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,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/lod/lod_dev.c
34 * Lustre Logical Object Device
36 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
37 * Author: Mikhail Pershin <mike.pershin@intel.com>
40 * The Logical Object Device (LOD) layer manages access to striped
41 * objects (both regular files and directories). It implements the DT
42 * device and object APIs and is responsible for creating, storing,
43 * and loading striping information as an extended attribute of the
44 * underlying OSD object. LOD is the server side analog of the LOV and
45 * LMV layers on the client side.
47 * Metadata LU object stack (layers of the same compound LU object,
48 * all have the same FID):
58 * During LOD object initialization the localness or remoteness of the
59 * object FID dictates the choice between OSD and OSP.
61 * An LOD object (file or directory) with N stripes (each has a
68 * S0 S1 S2 S3 S(N-1) OS[DP] objects, seen as DT objects by LOD
70 * When upper layers must access an object's stripes (which are
71 * themselves OST or MDT LU objects) LOD finds these objects by their
72 * FIDs and stores them as an array of DT object pointers on the
73 * object. Declarations and operations on LOD objects are received by
74 * LOD (as DT object operations) and performed on the underlying
75 * OS[DP] object and (as needed) on the stripes. From the perspective
76 * of LOD, a stripe-less file (created by mknod() or open with
77 * O_LOV_DELAY_CREATE) is an object which does not yet have stripes,
78 * while a non-striped directory (created by mkdir()) is an object
79 * which will never have stripes.
81 * The LOD layer also implements a small subset of the OBD device API
82 * to support MDT stack initialization and finalization (an MDD device
83 * connects and disconnects itself to and from the underlying LOD
84 * device), and pool management. In turn LOD uses the OBD device API
85 * to connect it self to the underlying OSD, and to connect itself to
86 * OSP devices representing the MDTs and OSTs that bear the stripes of
90 #define DEBUG_SUBSYSTEM S_MDS
92 #include <linux/kthread.h>
93 #include <obd_class.h>
94 #include <md_object.h>
95 #include <lustre_fid.h>
96 #include <lustre_param.h>
97 #include <lustre_update.h>
98 #include <lustre_log.h>
100 #include "lod_internal.h"
102 static const char lod_update_log_name[] = "update_log";
103 static const char lod_update_log_dir_name[] = "update_log_dir";
106 * Lookup target by FID.
108 * Lookup MDT/OST target index by FID. Type of the target can be
111 * \param[in] env LU environment provided by the caller
112 * \param[in] lod lod device
114 * \param[out] tgt result target index
115 * \param[in] type expected type of the target:
116 * LU_SEQ_RANGE_{MDT,OST,ANY}
118 * \retval 0 on success
119 * \retval negative negated errno on error
121 int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod,
122 const struct lu_fid *fid, __u32 *tgt, int *type)
124 struct lu_seq_range range = { 0 };
125 struct lu_server_fld *server_fld;
129 if (!fid_is_sane(fid)) {
130 CERROR("%s: invalid FID "DFID"\n", lod2obd(lod)->obd_name,
135 if (fid_is_idif(fid)) {
136 *tgt = fid_idif_ost_idx(fid);
137 *type = LU_SEQ_RANGE_OST;
141 if (fid_is_update_log(fid) || fid_is_update_log_dir(fid)) {
143 *type = LU_SEQ_RANGE_MDT;
147 if (!lod->lod_initialized || (!fid_seq_in_fldb(fid_seq(fid)))) {
148 LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);
150 *tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
151 *type = LU_SEQ_RANGE_MDT;
155 server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
156 if (server_fld == NULL)
159 fld_range_set_type(&range, *type);
160 rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
164 *tgt = range.lsr_index;
165 *type = range.lsr_flags;
167 CDEBUG(D_INFO, "%s: got tgt %x for sequence: "LPX64"\n",
168 lod2obd(lod)->obd_name, *tgt, fid_seq(fid));
173 /* Slab for OSD object allocation */
174 struct kmem_cache *lod_object_kmem;
176 /* Slab for dt_txn_callback */
177 struct kmem_cache *lod_txn_callback_kmem;
178 static struct lu_kmem_descr lod_caches[] = {
180 .ckd_cache = &lod_object_kmem,
181 .ckd_name = "lod_obj",
182 .ckd_size = sizeof(struct lod_object)
185 .ckd_cache = &lod_txn_callback_kmem,
186 .ckd_name = "lod_txn_callback",
187 .ckd_size = sizeof(struct dt_txn_callback)
194 static struct lu_device *lod_device_fini(const struct lu_env *env,
195 struct lu_device *d);
198 * Implementation of lu_device_operations::ldo_object_alloc() for LOD
200 * Allocates and initializes LOD's slice in the given object.
202 * see include/lu_object.h for the details.
204 static struct lu_object *lod_object_alloc(const struct lu_env *env,
205 const struct lu_object_header *hdr,
206 struct lu_device *dev)
208 struct lod_object *lod_obj;
209 struct lu_object *lu_obj;
212 OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, GFP_NOFS);
214 RETURN(ERR_PTR(-ENOMEM));
216 lu_obj = lod2lu_obj(lod_obj);
217 dt_object_init(&lod_obj->ldo_obj, NULL, dev);
218 lod_obj->ldo_obj.do_ops = &lod_obj_ops;
219 lu_obj->lo_ops = &lod_lu_obj_ops;
225 * Process the config log for all sub device.
227 * The function goes through all the targets in the given table
228 * and apply given configuration command on to the targets.
229 * Used to cleanup the targets at unmount.
231 * \param[in] env LU environment provided by the caller
232 * \param[in] lod lod device
233 * \param[in] ltd target's table to go through
234 * \param[in] lcfg configuration command to apply
236 * \retval 0 on success
237 * \retval negative negated errno on error
239 static int lod_sub_process_config(const struct lu_env *env,
240 struct lod_device *lod,
241 struct lod_tgt_descs *ltd,
242 struct lustre_cfg *lcfg)
244 struct lu_device *next;
249 if (ltd->ltd_tgts_size <= 0) {
250 lod_putref(lod, ltd);
253 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
254 struct lod_tgt_desc *tgt;
257 tgt = LTD_TGT(ltd, i);
258 LASSERT(tgt && tgt->ltd_tgt);
259 next = &tgt->ltd_tgt->dd_lu_dev;
260 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
262 CERROR("%s: error cleaning up LOD index %u: cmd %#x"
263 ": rc = %d\n", lod2obd(lod)->obd_name, i,
264 lcfg->lcfg_command, rc1);
268 lod_putref(lod, ltd);
272 struct lod_recovery_data {
273 struct lod_device *lrd_lod;
274 struct lod_tgt_desc *lrd_ltd;
275 struct ptlrpc_thread *lrd_thread;
281 * process update recovery record
283 * Add the update recovery recode to the update recovery list in
284 * lod_recovery_data. Then the recovery thread (target_recovery_thread)
285 * will redo these updates.
287 * \param[in]env execution environment
288 * \param[in]llh log handle of update record
289 * \param[in]rec update record to be replayed
290 * \param[in]data update recovery data which holds the necessary
291 * arguments for recovery (see struct lod_recovery_data)
293 * \retval 0 if the record is processed successfully.
294 * \retval negative errno if the record processing fails.
296 static int lod_process_recovery_updates(const struct lu_env *env,
297 struct llog_handle *llh,
298 struct llog_rec_hdr *rec,
301 struct lod_recovery_data *lrd = data;
302 struct llog_cookie *cookie = &lod_env_info(env)->lti_cookie;
303 struct lu_target *lut;
307 if (lrd->lrd_ltd == NULL) {
310 rc = lodname2mdt_index(lod2obd(lrd->lrd_lod)->obd_name, &index);
314 index = lrd->lrd_ltd->ltd_index;
318 llog_update_record_size((struct llog_update_record *)rec)) {
319 CERROR("%s broken update record! index %u "DOSTID":%u :"
320 " rc = %d\n", lod2obd(lrd->lrd_lod)->obd_name, index,
321 POSTID(&llh->lgh_id.lgl_oi), rec->lrh_index, -EIO);
325 cookie->lgc_lgl = llh->lgh_id;
326 cookie->lgc_index = rec->lrh_index;
327 cookie->lgc_subsys = LLOG_UPDATELOG_ORIG_CTXT;
329 CDEBUG(D_HA, "%s: process recovery updates "DOSTID":%u\n",
330 lod2obd(lrd->lrd_lod)->obd_name,
331 POSTID(&llh->lgh_id.lgl_oi), rec->lrh_index);
332 lut = lod2lu_dev(lrd->lrd_lod)->ld_site->ls_tgt;
334 if (lut->lut_obd->obd_stopping ||
335 lut->lut_obd->obd_abort_recovery)
338 return insert_update_records_to_replay_list(lut->lut_tdtd,
339 (struct llog_update_record *)rec,
344 * recovery thread for update log
346 * Start recovery thread and prepare the sub llog, then it will retrieve
347 * the update records from the correpondent MDT and do recovery.
349 * \param[in] arg pointer to the recovery data
351 * \retval 0 if recovery succeeds
352 * \retval negative errno if recovery failed.
354 static int lod_sub_recovery_thread(void *arg)
356 struct lod_recovery_data *lrd = arg;
357 struct lod_device *lod = lrd->lrd_lod;
358 struct dt_device *dt;
359 struct ptlrpc_thread *thread = lrd->lrd_thread;
360 struct llog_ctxt *ctxt = NULL;
362 struct lu_target *lut;
368 thread->t_flags = SVC_RUNNING;
369 wake_up(&thread->t_ctl_waitq);
371 rc = lu_env_init(&env, LCT_LOCAL | LCT_MD_THREAD);
374 CERROR("%s: can't initialize env: rc = %d\n",
375 lod2obd(lod)->obd_name, rc);
379 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
380 atomic_inc(&lut->lut_tdtd->tdtd_recovery_threads_count);
381 if (lrd->lrd_ltd == NULL)
384 dt = lrd->lrd_ltd->ltd_tgt;
387 rc = lod_sub_prep_llog(&env, lod, dt, lrd->lrd_idx);
389 /* Process the recovery record */
390 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
391 LLOG_UPDATELOG_ORIG_CTXT);
392 LASSERT(ctxt != NULL);
393 LASSERT(ctxt->loc_handle != NULL);
395 rc = llog_cat_process(&env, ctxt->loc_handle,
396 lod_process_recovery_updates, lrd, 0, 0);
400 struct lu_device *top_device;
402 top_device = lod->lod_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
403 /* Because the remote target might failover at the same time,
404 * let's retry here */
405 if ((rc == -ETIMEDOUT || rc == -EAGAIN || rc == -EIO) &&
406 dt != lod->lod_child &&
407 !top_device->ld_obd->obd_abort_recovery &&
408 !top_device->ld_obd->obd_stopping) {
410 if (ctxt->loc_handle != NULL)
418 CERROR("%s getting update log failed: rc = %d\n",
419 dt->dd_lu_dev.ld_obd->obd_name, rc);
422 spin_lock(&top_device->ld_obd->obd_dev_lock);
423 if (!top_device->ld_obd->obd_abort_recovery &&
424 !top_device->ld_obd->obd_stopping)
425 top_device->ld_obd->obd_abort_recovery = 1;
426 spin_unlock(&top_device->ld_obd->obd_dev_lock);
432 CDEBUG(D_HA, "%s retrieve update log: rc = %d\n",
433 dt->dd_lu_dev.ld_obd->obd_name, rc);
435 if (lrd->lrd_ltd == NULL)
436 lod->lod_child_got_update_log = 1;
438 lrd->lrd_ltd->ltd_got_update_log = 1;
440 if (lod->lod_child_got_update_log) {
441 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
442 struct lod_tgt_desc *tgt = NULL;
443 bool all_got_log = true;
446 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
447 tgt = LTD_TGT(ltd, i);
448 if (!tgt->ltd_got_update_log) {
455 CDEBUG(D_HA, "%s got update logs from all MDTs.\n",
456 lut->lut_obd->obd_name);
457 lut->lut_tdtd->tdtd_replay_ready = 1;
458 wake_up(&lut->lut_obd->obd_next_transno_waitq);
464 thread->t_flags = SVC_STOPPED;
465 atomic_dec(&lut->lut_tdtd->tdtd_recovery_threads_count);
466 wake_up(&lut->lut_tdtd->tdtd_recovery_threads_waitq);
467 wake_up(&thread->t_ctl_waitq);
473 * finish sub llog context
475 * Stop update recovery thread for the sub device, then cleanup the
476 * correspondent llog ctxt.
478 * \param[in] env execution environment
479 * \param[in] lod lod device to do update recovery
480 * \param[in] thread recovery thread on this sub device
482 void lod_sub_fini_llog(const struct lu_env *env,
483 struct dt_device *dt, struct ptlrpc_thread *thread)
485 struct obd_device *obd;
486 struct llog_ctxt *ctxt;
489 obd = dt->dd_lu_dev.ld_obd;
490 CDEBUG(D_INFO, "%s: finish sub llog\n", obd->obd_name);
491 /* Stop recovery thread first */
492 if (thread != NULL && thread->t_flags & SVC_RUNNING) {
493 thread->t_flags = SVC_STOPPING;
494 wake_up(&thread->t_ctl_waitq);
495 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
498 ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
502 if (ctxt->loc_handle != NULL)
503 llog_cat_close(env, ctxt->loc_handle);
505 llog_cleanup(env, ctxt);
511 * Extract MDT target index from a device name.
513 * a helper function to extract index from the given device name
514 * like "fsname-MDTxxxx-mdtlov"
516 * \param[in] lodname device name
517 * \param[out] mdt_index extracted index
519 * \retval 0 on success
520 * \retval -EINVAL if the name is invalid
522 int lodname2mdt_index(char *lodname, __u32 *mdt_index)
527 /* 1.8 configs don't have "-MDT0000" at the end */
528 ptr = strstr(lodname, "-MDT");
534 ptr = strrchr(lodname, '-');
536 CERROR("invalid MDT index in '%s'\n", lodname);
540 if (strncmp(ptr, "-mdtlov", 7) != 0) {
541 CERROR("invalid MDT index in '%s'\n", lodname);
545 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
546 CERROR("invalid MDT index in '%s'\n", lodname);
550 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
551 CERROR("invalid MDT index in '%s'\n", lodname);
555 index = simple_strtol(ptr - 4, &tmp, 16);
556 if (*tmp != '-' || index > INT_MAX) {
557 CERROR("invalid MDT index in '%s'\n", lodname);
565 * Init sub llog context
567 * Setup update llog ctxt for update recovery threads, then start the
568 * recovery thread (lod_sub_recovery_thread) to read update llog from
569 * the correspondent MDT to do update recovery.
571 * \param[in] env execution environment
572 * \param[in] lod lod device to do update recovery
573 * \param[in] dt sub dt device for which the recovery thread is
575 * \retval 0 if initialization succeeds.
576 * \retval negative errno if initialization fails.
578 int lod_sub_init_llog(const struct lu_env *env, struct lod_device *lod,
579 struct dt_device *dt)
581 struct obd_device *obd;
582 struct lod_recovery_data *lrd = NULL;
583 struct ptlrpc_thread *thread;
584 struct task_struct *task;
585 struct l_wait_info lwi = { 0 };
586 struct lod_tgt_desc *sub_ltd = NULL;
592 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &master_index);
600 if (lod->lod_child == dt) {
601 thread = &lod->lod_child_recovery_thread;
602 index = master_index;
604 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
605 struct lod_tgt_desc *tgt = NULL;
608 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
609 tgt = LTD_TGT(ltd, i);
610 if (tgt->ltd_tgt == dt) {
611 index = tgt->ltd_index;
616 LASSERT(sub_ltd != NULL);
617 OBD_ALLOC_PTR(sub_ltd->ltd_recovery_thread);
618 if (sub_ltd->ltd_recovery_thread == NULL)
619 GOTO(free_lrd, rc = -ENOMEM);
621 thread = sub_ltd->ltd_recovery_thread;
624 CDEBUG(D_INFO, "%s init sub log %s\n", lod2obd(lod)->obd_name,
625 dt->dd_lu_dev.ld_obd->obd_name);
627 lrd->lrd_ltd = sub_ltd;
628 lrd->lrd_thread = thread;
629 lrd->lrd_idx = index;
630 init_waitqueue_head(&thread->t_ctl_waitq);
632 obd = dt->dd_lu_dev.ld_obd;
633 obd->obd_lvfs_ctxt.dt = dt;
634 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_UPDATELOG_ORIG_CTXT,
635 NULL, &llog_common_cat_ops);
637 CERROR("%s: cannot setup updatelog llog: rc = %d\n",
639 GOTO(free_thread, rc);
642 /* Start the recovery thread */
643 task = kthread_run(lod_sub_recovery_thread, lrd, "lod%04x_rec%04x",
644 master_index, index);
647 CERROR("%s: cannot start recovery thread: rc = %d\n",
652 l_wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_RUNNING ||
653 thread->t_flags & SVC_STOPPED, &lwi);
657 lod_sub_fini_llog(env, dt, thread);
659 if (lod->lod_child != dt) {
660 OBD_FREE_PTR(sub_ltd->ltd_recovery_thread);
661 sub_ltd->ltd_recovery_thread = NULL;
669 * Stop sub recovery thread
671 * Stop sub recovery thread on all subs.
673 * \param[in] env execution environment
674 * \param[in] lod lod device to do update recovery
676 static void lod_sub_stop_recovery_threads(const struct lu_env *env,
677 struct lod_device *lod)
679 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
680 struct ptlrpc_thread *thread;
683 /* Stop the update log commit cancel threads and finish master
685 thread = &lod->lod_child_recovery_thread;
686 /* Stop recovery thread first */
687 if (thread != NULL && thread->t_flags & SVC_RUNNING) {
688 thread->t_flags = SVC_STOPPING;
689 wake_up(&thread->t_ctl_waitq);
690 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
694 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
695 struct lod_tgt_desc *tgt;
697 tgt = LTD_TGT(ltd, i);
698 thread = tgt->ltd_recovery_thread;
699 if (thread != NULL && thread->t_flags & SVC_RUNNING) {
700 thread->t_flags = SVC_STOPPING;
701 wake_up(&thread->t_ctl_waitq);
702 wait_event(thread->t_ctl_waitq,
703 thread->t_flags & SVC_STOPPED);
704 OBD_FREE_PTR(tgt->ltd_recovery_thread);
705 tgt->ltd_recovery_thread = NULL;
709 lod_putref(lod, ltd);
713 * finish all sub llog
715 * cleanup all of sub llog ctxt on the LOD.
717 * \param[in] env execution environment
718 * \param[in] lod lod device to do update recovery
720 static void lod_sub_fini_all_llogs(const struct lu_env *env,
721 struct lod_device *lod)
723 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
726 /* Stop the update log commit cancel threads and finish master
728 lod_sub_fini_llog(env, lod->lod_child,
729 &lod->lod_child_recovery_thread);
731 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
732 struct lod_tgt_desc *tgt;
734 tgt = LTD_TGT(ltd, i);
735 lod_sub_fini_llog(env, tgt->ltd_tgt,
736 tgt->ltd_recovery_thread);
739 lod_putref(lod, ltd);
742 static char *lod_show_update_logs_retrievers(void *data, int *size, int *count)
744 struct lod_device *lod = (struct lod_device *)data;
745 struct lu_target *lut = lod2lu_dev(lod)->ld_site->ls_tgt;
746 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
747 struct lod_tgt_desc *tgt = NULL;
753 *count = atomic_read(&lut->lut_tdtd->tdtd_recovery_threads_count);
759 *size = 5 * *count + 1;
760 OBD_ALLOC(buf, *size);
765 memset(buf, 0, *size);
767 if (!lod->lod_child_got_update_log) {
768 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &i);
769 LASSERTF(rc == 0, "Fail to parse target index: rc = %d\n", rc);
771 rc = snprintf(buf + len, *size - len, " %04x", i);
778 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
779 tgt = LTD_TGT(ltd, i);
780 if (!tgt->ltd_got_update_log) {
781 rc = snprintf(buf + len, *size - len, " %04x", i);
782 if (unlikely(rc <= 0))
794 * Prepare distribute txn
796 * Prepare distribute txn structure for LOD
798 * \param[in] env execution environment
799 * \param[in] lod_device LOD device
801 * \retval 0 if preparation succeeds.
802 * \retval negative errno if preparation fails.
804 static int lod_prepare_distribute_txn(const struct lu_env *env,
805 struct lod_device *lod)
807 struct target_distribute_txn_data *tdtd;
808 struct lu_target *lut;
812 /* Init update recovery data */
817 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
818 tdtd->tdtd_dt = &lod->lod_dt_dev;
819 rc = distribute_txn_init(env, lut, tdtd,
820 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id);
823 CERROR("%s: cannot init distribute txn: rc = %d\n",
824 lod2obd(lod)->obd_name, rc);
829 tdtd->tdtd_show_update_logs_retrievers =
830 lod_show_update_logs_retrievers;
831 tdtd->tdtd_show_retrievers_cbdata = lod;
833 lut->lut_tdtd = tdtd;
839 * Finish distribute txn
841 * Release the resource holding by distribute txn, i.e. stop distribute
844 * \param[in] env execution environment
845 * \param[in] lod lod device
847 static void lod_fini_distribute_txn(const struct lu_env *env,
848 struct lod_device *lod)
850 struct lu_target *lut;
852 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
853 if (lut->lut_tdtd == NULL)
856 distribute_txn_fini(env, lut->lut_tdtd);
858 OBD_FREE_PTR(lut->lut_tdtd);
859 lut->lut_tdtd = NULL;
863 * Implementation of lu_device_operations::ldo_process_config() for LOD
865 * The method is called by the configuration subsystem during setup,
866 * cleanup and when the configuration changes. The method processes
867 * few specific commands like adding/removing the targets, changing
868 * the runtime parameters.
870 * \param[in] env LU environment provided by the caller
871 * \param[in] dev lod device
872 * \param[in] lcfg configuration command to apply
874 * \retval 0 on success
875 * \retval negative negated errno on error
877 * The examples are below.
879 * Add osc config log:
880 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
881 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
882 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
883 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
884 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
885 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
887 * Add mdc config log:
888 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
889 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
890 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
891 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
892 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
893 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
895 static int lod_process_config(const struct lu_env *env,
896 struct lu_device *dev,
897 struct lustre_cfg *lcfg)
899 struct lod_device *lod = lu2lod_dev(dev);
900 struct lu_device *next = &lod->lod_child->dd_lu_dev;
905 switch(lcfg->lcfg_command) {
906 case LCFG_LOV_DEL_OBD:
907 case LCFG_LOV_ADD_INA:
908 case LCFG_LOV_ADD_OBD:
913 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
914 * modify_mdc_tgts add 0:lustre-MDT0001
915 * 1:lustre-MDT0001-mdc0002
917 arg1 = lustre_cfg_string(lcfg, 1);
919 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
920 GOTO(out, rc = -EINVAL);
921 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
922 GOTO(out, rc = -EINVAL);
924 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
927 rc = lodname2mdt_index(lustre_cfg_string(lcfg, 0),
932 rc = lod_add_device(env, lod, arg1, index, gen,
933 mdt_index, LUSTRE_OSC_NAME, 1);
934 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
936 rc = lod_add_device(env, lod, arg1, index, gen,
937 mdt_index, LUSTRE_MDC_NAME, 1);
938 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
939 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
941 rc = lod_add_device(env, lod, arg1, index, gen,
942 mdt_index, LUSTRE_OSC_NAME, 0);
944 rc = lod_del_device(env, lod,
946 arg1, index, gen, true);
953 struct obd_device *obd;
956 /* Check if it is activate/deactivate mdc
957 * lustre-MDTXXXX-osp-MDTXXXX.active=1 */
958 param = lustre_cfg_buf(lcfg, 1);
959 if (strstr(param, "osp") != NULL &&
960 strstr(param, ".active=") != NULL) {
961 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
962 struct lod_tgt_desc *sub_tgt = NULL;
967 ptr = strstr(param, ".");
969 obd = class_name2obd(param);
971 CERROR("%s: can not find %s: rc = %d\n",
972 lod2obd(lod)->obd_name, param, -EINVAL);
977 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
978 struct lod_tgt_desc *tgt;
980 tgt = LTD_TGT(ltd, i);
981 if (tgt->ltd_tgt->dd_lu_dev.ld_obd == obd) {
987 if (sub_tgt == NULL) {
988 CERROR("%s: can not find %s: rc = %d\n",
989 lod2obd(lod)->obd_name, param, -EINVAL);
995 tmp = strstr(param, "=");
998 struct llog_ctxt *ctxt;
1000 obd = sub_tgt->ltd_tgt->dd_lu_dev.ld_obd;
1001 ctxt = llog_get_context(obd,
1002 LLOG_UPDATELOG_ORIG_CTXT);
1004 rc = llog_setup(env, obd, &obd->obd_olg,
1005 LLOG_UPDATELOG_ORIG_CTXT,
1006 NULL, &llog_common_cat_ops);
1010 llog_ctxt_put(ctxt);
1012 rc = lod_sub_prep_llog(env, lod,
1014 sub_tgt->ltd_index);
1016 sub_tgt->ltd_active = 1;
1018 lod_sub_fini_llog(env, sub_tgt->ltd_tgt,
1020 sub_tgt->ltd_active = 0;
1026 rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
1032 case LCFG_PRE_CLEANUP: {
1033 if (lod->lod_md_root != NULL) {
1034 lu_object_put(env, &lod->lod_md_root->ldo_obj.do_lu);
1035 lod->lod_md_root = NULL;
1038 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1039 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1040 next = &lod->lod_child->dd_lu_dev;
1041 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1043 CDEBUG(D_HA, "%s: can't process %u: %d\n",
1044 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1046 lod_sub_stop_recovery_threads(env, lod);
1047 lod_fini_distribute_txn(env, lod);
1048 lod_sub_fini_all_llogs(env, lod);
1051 case LCFG_CLEANUP: {
1053 * do cleanup on underlying storage only when
1054 * all OSPs are cleaned up, as they use that OSD as well
1056 lu_dev_del_linkage(dev->ld_site, dev);
1057 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1058 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1059 next = &lod->lod_child->dd_lu_dev;
1060 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1062 CERROR("%s: can't process %u: %d\n",
1063 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1065 rc = obd_disconnect(lod->lod_child_exp);
1067 CERROR("error in disconnect from storage: %d\n", rc);
1071 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
1072 lcfg->lcfg_command);
1082 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
1084 * The method is called once the recovery is complete. This implementation
1085 * distributes the notification to all the known targets.
1087 * see include/lu_object.h for the details
1089 static int lod_recovery_complete(const struct lu_env *env,
1090 struct lu_device *dev)
1092 struct lod_device *lod = lu2lod_dev(dev);
1093 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1098 LASSERT(lod->lod_recovery_completed == 0);
1099 lod->lod_recovery_completed = 1;
1101 rc = next->ld_ops->ldo_recovery_complete(env, next);
1103 lod_getref(&lod->lod_ost_descs);
1104 if (lod->lod_osts_size > 0) {
1105 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
1106 struct lod_tgt_desc *tgt;
1107 tgt = OST_TGT(lod, i);
1108 LASSERT(tgt && tgt->ltd_tgt);
1109 next = &tgt->ltd_ost->dd_lu_dev;
1110 rc = next->ld_ops->ldo_recovery_complete(env, next);
1112 CERROR("%s: can't complete recovery on #%d:"
1113 "%d\n", lod2obd(lod)->obd_name, i, rc);
1116 lod_putref(lod, &lod->lod_ost_descs);
1121 * Init update logs on all sub device
1123 * LOD initialize update logs on all of sub devices. Because the initialization
1124 * process might need FLD lookup, see llog_osd_open()->dt_locate()->...->
1125 * lod_object_init(), this API has to be called after LOD is initialized.
1126 * \param[in] env execution environment
1127 * \param[in] lod lod device
1129 * \retval 0 if update log is initialized successfully.
1130 * \retval negative errno if initialization fails.
1132 static int lod_sub_init_llogs(const struct lu_env *env, struct lod_device *lod)
1134 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1139 /* llog must be setup after LOD is initialized, because llog
1140 * initialization include FLD lookup */
1141 LASSERT(lod->lod_initialized);
1143 /* Init the llog in its own stack */
1144 rc = lod_sub_init_llog(env, lod, lod->lod_child);
1148 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
1149 struct lod_tgt_desc *tgt;
1151 tgt = LTD_TGT(ltd, i);
1152 rc = lod_sub_init_llog(env, lod, tgt->ltd_tgt);
1161 * Implementation of lu_device_operations::ldo_prepare() for LOD
1163 * see include/lu_object.h for the details.
1165 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
1166 struct lu_device *cdev)
1168 struct lod_device *lod = lu2lod_dev(cdev);
1169 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1170 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1172 struct dt_object *root;
1173 struct dt_object *dto;
1177 rc = next->ld_ops->ldo_prepare(env, pdev, next);
1179 CERROR("%s: prepare bottom error: rc = %d\n",
1180 lod2obd(lod)->obd_name, rc);
1184 lod->lod_initialized = 1;
1186 rc = dt_root_get(env, lod->lod_child, fid);
1190 root = dt_locate(env, lod->lod_child, fid);
1192 RETURN(PTR_ERR(root));
1194 /* Create update log object */
1195 index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1196 lu_update_log_fid(fid, index);
1198 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1200 lod_update_log_name,
1201 S_IFREG | S_IRUGO | S_IWUSR);
1203 GOTO(out_put, rc = PTR_ERR(dto));
1205 lu_object_put(env, &dto->do_lu);
1207 /* Create update log dir */
1208 lu_update_log_dir_fid(fid, index);
1209 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1211 lod_update_log_dir_name,
1212 S_IFDIR | S_IRUGO | S_IWUSR);
1214 GOTO(out_put, rc = PTR_ERR(dto));
1216 lu_object_put(env, &dto->do_lu);
1218 rc = lod_prepare_distribute_txn(env, lod);
1222 rc = lod_sub_init_llogs(env, lod);
1227 lu_object_put(env, &root->do_lu);
1232 const struct lu_device_operations lod_lu_ops = {
1233 .ldo_object_alloc = lod_object_alloc,
1234 .ldo_process_config = lod_process_config,
1235 .ldo_recovery_complete = lod_recovery_complete,
1236 .ldo_prepare = lod_prepare,
1240 * Implementation of dt_device_operations::dt_root_get() for LOD
1242 * see include/dt_object.h for the details.
1244 static int lod_root_get(const struct lu_env *env,
1245 struct dt_device *dev, struct lu_fid *f)
1247 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
1251 * Implementation of dt_device_operations::dt_statfs() for LOD
1253 * see include/dt_object.h for the details.
1255 static int lod_statfs(const struct lu_env *env,
1256 struct dt_device *dev, struct obd_statfs *sfs)
1258 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
1262 * Implementation of dt_device_operations::dt_trans_create() for LOD
1264 * Creates a transaction using local (to this node) OSD.
1266 * see include/dt_object.h for the details.
1268 static struct thandle *lod_trans_create(const struct lu_env *env,
1269 struct dt_device *dt)
1273 th = top_trans_create(env, dt2lod_dev(dt)->lod_child);
1283 * Implementation of dt_device_operations::dt_trans_start() for LOD
1285 * Starts the set of local transactions using the targets involved
1286 * in declare phase. Initial support for the distributed transactions.
1288 * see include/dt_object.h for the details.
1290 static int lod_trans_start(const struct lu_env *env, struct dt_device *dt,
1293 return top_trans_start(env, dt2lod_dev(dt)->lod_child, th);
1296 static int lod_trans_cb_add(struct thandle *th,
1297 struct dt_txn_commit_cb *dcb)
1299 struct top_thandle *top_th = container_of(th, struct top_thandle,
1301 return dt_trans_cb_add(top_th->tt_master_sub_thandle, dcb);
1305 * add noop update to the update records
1307 * Add noop updates to the update records, which is only used in
1310 * \param[in] env execution environment
1311 * \param[in] dt dt device of lod
1312 * \param[in] th thandle
1313 * \param[in] count the count of update records to be added.
1315 * \retval 0 if adding succeeds.
1316 * \retval negative errno if adding fails.
1318 static int lod_add_noop_records(const struct lu_env *env,
1319 struct dt_device *dt, struct thandle *th,
1322 struct top_thandle *top_th;
1323 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1327 top_th = container_of(th, struct top_thandle, tt_super);
1328 if (top_th->tt_multiple_thandle == NULL)
1332 for (i = 0; i < count; i++) {
1333 rc = update_record_pack(noop, th, fid);
1341 * Implementation of dt_device_operations::dt_trans_stop() for LOD
1343 * Stops the set of local transactions using the targets involved
1344 * in declare phase. Initial support for the distributed transactions.
1346 * see include/dt_object.h for the details.
1348 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
1351 if (OBD_FAIL_CHECK(OBD_FAIL_SPLIT_UPDATE_REC)) {
1354 rc = lod_add_noop_records(env, dt, th, 5000);
1358 return top_trans_stop(env, dt2lod_dev(dt)->lod_child, th);
1362 * Implementation of dt_device_operations::dt_conf_get() for LOD
1364 * Currently returns the configuration provided by the local OSD.
1366 * see include/dt_object.h for the details.
1368 static void lod_conf_get(const struct lu_env *env,
1369 const struct dt_device *dev,
1370 struct dt_device_param *param)
1372 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
1376 * Implementation of dt_device_operations::dt_sync() for LOD
1378 * Syncs all known OST targets. Very very expensive and used
1379 * rarely by LFSCK now. Should not be used in general.
1381 * see include/dt_object.h for the details.
1383 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
1385 struct lod_device *lod = dt2lod_dev(dev);
1386 struct lod_ost_desc *ost;
1391 lod_getref(&lod->lod_ost_descs);
1392 lod_foreach_ost(lod, i) {
1393 ost = OST_TGT(lod, i);
1394 LASSERT(ost && ost->ltd_ost);
1395 rc = dt_sync(env, ost->ltd_ost);
1397 CERROR("%s: can't sync %u: %d\n",
1398 lod2obd(lod)->obd_name, i, rc);
1402 lod_putref(lod, &lod->lod_ost_descs);
1404 rc = dt_sync(env, lod->lod_child);
1410 * Implementation of dt_device_operations::dt_ro() for LOD
1412 * Turns local OSD read-only, used for the testing only.
1414 * see include/dt_object.h for the details.
1416 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
1418 return dt_ro(env, dt2lod_dev(dev)->lod_child);
1422 * Implementation of dt_device_operations::dt_commit_async() for LOD
1424 * Asks local OSD to commit sooner.
1426 * see include/dt_object.h for the details.
1428 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
1430 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
1433 static const struct dt_device_operations lod_dt_ops = {
1434 .dt_root_get = lod_root_get,
1435 .dt_statfs = lod_statfs,
1436 .dt_trans_create = lod_trans_create,
1437 .dt_trans_start = lod_trans_start,
1438 .dt_trans_stop = lod_trans_stop,
1439 .dt_conf_get = lod_conf_get,
1440 .dt_sync = lod_sync,
1442 .dt_commit_async = lod_commit_async,
1443 .dt_trans_cb_add = lod_trans_cb_add,
1447 * Connect to a local OSD.
1449 * Used to connect to the local OSD at mount. OSD name is taken from the
1450 * configuration command passed. This connection is used to identify LU
1451 * site and pin the OSD from early removal.
1453 * \param[in] env LU environment provided by the caller
1454 * \param[in] lod lod device
1455 * \param[in] cfg configuration command to apply
1457 * \retval 0 on success
1458 * \retval negative negated errno on error
1460 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
1461 struct lustre_cfg *cfg)
1463 struct obd_connect_data *data = NULL;
1464 struct obd_device *obd;
1465 char *nextdev = NULL, *p, *s;
1470 LASSERT(lod->lod_child_exp == NULL);
1472 /* compatibility hack: we still use old config logs
1473 * which specify LOV, but we need to learn underlying
1474 * OSD device, which is supposed to be:
1475 * <fsname>-MDTxxxx-osd
1477 * 2.x MGS generates lines like the following:
1478 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
1479 * 1.8 MGS generates lines like the following:
1480 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
1482 * we use "-MDT" to differentiate 2.x from 1.8 */
1484 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
1485 len = strlen(p) + 6;
1486 OBD_ALLOC(nextdev, len);
1487 if (nextdev == NULL)
1488 GOTO(out, rc = -ENOMEM);
1491 s = strstr(nextdev, "-mdtlov");
1492 if (unlikely(s == NULL)) {
1493 CERROR("unable to parse device name %s\n",
1494 lustre_cfg_string(cfg, 0));
1495 GOTO(out, rc = -EINVAL);
1498 if (strstr(nextdev, "-MDT")) {
1503 strcpy(s, "-MDT0000-osd");
1506 CERROR("unable to parse device name %s\n",
1507 lustre_cfg_string(cfg, 0));
1508 GOTO(out, rc = -EINVAL);
1511 OBD_ALLOC_PTR(data);
1513 GOTO(out, rc = -ENOMEM);
1515 obd = class_name2obd(nextdev);
1517 CERROR("can not locate next device: %s\n", nextdev);
1518 GOTO(out, rc = -ENOTCONN);
1521 data->ocd_connect_flags = OBD_CONNECT_VERSION;
1522 data->ocd_version = LUSTRE_VERSION_CODE;
1524 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
1527 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
1531 lod->lod_dt_dev.dd_lu_dev.ld_site =
1532 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
1533 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
1534 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
1540 OBD_FREE(nextdev, len);
1545 * Allocate and initialize target table.
1547 * A helper function to initialize the target table and allocate
1548 * a bitmap of the available targets.
1550 * \param[in] ltd target's table to initialize
1552 * \retval 0 on success
1553 * \retval negative negated errno on error
1555 static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
1557 mutex_init(<d->ltd_mutex);
1558 init_rwsem(<d->ltd_rw_sem);
1560 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
1561 * added to the LOD, see lod_add_device() */
1562 ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
1563 if (ltd->ltd_tgt_bitmap == NULL)
1566 ltd->ltd_tgts_size = 32;
1569 ltd->ltd_death_row = 0;
1570 ltd->ltd_refcount = 0;
1575 * Initialize LOD device at setup.
1577 * Initializes the given LOD device using the original configuration command.
1578 * The function initiates a connection to the local OSD and initializes few
1579 * internal structures like pools, target tables, etc.
1581 * \param[in] env LU environment provided by the caller
1582 * \param[in] lod lod device
1583 * \param[in] ldt not used
1584 * \param[in] cfg configuration command
1586 * \retval 0 on success
1587 * \retval negative negated errno on error
1589 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
1590 struct lu_device_type *ldt, struct lustre_cfg *cfg)
1592 struct dt_device_param ddp;
1593 struct obd_device *obd;
1597 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1599 CERROR("Cannot find obd with name %s\n",
1600 lustre_cfg_string(cfg, 0));
1604 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1605 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
1606 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
1607 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
1609 rc = lod_connect_to_osd(env, lod, cfg);
1613 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
1614 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
1616 /* setup obd to be used with old lov code */
1617 rc = lod_pools_init(lod, cfg);
1619 GOTO(out_disconnect, rc);
1621 rc = lod_procfs_init(lod);
1623 GOTO(out_pools, rc);
1625 spin_lock_init(&lod->lod_lock);
1626 spin_lock_init(&lod->lod_connects_lock);
1627 lod_tgt_desc_init(&lod->lod_mdt_descs);
1628 lod_tgt_desc_init(&lod->lod_ost_descs);
1633 lod_pools_fini(lod);
1635 obd_disconnect(lod->lod_child_exp);
1640 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
1642 * Releases the memory allocated for LOD device.
1644 * see include/lu_object.h for the details.
1646 static struct lu_device *lod_device_free(const struct lu_env *env,
1647 struct lu_device *lu)
1649 struct lod_device *lod = lu2lod_dev(lu);
1650 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1653 LASSERTF(atomic_read(&lu->ld_ref) == 0, "lu is %p\n", lu);
1654 dt_device_fini(&lod->lod_dt_dev);
1660 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
1662 * Allocates LOD device and calls the helpers to initialize it.
1664 * see include/lu_object.h for the details.
1666 static struct lu_device *lod_device_alloc(const struct lu_env *env,
1667 struct lu_device_type *type,
1668 struct lustre_cfg *lcfg)
1670 struct lod_device *lod;
1671 struct lu_device *lu_dev;
1675 lu_dev = ERR_PTR(-ENOMEM);
1679 lu_dev = lod2lu_dev(lod);
1680 dt_device_init(&lod->lod_dt_dev, type);
1681 rc = lod_init0(env, lod, type, lcfg);
1683 lod_device_free(env, lu_dev);
1684 lu_dev = ERR_PTR(rc);
1692 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
1694 * Releases the internal resources used by LOD device.
1696 * see include/lu_object.h for the details.
1698 static struct lu_device *lod_device_fini(const struct lu_env *env,
1699 struct lu_device *d)
1701 struct lod_device *lod = lu2lod_dev(d);
1705 lod_pools_fini(lod);
1707 lod_procfs_fini(lod);
1709 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs, true);
1711 CERROR("%s:can not fini ost descs %d\n",
1712 lod2obd(lod)->obd_name, rc);
1714 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs, false);
1716 CERROR("%s:can not fini mdt descs %d\n",
1717 lod2obd(lod)->obd_name, rc);
1723 * Implementation of obd_ops::o_connect() for LOD
1725 * Used to track all the users of this specific LOD device,
1726 * so the device stays up until the last user disconnected.
1728 * \param[in] env LU environment provided by the caller
1729 * \param[out] exp export the caller will be using to access LOD
1730 * \param[in] obd OBD device representing LOD device
1731 * \param[in] cluuid unique identifier of the caller
1732 * \param[in] data not used
1733 * \param[in] localdata not used
1735 * \retval 0 on success
1736 * \retval negative negated errno on error
1738 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
1739 struct obd_device *obd, struct obd_uuid *cluuid,
1740 struct obd_connect_data *data, void *localdata)
1742 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1743 struct lustre_handle conn;
1747 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1749 rc = class_connect(&conn, obd, cluuid);
1753 *exp = class_conn2export(&conn);
1755 spin_lock(&lod->lod_connects_lock);
1756 lod->lod_connects++;
1757 /* at the moment we expect the only user */
1758 LASSERT(lod->lod_connects == 1);
1759 spin_unlock(&lod->lod_connects_lock);
1766 * Implementation of obd_ops::o_disconnect() for LOD
1768 * When the caller doesn't need to use this LOD instance, it calls
1769 * obd_disconnect() and LOD releases corresponding export/reference count.
1770 * Once all the users gone, LOD device is released.
1772 * \param[in] exp export provided to the caller in obd_connect()
1774 * \retval 0 on success
1775 * \retval negative negated errno on error
1777 static int lod_obd_disconnect(struct obd_export *exp)
1779 struct obd_device *obd = exp->exp_obd;
1780 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1781 int rc, release = 0;
1784 /* Only disconnect the underlying layers on the final disconnect. */
1785 spin_lock(&lod->lod_connects_lock);
1786 lod->lod_connects--;
1787 if (lod->lod_connects != 0) {
1788 /* why should there be more than 1 connect? */
1789 spin_unlock(&lod->lod_connects_lock);
1790 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
1794 spin_unlock(&lod->lod_connects_lock);
1796 /* the last user of lod has gone, let's release the device */
1800 rc = class_disconnect(exp); /* bz 9811 */
1802 if (rc == 0 && release)
1803 class_manual_cleanup(obd);
1807 LU_KEY_INIT(lod, struct lod_thread_info);
1809 static void lod_key_fini(const struct lu_context *ctx,
1810 struct lu_context_key *key, void *data)
1812 struct lod_thread_info *info = data;
1813 /* allocated in lod_get_lov_ea
1814 * XXX: this is overload, a tread may have such store but used only
1815 * once. Probably better would be pool of such stores per LOD.
1817 if (info->lti_ea_store) {
1818 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
1819 info->lti_ea_store = NULL;
1820 info->lti_ea_store_size = 0;
1822 lu_buf_free(&info->lti_linkea_buf);
1826 /* context key: lod_thread_key */
1827 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
1829 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
1831 static struct lu_device_type_operations lod_device_type_ops = {
1832 .ldto_init = lod_type_init,
1833 .ldto_fini = lod_type_fini,
1835 .ldto_start = lod_type_start,
1836 .ldto_stop = lod_type_stop,
1838 .ldto_device_alloc = lod_device_alloc,
1839 .ldto_device_free = lod_device_free,
1841 .ldto_device_fini = lod_device_fini
1844 static struct lu_device_type lod_device_type = {
1845 .ldt_tags = LU_DEVICE_DT,
1846 .ldt_name = LUSTRE_LOD_NAME,
1847 .ldt_ops = &lod_device_type_ops,
1848 .ldt_ctx_tags = LCT_MD_THREAD,
1852 * Implementation of obd_ops::o_get_info() for LOD
1854 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
1855 * the caller binary status whether LOD has seen connection to any OST target.
1856 * It will also check if the MDT update log context being initialized (if
1859 * \param[in] env LU environment provided by the caller
1860 * \param[in] exp export of the caller
1861 * \param[in] keylen len of the key
1862 * \param[in] key the key
1863 * \param[in] vallen not used
1864 * \param[in] val not used
1866 * \retval 0 if a connection was seen
1867 * \retval -EAGAIN if LOD isn't running yet or no
1868 * connection has been seen yet
1869 * \retval -EINVAL if not supported key is requested
1871 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1872 __u32 keylen, void *key, __u32 *vallen, void *val)
1876 if (KEY_IS(KEY_OSP_CONNECTED)) {
1877 struct obd_device *obd = exp->exp_obd;
1878 struct lod_device *d;
1879 struct lod_tgt_desc *tgt;
1883 if (!obd->obd_set_up || obd->obd_stopping)
1886 d = lu2lod_dev(obd->obd_lu_dev);
1887 lod_getref(&d->lod_ost_descs);
1888 lod_foreach_ost(d, i) {
1889 tgt = OST_TGT(d, i);
1890 LASSERT(tgt && tgt->ltd_tgt);
1891 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
1893 /* one healthy device is enough */
1897 lod_putref(d, &d->lod_ost_descs);
1899 lod_getref(&d->lod_mdt_descs);
1900 lod_foreach_mdt(d, i) {
1901 struct llog_ctxt *ctxt;
1903 tgt = MDT_TGT(d, i);
1904 LASSERT(tgt != NULL);
1905 LASSERT(tgt->ltd_tgt != NULL);
1906 if (!tgt->ltd_active)
1909 ctxt = llog_get_context(tgt->ltd_tgt->dd_lu_dev.ld_obd,
1910 LLOG_UPDATELOG_ORIG_CTXT);
1912 CDEBUG(D_INFO, "%s: %s is not ready.\n",
1914 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
1918 if (ctxt->loc_handle == NULL) {
1919 CDEBUG(D_INFO, "%s: %s is not ready.\n",
1921 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
1923 llog_ctxt_put(ctxt);
1926 llog_ctxt_put(ctxt);
1928 lod_putref(d, &d->lod_mdt_descs);
1936 static int lod_obd_set_info_async(const struct lu_env *env,
1937 struct obd_export *exp,
1938 __u32 keylen, void *key,
1939 __u32 vallen, void *val,
1940 struct ptlrpc_request_set *set)
1942 struct obd_device *obd = class_exp2obd(exp);
1943 struct lod_device *d;
1944 struct lod_tgt_desc *tgt;
1951 set = ptlrpc_prep_set();
1956 d = lu2lod_dev(obd->obd_lu_dev);
1957 lod_getref(&d->lod_ost_descs);
1958 lod_foreach_ost(d, i) {
1959 tgt = OST_TGT(d, i);
1960 LASSERT(tgt && tgt->ltd_tgt);
1961 if (!tgt->ltd_active)
1964 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1966 if (rc2 != 0 && rc == 0)
1969 lod_putref(d, &d->lod_ost_descs);
1971 lod_getref(&d->lod_mdt_descs);
1972 lod_foreach_mdt(d, i) {
1973 tgt = MDT_TGT(d, i);
1974 LASSERT(tgt && tgt->ltd_tgt);
1975 if (!tgt->ltd_active)
1977 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1979 if (rc2 != 0 && rc == 0)
1982 lod_putref(d, &d->lod_mdt_descs);
1986 rc2 = ptlrpc_set_wait(set);
1987 if (rc2 == 0 && rc == 0)
1989 ptlrpc_set_destroy(set);
1994 static struct obd_ops lod_obd_device_ops = {
1995 .o_owner = THIS_MODULE,
1996 .o_connect = lod_obd_connect,
1997 .o_disconnect = lod_obd_disconnect,
1998 .o_get_info = lod_obd_get_info,
1999 .o_set_info_async = lod_obd_set_info_async,
2000 .o_pool_new = lod_pool_new,
2001 .o_pool_rem = lod_pool_remove,
2002 .o_pool_add = lod_pool_add,
2003 .o_pool_del = lod_pool_del,
2006 static int __init lod_init(void)
2008 struct obd_type *type;
2011 rc = lu_kmem_init(lod_caches);
2015 rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
2016 LUSTRE_LOD_NAME, &lod_device_type);
2018 lu_kmem_fini(lod_caches);
2022 /* create "lov" entry in procfs for compatibility purposes */
2023 type = class_search_type(LUSTRE_LOV_NAME);
2024 if (type != NULL && type->typ_procroot != NULL)
2027 type = class_search_type(LUSTRE_LOD_NAME);
2028 type->typ_procsym = lprocfs_register("lov", proc_lustre_root,
2030 if (IS_ERR(type->typ_procsym)) {
2031 CERROR("lod: can't create compat entry \"lov\": %d\n",
2032 (int)PTR_ERR(type->typ_procsym));
2033 type->typ_procsym = NULL;
2038 static void __exit lod_exit(void)
2040 class_unregister_type(LUSTRE_LOD_NAME);
2041 lu_kmem_fini(lod_caches);
2044 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2045 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
2046 MODULE_VERSION(LUSTRE_VERSION_STRING);
2047 MODULE_LICENSE("GPL");
2049 module_init(lod_init);
2050 module_exit(lod_exit);