocking
2564 * an LDLM lock cancel. Also checks whether the request is covered by an LDLM
2567 * \param[in] req the incoming request
2569 * \retval 1 if \a req is blocking an LDLM lock cancel
2570 * \retval 0 if it is not
2571 * \retval -ESTALE if lock is not found
2573 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
2575 struct tgt_session_info *tsi;
2576 struct obd_ioobj *ioo;
2577 struct niobuf_remote *rnb;
2579 struct ldlm_prolong_args pa = { 0 };
2583 /* Don't use tgt_ses_info() to get session info, because lock_match()
2584 * can be called while request has no processing thread yet. */
2585 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2588 * Use LASSERT below because malformed RPCs should have
2589 * been filtered out in tgt_hpreq_handler().
2591 opc = lustre_msg_get_opc(req->rq_reqmsg);
2592 LASSERT(opc == OST_READ || opc == OST_WRITE);
2594 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2595 LASSERT(ioo != NULL);
2597 rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2598 LASSERT(rnb != NULL);
2599 LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
2601 pa.lpa_mode = LCK_PW | LCK_GROUP;
2602 if (opc == OST_READ)
2603 pa.lpa_mode |= LCK_PR;
2605 pa.lpa_extent.start = rnb->rnb_offset;
2606 rnb += ioo->ioo_bufcnt - 1;
2607 pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1;
2609 DEBUG_REQ(D_RPCTRACE, req,
2610 "%s %s: refresh rw locks for "DFID" (%llu->%llu)",
2611 tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid),
2612 pa.lpa_extent.start, pa.lpa_extent.end);
2614 ofd_prolong_extent_locks(tsi, &pa);
2616 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n",
2617 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2619 if (pa.lpa_blocks_cnt > 0)
2622 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2626 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
2628 * Called after the request has been handled. It refreshes lock timeout again
2629 * so that client has more time to send lock cancel RPC.
2631 * \param[in] req request which is being processed.
2633 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
2635 ofd_rw_hpreq_check(req);
2639 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
2641 * This function checks if the given lock is the same by its resname, mode
2642 * and extent as one taken from the request.
2643 * It is used to give priority to punch/truncate RPCs that might lead to
2644 * the fastest release of that lock when a lock is contended.
2646 * \param[in] req ptlrpc_request being processed
2647 * \param[in] lock contended lock to match
2649 * \retval 1 if lock is matched
2650 * \retval 0 otherwise
2652 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
2653 struct ldlm_lock *lock)
2655 struct tgt_session_info *tsi;
2657 struct ldlm_extent ext;
2661 /* Don't use tgt_ses_info() to get session info, because lock_match()
2662 * can be called while request has no processing thread yet. */
2663 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2666 * Use LASSERT below because malformed RPCs should have
2667 * been filtered out in tgt_hpreq_handler().
2669 LASSERT(tsi->tsi_ost_body != NULL);
2670 if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
2671 tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2674 oa = &tsi->tsi_ost_body->oa;
2675 ext.start = oa->o_size;
2676 ext.end = oa->o_blocks;
2678 LASSERT(lock->l_resource != NULL);
2679 if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name))
2682 if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
2685 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2689 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
2691 * High-priority queue request check for whether the given punch request
2692 * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is
2693 * covered by an LDLM lock.
2697 * \param[in] req the incoming request
2699 * \retval 1 if \a req is blocking an LDLM lock cancel
2700 * \retval 0 if it is not
2701 * \retval -ESTALE if lock is not found
2703 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
2705 struct tgt_session_info *tsi;
2707 struct ldlm_prolong_args pa = { 0 };
2711 /* Don't use tgt_ses_info() to get session info, because lock_match()
2712 * can be called while request has no processing thread yet. */
2713 tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
2714 LASSERT(tsi != NULL);
2715 oa = &tsi->tsi_ost_body->oa;
2717 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
2718 oa->o_flags & OBD_FL_SRVLOCK));
2720 pa.lpa_mode = LCK_PW | LCK_GROUP;
2721 pa.lpa_extent.start = oa->o_size;
2722 pa.lpa_extent.end = oa->o_blocks;
2725 "%s: refresh locks: %llu/%llu (%llu->%llu)\n",
2726 tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
2727 tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end);
2729 ofd_prolong_extent_locks(tsi, &pa);
2731 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2732 tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req);
2734 if (pa.lpa_blocks_cnt > 0)
2737 RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE);
2741 * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
2743 * Called after the request has been handled. It refreshes lock timeout again
2744 * so that client has more time to send lock cancel RPC.
2746 * \param[in] req request which is being processed.
2748 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
2750 ofd_punch_hpreq_check(req);
2753 static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
2754 .hpreq_lock_match = ofd_rw_hpreq_lock_match,
2755 .hpreq_check = ofd_rw_hpreq_check,
2756 .hpreq_fini = ofd_rw_hpreq_fini
2759 static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
2760 .hpreq_lock_match = ofd_punch_hpreq_lock_match,
2761 .hpreq_check = ofd_punch_hpreq_check,
2762 .hpreq_fini = ofd_punch_hpreq_fini
2766 * Assign high priority operations to an IO request.
2768 * Check if the incoming request is a candidate for
2769 * high-priority processing. If it is, assign it a high
2770 * priority operations table.
2772 * \param[in] tsi target session environment for this request
2774 static void ofd_hp_brw(struct tgt_session_info *tsi)
2776 struct niobuf_remote *rnb;
2777 struct obd_ioobj *ioo;
2781 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2782 LASSERT(ioo != NULL); /* must exist after request preprocessing */
2783 if (ioo->ioo_bufcnt > 0) {
2784 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2785 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2787 /* no high priority if server lock is needed */
2788 if (rnb->rnb_flags & OBD_BRW_SRVLOCK ||
2789 (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg)
2793 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2797 * Assign high priority operations to an punch request.
2799 * Check if the incoming request is a candidate for
2800 * high-priority processing. If it is, assign it a high
2801 * priority operations table.
2803 * \param[in] tsi target session environment for this request
2805 static void ofd_hp_punch(struct tgt_session_info *tsi)
2807 LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2808 /* no high-priority if server lock is needed */
2809 if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2810 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) ||
2811 tgt_conn_flags(tsi) & OBD_CONNECT_MDS ||
2812 lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY)
2814 tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2817 #define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
2818 #define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
2819 #define OST_BRW_READ OST_READ
2820 #define OST_BRW_WRITE OST_WRITE
2823 * Table of OFD-specific request handlers
2825 * This table contains all opcodes accepted by OFD and
2826 * specifies handlers for them. The tgt_request_handler()
2827 * uses such table from each target to process incoming
2830 static struct tgt_handler ofd_tgt_handlers[] = {
2831 TGT_RPC_HANDLER(OST_FIRST_OPC,
2832 0, OST_CONNECT, tgt_connect,
2833 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2834 TGT_RPC_HANDLER(OST_FIRST_OPC,
2835 0, OST_DISCONNECT, tgt_disconnect,
2836 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2837 TGT_RPC_HANDLER(OST_FIRST_OPC,
2838 0, OST_SET_INFO, ofd_set_info_hdl,
2839 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2840 TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl),
2841 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl),
2842 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2843 OST_SETATTR, ofd_setattr_hdl),
2844 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2845 OST_CREATE, ofd_create_hdl),
2846 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE,
2847 OST_DESTROY, ofd_destroy_hdl),
2848 TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl),
2849 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
2851 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2852 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write,
2854 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
2855 OST_PUNCH, ofd_punch_hdl,
2857 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl),
2858 TGT_OST_HDL(0, OST_QUOTACTL, ofd_quotactl),
2859 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl),
2860 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE, ofd_fallocate_hdl),
2861 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
2864 static struct tgt_opc_slice ofd_common_slice[] = {
2866 .tos_opc_start = OST_FIRST_OPC,
2867 .tos_opc_end = OST_LAST_OPC,
2868 .tos_hs = ofd_tgt_handlers
2871 .tos_opc_start = OBD_FIRST_OPC,
2872 .tos_opc_end = OBD_LAST_OPC,
2873 .tos_hs = tgt_obd_handlers
2876 .tos_opc_start = LDLM_FIRST_OPC,
2877 .tos_opc_end = LDLM_LAST_OPC,
2878 .tos_hs = tgt_dlm_handlers
2881 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
2882 .tos_opc_end = OUT_UPDATE_LAST_OPC,
2883 .tos_hs = tgt_out_handlers
2886 .tos_opc_start = SEQ_FIRST_OPC,
2887 .tos_opc_end = SEQ_LAST_OPC,
2888 .tos_hs = seq_handlers
2891 .tos_opc_start = LFSCK_FIRST_OPC,
2892 .tos_opc_end = LFSCK_LAST_OPC,
2893 .tos_hs = tgt_lfsck_handlers
2896 .tos_opc_start = SEC_FIRST_OPC,
2897 .tos_opc_end = SEC_LAST_OPC,
2898 .tos_hs = tgt_sec_ctx_handlers
2905 /* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
2906 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2909 * Implementation of lu_context_key::lct_key_exit.
2911 * Optional method called on lu_context_exit() for all allocated
2913 * It is used in OFD to sanitize context values which may be re-used
2914 * during another request processing by the same thread.
2916 * \param[in] ctx execution context
2917 * \param[in] key context key
2918 * \param[in] data ofd_thread_info
2920 static void ofd_key_exit(const struct lu_context *ctx,
2921 struct lu_context_key *key, void *data)
2923 struct ofd_thread_info *info = data;
2925 info->fti_env = NULL;
2926 info->fti_exp = NULL;
2929 info->fti_pre_version = 0;
2931 memset(&info->fti_attr, 0, sizeof info->fti_attr);
2934 struct lu_context_key ofd_thread_key = {
2935 .lct_tags = LCT_DT_THREAD,
2936 .lct_init = ofd_key_init,
2937 .lct_fini = ofd_key_fini,
2938 .lct_exit = ofd_key_exit
2942 * Initialize OFD device according to parameters in the config log \a cfg.
2944 * This is the main starting point of OFD initialization. It fills all OFD
2945 * parameters with their initial values and calls other initializing functions
2946 * to set up all OFD subsystems.
2948 * \param[in] env execution environment
2949 * \param[in] m OFD device
2950 * \param[in] ldt LU device type of OFD
2951 * \param[in] cfg configuration log
2953 * \retval 0 if successful
2954 * \retval negative value on error
2956 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2957 struct lu_device_type *ldt, struct lustre_cfg *cfg)
2959 const char *dev = lustre_cfg_string(cfg, 0);
2960 struct ofd_thread_info *info = NULL;
2961 struct obd_device *obd;
2962 struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
2963 DECLARE_BITMAP(lmd_flags, LMD_FLG_NUM_FLAGS);
2965 struct nm_config_file *nodemap_config;
2966 struct obd_device_target *obt;
2971 obd = class_name2obd(dev);
2973 CERROR("Cannot find obd with name %s\n", dev);
2977 rc = lu_env_refill((struct lu_env *)env);
2981 obt = obd_obt_init(obd);
2983 spin_lock_init(&m->ofd_flags_lock);
2984 m->ofd_raid_degraded = 0;
2985 m->ofd_sync_journal = 0;
2987 m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2989 m->ofd_seq_count = 0;
2990 INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2991 spin_lock_init(&m->ofd_inconsistency_lock);
2993 m->ofd_access_log_mask = -1; /* Log all accesses if enabled. */
2995 spin_lock_init(&m->ofd_batch_lock);
2996 init_rwsem(&m->ofd_lastid_rwsem);
2998 m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2999 m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
3000 /* set this lu_device to obd, because error handling need it */
3001 obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
3003 /* No connection accepted until configurations will finish */
3004 spin_lock(&obd->obd_dev_lock);
3005 obd->obd_no_conn = 1;
3006 spin_unlock(&obd->obd_dev_lock);
3007 obd->obd_replayable = 1;
3008 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
3009 char *str = lustre_cfg_string(cfg, 4);
3011 if (strchr(str, 'n')) {
3012 CWARN("%s: recovery disabled\n", obd->obd_name);
3013 obd->obd_replayable = 0;
3017 info = ofd_info_init(env, NULL);
3021 rc = ofd_stack_init(env, m, cfg, lmd_flags);
3023 CERROR("%s: can't init device stack, rc %d\n",
3028 snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
3029 "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
3030 m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
3031 LDLM_NAMESPACE_SERVER,
3032 LDLM_NAMESPACE_GREEDY,
3034 if (IS_ERR(m->ofd_namespace)) {
3035 rc = PTR_ERR(m->ofd_namespace);
3036 CERROR("%s: unable to create server namespace: rc = %d\n",
3038 m->ofd_namespace = NULL;
3039 GOTO(err_fini_stack, rc);
3041 /* set obd_namespace for compatibility with old code */
3042 obd->obd_namespace = m->ofd_namespace;
3043 ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
3044 m->ofd_namespace->ns_lvbo = &ofd_lvbo;
3045 m->ofd_namespace->ns_lvbp = m;
3047 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3048 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
3050 rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
3051 OBD_FAIL_OST_ALL_REQUEST_NET,
3052 OBD_FAIL_OST_ALL_REPLY_NET);
3054 GOTO(err_free_ns, rc);
3056 if (test_bit(LMD_FLG_SKIP_LFSCK, lmd_flags))
3057 m->ofd_skip_lfsck = 1;
3058 if (test_bit(LMD_FLG_LOCAL_RECOV, lmd_flags))
3059 m->ofd_lut.lut_local_recovery = 1;
3061 rc = ofd_tunables_init(m);
3063 GOTO(err_fini_lut, rc);
3065 tgd->tgd_reserved_pcnt = 0;
3067 m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
3068 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
3069 if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
3070 OFD_PRECREATE_SMALL_FS)
3071 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
3072 m->ofd_atime_diff = OFD_DEF_ATIME_DIFF;
3074 rc = ofd_fs_setup(env, m, obd);
3076 GOTO(err_fini_proc, rc);
3078 fid.f_seq = FID_SEQ_LOCAL_NAME;
3081 rc = local_oid_storage_init(env, m->ofd_osd, &fid,
3084 GOTO(err_fini_fs, rc);
3086 nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd,
3088 if (IS_ERR(nodemap_config)) {
3089 rc = PTR_ERR(nodemap_config);
3091 GOTO(err_fini_los, rc);
3093 obt->obt_nodemap_config_file = nodemap_config;
3096 rc = ofd_start_inconsistency_verification_thread(m);
3098 GOTO(err_fini_nm, rc);
3100 tgt_adapt_sptlrpc_conf(&m->ofd_lut);
3105 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
3106 obt->obt_nodemap_config_file = NULL;
3108 local_oid_storage_fini(env, m->ofd_los);
3111 ofd_fs_cleanup(env, m);
3115 tgt_fini(env, &m->ofd_lut);
3117 ldlm_namespace_free(m->ofd_namespace, NULL, obd->obd_force);
3118 obd->obd_namespace = m->ofd_namespace = NULL;
3120 ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
3125 * Stop the OFD device
3127 * This function stops the OFD device and all its subsystems.
3128 * This is the end of OFD lifecycle.
3130 * \param[in] env execution environment
3131 * \param[in] m OFD device
3133 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
3135 struct obd_device *obd = ofd_obd(m);
3136 struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
3137 struct lfsck_stop stop;
3139 stop.ls_status = LS_PAUSED;
3141 lfsck_stop(env, m->ofd_osd, &stop);
3142 ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3143 target_recovery_fini(obd);
3144 if (m->ofd_namespace != NULL)
3145 ldlm_namespace_free_prior(m->ofd_namespace, NULL,
3146 d->ld_obd->obd_force);
3148 obd_exports_barrier(obd);
3149 obd_zombie_barrier();
3152 tgt_fini(env, &m->ofd_lut);
3153 ofd_stop_inconsistency_verification_thread(m);
3154 lfsck_degister(env, m->ofd_osd);
3155 ofd_fs_cleanup(env, m);
3156 nm_config_file_deregister_tgt(env,
3157 obd2obt(obd)->obt_nodemap_config_file);
3158 obd2obt(obd)->obt_nodemap_config_file = NULL;
3160 if (m->ofd_namespace != NULL) {
3161 ldlm_namespace_free_post(m->ofd_namespace);
3162 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
3165 ofd_access_log_delete(m->ofd_access_log);
3166 m->ofd_access_log = NULL;
3168 ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
3170 LASSERT(atomic_read(&d->ld_ref) == 0);
3171 server_put_mount(obd->obd_name, true);
3176 * Implementation of lu_device_type_operations::ldto_device_fini.
3178 * Finalize device. Dual to ofd_device_init(). It is called from
3179 * obd_precleanup() and stops the current device.
3181 * \param[in] env execution environment
3182 * \param[in] d LU device of OFD
3186 static struct lu_device *ofd_device_fini(const struct lu_env *env,
3187 struct lu_device *d)
3190 ofd_fini(env, ofd_dev(d));
3195 * Implementation of lu_device_type_operations::ldto_device_free.
3197 * Free OFD device. Dual to ofd_device_alloc().
3199 * \param[in] env execution environment
3200 * \param[in] d LU device of OFD
3204 static struct lu_device *ofd_device_free(const struct lu_env *env,
3205 struct lu_device *d)
3207 struct ofd_device *m = ofd_dev(d);
3209 dt_device_fini(&m->ofd_dt_dev);
3215 * Implementation of lu_device_type_operations::ldto_device_alloc.
3217 * This function allocates the new OFD device. It is called from
3218 * obd_setup() if OBD device had lu_device_type defined.
3220 * \param[in] env execution environment
3221 * \param[in] t lu_device_type of OFD device
3222 * \param[in] cfg configuration log
3224 * \retval pointer to the lu_device of just allocated OFD
3225 * \retval ERR_PTR of return value on error
3227 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
3228 struct lu_device_type *t,
3229 struct lustre_cfg *cfg)
3231 struct ofd_device *m;
3232 struct lu_device *l;
3237 return ERR_PTR(-ENOMEM);
3239 l = &m->ofd_dt_dev.dd_lu_dev;
3240 dt_device_init(&m->ofd_dt_dev, t);
3241 rc = ofd_init0(env, m, t, cfg);
3243 ofd_device_free(env, l);
3250 /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
3251 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
3253 static const struct lu_device_type_operations ofd_device_type_ops = {
3254 .ldto_init = ofd_type_init,
3255 .ldto_fini = ofd_type_fini,
3257 .ldto_start = ofd_type_start,
3258 .ldto_stop = ofd_type_stop,
3260 .ldto_device_alloc = ofd_device_alloc,
3261 .ldto_device_free = ofd_device_free,
3262 .ldto_device_fini = ofd_device_fini
3265 static struct lu_device_type ofd_device_type = {
3266 .ldt_tags = LU_DEVICE_DT,
3267 .ldt_name = LUSTRE_OST_NAME,
3268 .ldt_ops = &ofd_device_type_ops,
3269 .ldt_ctx_tags = LCT_DT_THREAD
3273 * Initialize OFD module.
3275 * This function is called upon module loading. It registers OFD device type
3276 * and prepares all in-memory structures used by all OFD devices.
3278 * \retval 0 if successful
3279 * \retval negative value on error
3281 static int __init ofd_init(void)
3285 rc = libcfs_setup();
3289 rc = lu_kmem_init(ofd_caches);
3293 rc = ofd_access_log_module_init();
3297 rc = class_register_type(&ofd_obd_ops, NULL, true,
3298 LUSTRE_OST_NAME, &ofd_device_type);
3300 goto out_ofd_access_log;
3305 ofd_access_log_module_exit();
3307 lu_kmem_fini(ofd_caches);
3315 * This function is called upon OFD module unloading.
3316 * It frees all related structures and unregisters OFD device type.
3318 static void __exit ofd_exit(void)
3320 class_unregister_type(LUSTRE_OST_NAME);
3321 ofd_access_log_module_exit();
3322 lu_kmem_fini(ofd_caches);
3325 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3326 MODULE_DESCRIPTION("Lustre Object Filtering Device");
3327 MODULE_VERSION(LUSTRE_VERSION_STRING);
3328 MODULE_LICENSE("GPL");
3330 module_init(ofd_init);
3331 module_exit(ofd_exit);