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, 2017, 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 <uapi/linux/lustre/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;
130 if (!fid_is_sane(fid)) {
131 CERROR("%s: invalid FID "DFID"\n", lod2obd(lod)->obd_name,
136 if (fid_is_idif(fid)) {
137 *tgt = fid_idif_ost_idx(fid);
138 *type = LU_SEQ_RANGE_OST;
142 if (fid_is_update_log(fid) || fid_is_update_log_dir(fid)) {
144 *type = LU_SEQ_RANGE_MDT;
148 if (!lod->lod_initialized || (!fid_seq_in_fldb(fid_seq(fid)))) {
149 LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);
151 *tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
152 *type = LU_SEQ_RANGE_MDT;
156 server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
160 fld_range_set_type(&range, *type);
161 rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
165 *tgt = range.lsr_index;
166 *type = range.lsr_flags;
168 CDEBUG(D_INFO, "%s: got tgt %x for sequence: %#llx\n",
169 lod2obd(lod)->obd_name, *tgt, fid_seq(fid));
174 /* Slab for OSD object allocation */
175 struct kmem_cache *lod_object_kmem;
177 /* Slab for dt_txn_callback */
178 struct kmem_cache *lod_txn_callback_kmem;
179 static struct lu_kmem_descr lod_caches[] = {
181 .ckd_cache = &lod_object_kmem,
182 .ckd_name = "lod_obj",
183 .ckd_size = sizeof(struct lod_object)
186 .ckd_cache = &lod_txn_callback_kmem,
187 .ckd_name = "lod_txn_callback",
188 .ckd_size = sizeof(struct dt_txn_callback)
195 static struct lu_device *lod_device_fini(const struct lu_env *env,
196 struct lu_device *d);
199 * Implementation of lu_device_operations::ldo_object_alloc() for LOD
201 * Allocates and initializes LOD's slice in the given object.
203 * see include/lu_object.h for the details.
205 static struct lu_object *lod_object_alloc(const struct lu_env *env,
206 const struct lu_object_header *hdr,
207 struct lu_device *dev)
209 struct lod_object *lod_obj;
210 struct lu_object *lu_obj;
214 OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, GFP_NOFS);
216 RETURN(ERR_PTR(-ENOMEM));
218 mutex_init(&lod_obj->ldo_layout_mutex);
219 lu_obj = lod2lu_obj(lod_obj);
220 dt_object_init(&lod_obj->ldo_obj, NULL, dev);
221 lod_obj->ldo_obj.do_ops = &lod_obj_ops;
222 lu_obj->lo_ops = &lod_lu_obj_ops;
228 * Process the config log for all sub device.
230 * The function goes through all the targets in the given table
231 * and apply given configuration command on to the targets.
232 * Used to cleanup the targets at unmount.
234 * \param[in] env LU environment provided by the caller
235 * \param[in] lod lod device
236 * \param[in] ltd target's table to go through
237 * \param[in] lcfg configuration command to apply
239 * \retval 0 on success
240 * \retval negative negated errno on error
242 static int lod_sub_process_config(const struct lu_env *env,
243 struct lod_device *lod,
244 struct lod_tgt_descs *ltd,
245 struct lustre_cfg *lcfg)
247 struct lu_device *next;
252 if (ltd->ltd_tgts_size <= 0) {
253 lod_putref(lod, ltd);
256 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
257 struct lod_tgt_desc *tgt;
260 tgt = LTD_TGT(ltd, i);
261 LASSERT(tgt && tgt->ltd_tgt);
262 next = &tgt->ltd_tgt->dd_lu_dev;
263 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
265 CERROR("%s: error cleaning up LOD index %u: cmd %#x : rc = %d\n",
266 lod2obd(lod)->obd_name, i, lcfg->lcfg_command,
271 lod_putref(lod, ltd);
275 struct lod_recovery_data {
276 struct lod_device *lrd_lod;
277 struct lod_tgt_desc *lrd_ltd;
278 struct ptlrpc_thread *lrd_thread;
284 * process update recovery record
286 * Add the update recovery recode to the update recovery list in
287 * lod_recovery_data. Then the recovery thread (target_recovery_thread)
288 * will redo these updates.
290 * \param[in]env execution environment
291 * \param[in]llh log handle of update record
292 * \param[in]rec update record to be replayed
293 * \param[in]data update recovery data which holds the necessary
294 * arguments for recovery (see struct lod_recovery_data)
296 * \retval 0 if the record is processed successfully.
297 * \retval negative errno if the record processing fails.
299 static int lod_process_recovery_updates(const struct lu_env *env,
300 struct llog_handle *llh,
301 struct llog_rec_hdr *rec,
304 struct lod_recovery_data *lrd = data;
305 struct llog_cookie *cookie = &lod_env_info(env)->lti_cookie;
306 struct lu_target *lut;
314 rc = lodname2mdt_index(lod2obd(lrd->lrd_lod)->obd_name, &index);
318 index = lrd->lrd_ltd->ltd_index;
322 llog_update_record_size((struct llog_update_record *)rec)) {
323 CERROR("%s: broken update record! index %u "DFID".%u: rc = %d\n",
324 lod2obd(lrd->lrd_lod)->obd_name, index,
325 PFID(&llh->lgh_id.lgl_oi.oi_fid), rec->lrh_index, -EIO);
329 cookie->lgc_lgl = llh->lgh_id;
330 cookie->lgc_index = rec->lrh_index;
331 cookie->lgc_subsys = LLOG_UPDATELOG_ORIG_CTXT;
333 CDEBUG(D_HA, "%s: process recovery updates "DFID".%u\n",
334 lod2obd(lrd->lrd_lod)->obd_name,
335 PFID(&llh->lgh_id.lgl_oi.oi_fid), rec->lrh_index);
336 lut = lod2lu_dev(lrd->lrd_lod)->ld_site->ls_tgt;
338 if (lut->lut_obd->obd_stopping ||
339 lut->lut_obd->obd_abort_recovery)
342 return insert_update_records_to_replay_list(lut->lut_tdtd,
343 (struct llog_update_record *)rec,
348 * recovery thread for update log
350 * Start recovery thread and prepare the sub llog, then it will retrieve
351 * the update records from the correpondent MDT and do recovery.
353 * \param[in] arg pointer to the recovery data
355 * \retval 0 if recovery succeeds
356 * \retval negative errno if recovery failed.
358 static int lod_sub_recovery_thread(void *arg)
360 struct lod_recovery_data *lrd = arg;
361 struct lod_device *lod = lrd->lrd_lod;
362 struct dt_device *dt;
363 struct ptlrpc_thread *thread = lrd->lrd_thread;
364 struct llog_ctxt *ctxt = NULL;
366 struct lu_target *lut;
367 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
368 struct lod_tgt_desc *tgt = NULL;
376 thread->t_flags = SVC_RUNNING;
377 wake_up(&thread->t_ctl_waitq);
379 rc = lu_env_init(&env, LCT_LOCAL | LCT_MD_THREAD);
382 CERROR("%s: can't initialize env: rc = %d\n",
383 lod2obd(lod)->obd_name, rc);
387 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
388 atomic_inc(&lut->lut_tdtd->tdtd_recovery_threads_count);
392 dt = lrd->lrd_ltd->ltd_tgt;
394 start = ktime_get_real_seconds();
397 rc = lod_sub_prep_llog(&env, lod, dt, lrd->lrd_idx);
398 if (!rc && !lod->lod_child->dd_rdonly) {
399 /* Process the recovery record */
400 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
401 LLOG_UPDATELOG_ORIG_CTXT);
402 LASSERT(ctxt != NULL);
403 LASSERT(ctxt->loc_handle != NULL);
405 rc = llog_cat_process(&env, ctxt->loc_handle,
406 lod_process_recovery_updates, lrd, 0, 0);
410 struct lu_device *top_device;
412 top_device = lod->lod_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
414 * Because the remote target might failover at the same time,
417 if ((rc == -ETIMEDOUT || rc == -EAGAIN || rc == -EIO) &&
418 dt != lod->lod_child &&
419 !top_device->ld_obd->obd_abort_recovery &&
420 !top_device->ld_obd->obd_stopping) {
422 if (ctxt->loc_handle)
428 CDEBUG(D_HA, "%s get update log failed %d, retry\n",
429 dt->dd_lu_dev.ld_obd->obd_name, rc);
433 CERROR("%s get update log failed: rc = %d\n",
434 dt->dd_lu_dev.ld_obd->obd_name, rc);
437 spin_lock(&top_device->ld_obd->obd_dev_lock);
438 if (!top_device->ld_obd->obd_abort_recovery &&
439 !top_device->ld_obd->obd_stopping)
440 top_device->ld_obd->obd_abort_recovery = 1;
441 spin_unlock(&top_device->ld_obd->obd_dev_lock);
447 CDEBUG(D_HA, "%s retrieved update log, duration %lld, retries %d\n",
448 dt->dd_lu_dev.ld_obd->obd_name, ktime_get_real_seconds() - start,
451 spin_lock(&lod->lod_lock);
453 lod->lod_child_got_update_log = 1;
455 lrd->lrd_ltd->ltd_got_update_log = 1;
457 if (!lod->lod_child_got_update_log) {
458 spin_unlock(&lod->lod_lock);
462 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
463 tgt = LTD_TGT(ltd, i);
464 if (!tgt->ltd_got_update_log) {
465 spin_unlock(&lod->lod_lock);
469 lut->lut_tdtd->tdtd_replay_ready = 1;
470 spin_unlock(&lod->lod_lock);
472 CDEBUG(D_HA, "%s got update logs from all MDTs.\n",
473 lut->lut_obd->obd_name);
474 wake_up(&lut->lut_obd->obd_next_transno_waitq);
479 thread->t_flags = SVC_STOPPED;
480 atomic_dec(&lut->lut_tdtd->tdtd_recovery_threads_count);
481 wake_up(&lut->lut_tdtd->tdtd_recovery_threads_waitq);
482 wake_up(&thread->t_ctl_waitq);
488 * finish sub llog context
490 * Stop update recovery thread for the sub device, then cleanup the
491 * correspondent llog ctxt.
493 * \param[in] env execution environment
494 * \param[in] lod lod device to do update recovery
495 * \param[in] thread recovery thread on this sub device
497 void lod_sub_fini_llog(const struct lu_env *env,
498 struct dt_device *dt, struct ptlrpc_thread *thread)
500 struct obd_device *obd;
501 struct llog_ctxt *ctxt;
505 obd = dt->dd_lu_dev.ld_obd;
506 CDEBUG(D_INFO, "%s: finish sub llog\n", obd->obd_name);
507 /* Stop recovery thread first */
508 if (thread && thread->t_flags & SVC_RUNNING) {
509 thread->t_flags = SVC_STOPPING;
510 wake_up(&thread->t_ctl_waitq);
511 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
514 ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
518 if (ctxt->loc_handle)
519 llog_cat_close(env, ctxt->loc_handle);
521 llog_cleanup(env, ctxt);
527 * Extract MDT target index from a device name.
529 * a helper function to extract index from the given device name
530 * like "fsname-MDTxxxx-mdtlov"
532 * \param[in] lodname device name
533 * \param[out] mdt_index extracted index
535 * \retval 0 on success
536 * \retval -EINVAL if the name is invalid
538 int lodname2mdt_index(char *lodname, u32 *mdt_index)
544 /* 1.8 configs don't have "-MDT0000" at the end */
545 ptr = strstr(lodname, "-MDT");
551 ptr = strrchr(lodname, '-');
554 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
558 if (strncmp(ptr, "-mdtlov", 7) != 0) {
560 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
564 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
566 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
570 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
572 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
576 index = simple_strtol(ptr - 4, &tmp, 16);
577 if (*tmp != '-' || index > INT_MAX) {
579 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
587 * Init sub llog context
589 * Setup update llog ctxt for update recovery threads, then start the
590 * recovery thread (lod_sub_recovery_thread) to read update llog from
591 * the correspondent MDT to do update recovery.
593 * \param[in] env execution environment
594 * \param[in] lod lod device to do update recovery
595 * \param[in] dt sub dt device for which the recovery thread is
597 * \retval 0 if initialization succeeds.
598 * \retval negative errno if initialization fails.
600 int lod_sub_init_llog(const struct lu_env *env, struct lod_device *lod,
601 struct dt_device *dt)
603 struct obd_device *obd;
604 struct lod_recovery_data *lrd = NULL;
605 struct ptlrpc_thread *thread;
606 struct task_struct *task;
607 struct l_wait_info lwi = { 0 };
608 struct lod_tgt_desc *sub_ltd = NULL;
615 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &master_index);
623 if (lod->lod_child == dt) {
624 thread = &lod->lod_child_recovery_thread;
625 index = master_index;
627 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
628 struct lod_tgt_desc *tgt = NULL;
631 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
632 tgt = LTD_TGT(ltd, i);
633 if (tgt->ltd_tgt == dt) {
634 index = tgt->ltd_index;
639 LASSERT(sub_ltd != NULL);
640 OBD_ALLOC_PTR(sub_ltd->ltd_recovery_thread);
641 if (!sub_ltd->ltd_recovery_thread)
642 GOTO(free_lrd, rc = -ENOMEM);
644 thread = sub_ltd->ltd_recovery_thread;
647 CDEBUG(D_INFO, "%s init sub log %s\n", lod2obd(lod)->obd_name,
648 dt->dd_lu_dev.ld_obd->obd_name);
650 lrd->lrd_ltd = sub_ltd;
651 lrd->lrd_thread = thread;
652 lrd->lrd_idx = index;
653 init_waitqueue_head(&thread->t_ctl_waitq);
655 obd = dt->dd_lu_dev.ld_obd;
656 obd->obd_lvfs_ctxt.dt = dt;
657 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_UPDATELOG_ORIG_CTXT,
658 NULL, &llog_common_cat_ops);
660 CERROR("%s: cannot setup updatelog llog: rc = %d\n",
662 GOTO(free_thread, rc);
665 /* Start the recovery thread */
666 task = kthread_run(lod_sub_recovery_thread, lrd, "lod%04x_rec%04x",
667 master_index, index);
670 CERROR("%s: cannot start recovery thread: rc = %d\n",
675 l_wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_RUNNING ||
676 thread->t_flags & SVC_STOPPED, &lwi);
680 lod_sub_fini_llog(env, dt, thread);
682 if (lod->lod_child != dt) {
683 OBD_FREE_PTR(sub_ltd->ltd_recovery_thread);
684 sub_ltd->ltd_recovery_thread = NULL;
692 * Stop sub recovery thread
694 * Stop sub recovery thread on all subs.
696 * \param[in] env execution environment
697 * \param[in] lod lod device to do update recovery
699 static void lod_sub_stop_recovery_threads(const struct lu_env *env,
700 struct lod_device *lod)
702 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
703 struct ptlrpc_thread *thread;
707 * Stop the update log commit cancel threads and finish master
710 thread = &lod->lod_child_recovery_thread;
711 /* Stop recovery thread first */
712 if (thread && thread->t_flags & SVC_RUNNING) {
713 thread->t_flags = SVC_STOPPING;
714 wake_up(&thread->t_ctl_waitq);
715 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
719 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
720 struct lod_tgt_desc *tgt;
722 tgt = LTD_TGT(ltd, i);
723 thread = tgt->ltd_recovery_thread;
724 if (thread && thread->t_flags & SVC_RUNNING) {
725 thread->t_flags = SVC_STOPPING;
726 wake_up(&thread->t_ctl_waitq);
727 wait_event(thread->t_ctl_waitq,
728 thread->t_flags & SVC_STOPPED);
729 OBD_FREE_PTR(tgt->ltd_recovery_thread);
730 tgt->ltd_recovery_thread = NULL;
734 lod_putref(lod, ltd);
738 * finish all sub llog
740 * cleanup all of sub llog ctxt on the LOD.
742 * \param[in] env execution environment
743 * \param[in] lod lod device to do update recovery
745 static void lod_sub_fini_all_llogs(const struct lu_env *env,
746 struct lod_device *lod)
748 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
752 * Stop the update log commit cancel threads and finish master
755 lod_sub_fini_llog(env, lod->lod_child,
756 &lod->lod_child_recovery_thread);
758 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
759 struct lod_tgt_desc *tgt;
761 tgt = LTD_TGT(ltd, i);
762 lod_sub_fini_llog(env, tgt->ltd_tgt,
763 tgt->ltd_recovery_thread);
766 lod_putref(lod, ltd);
769 static char *lod_show_update_logs_retrievers(void *data, int *size, int *count)
771 struct lod_device *lod = (struct lod_device *)data;
772 struct lu_target *lut = lod2lu_dev(lod)->ld_site->ls_tgt;
773 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
774 struct lod_tgt_desc *tgt = NULL;
780 *count = atomic_read(&lut->lut_tdtd->tdtd_recovery_threads_count);
786 *size = 5 * *count + 1;
787 OBD_ALLOC(buf, *size);
792 memset(buf, 0, *size);
794 if (!lod->lod_child_got_update_log) {
795 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &i);
796 LASSERTF(rc == 0, "Fail to parse target index: rc = %d\n", rc);
798 rc = snprintf(buf + len, *size - len, " %04x", i);
805 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
806 tgt = LTD_TGT(ltd, i);
807 if (!tgt->ltd_got_update_log) {
808 rc = snprintf(buf + len, *size - len, " %04x", i);
809 if (unlikely(rc <= 0))
821 * Prepare distribute txn
823 * Prepare distribute txn structure for LOD
825 * \param[in] env execution environment
826 * \param[in] lod_device LOD device
828 * \retval 0 if preparation succeeds.
829 * \retval negative errno if preparation fails.
831 static int lod_prepare_distribute_txn(const struct lu_env *env,
832 struct lod_device *lod)
834 struct target_distribute_txn_data *tdtd;
835 struct lu_target *lut;
840 /* Init update recovery data */
845 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
846 tdtd->tdtd_dt = &lod->lod_dt_dev;
847 rc = distribute_txn_init(env, lut, tdtd,
848 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id);
851 CERROR("%s: cannot init distribute txn: rc = %d\n",
852 lod2obd(lod)->obd_name, rc);
857 tdtd->tdtd_show_update_logs_retrievers =
858 lod_show_update_logs_retrievers;
859 tdtd->tdtd_show_retrievers_cbdata = lod;
861 lut->lut_tdtd = tdtd;
867 * Finish distribute txn
869 * Release the resource holding by distribute txn, i.e. stop distribute
872 * \param[in] env execution environment
873 * \param[in] lod lod device
875 static void lod_fini_distribute_txn(const struct lu_env *env,
876 struct lod_device *lod)
878 struct lu_target *lut;
880 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
881 target_recovery_fini(lut->lut_obd);
885 distribute_txn_fini(env, lut->lut_tdtd);
887 OBD_FREE_PTR(lut->lut_tdtd);
888 lut->lut_tdtd = NULL;
892 * Implementation of lu_device_operations::ldo_process_config() for LOD
894 * The method is called by the configuration subsystem during setup,
895 * cleanup and when the configuration changes. The method processes
896 * few specific commands like adding/removing the targets, changing
897 * the runtime parameters.
899 * \param[in] env LU environment provided by the caller
900 * \param[in] dev lod device
901 * \param[in] lcfg configuration command to apply
903 * \retval 0 on success
904 * \retval negative negated errno on error
906 * The examples are below.
908 * Add osc config log:
909 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
910 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
911 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
912 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
913 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
914 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
916 * Add mdc config log:
917 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
918 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
919 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
920 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
921 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
922 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
924 static int lod_process_config(const struct lu_env *env,
925 struct lu_device *dev,
926 struct lustre_cfg *lcfg)
928 struct lod_device *lod = lu2lod_dev(dev);
929 struct lu_device *next = &lod->lod_child->dd_lu_dev;
935 switch (lcfg->lcfg_command) {
936 case LCFG_LOV_DEL_OBD:
937 case LCFG_LOV_ADD_INA:
938 case LCFG_LOV_ADD_OBD:
944 * lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
945 * modify_mdc_tgts add 0:lustre-MDT0001
946 * 1:lustre-MDT0001-mdc0002
949 arg1 = lustre_cfg_string(lcfg, 1);
951 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
952 GOTO(out, rc = -EINVAL);
953 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
954 GOTO(out, rc = -EINVAL);
956 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
959 rc = lodname2mdt_index(lustre_cfg_string(lcfg, 0),
964 rc = lod_add_device(env, lod, arg1, index, gen,
965 mdt_index, LUSTRE_OSC_NAME, 1);
966 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
968 rc = lod_add_device(env, lod, arg1, index, gen,
969 mdt_index, LUSTRE_MDC_NAME, 1);
970 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
971 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
973 rc = lod_add_device(env, lod, arg1, index, gen,
974 mdt_index, LUSTRE_OSC_NAME, 0);
976 rc = lod_del_device(env, lod,
978 arg1, index, gen, true);
985 struct obd_device *obd;
990 * Check if it is activate/deactivate mdc
991 * lustre-MDTXXXX-osp-MDTXXXX.active=1
993 param = lustre_cfg_buf(lcfg, 1);
994 if (strstr(param, "osp") && strstr(param, ".active=")) {
995 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
996 struct lod_tgt_desc *sub_tgt = NULL;
1001 ptr = strstr(param, ".");
1003 obd = class_name2obd(param);
1005 CERROR("%s: can not find %s: rc = %d\n",
1006 lod2obd(lod)->obd_name, param, -EINVAL);
1011 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
1012 struct lod_tgt_desc *tgt;
1014 tgt = LTD_TGT(ltd, i);
1015 if (tgt->ltd_tgt->dd_lu_dev.ld_obd == obd) {
1022 CERROR("%s: can not find %s: rc = %d\n",
1023 lod2obd(lod)->obd_name, param, -EINVAL);
1029 tmp = strstr(param, "=");
1032 struct llog_ctxt *ctxt;
1034 obd = sub_tgt->ltd_tgt->dd_lu_dev.ld_obd;
1035 ctxt = llog_get_context(obd,
1036 LLOG_UPDATELOG_ORIG_CTXT);
1038 rc = llog_setup(env, obd, &obd->obd_olg,
1039 LLOG_UPDATELOG_ORIG_CTXT,
1040 NULL, &llog_common_cat_ops);
1044 llog_ctxt_put(ctxt);
1046 rc = lod_sub_prep_llog(env, lod,
1048 sub_tgt->ltd_index);
1050 sub_tgt->ltd_active = 1;
1052 lod_sub_fini_llog(env, sub_tgt->ltd_tgt,
1054 sub_tgt->ltd_active = 0;
1060 if (strstr(param, PARAM_LOD) != NULL)
1061 count = class_modify_config(lcfg, PARAM_LOD,
1062 &lod->lod_dt_dev.dd_kobj);
1064 count = class_modify_config(lcfg, PARAM_LOV,
1065 &lod->lod_dt_dev.dd_kobj);
1066 rc = count > 0 ? 0 : count;
1069 case LCFG_PRE_CLEANUP: {
1070 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1071 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1072 next = &lod->lod_child->dd_lu_dev;
1073 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1075 CDEBUG(D_HA, "%s: can't process %u: %d\n",
1076 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1078 lod_sub_stop_recovery_threads(env, lod);
1079 lod_fini_distribute_txn(env, lod);
1080 lod_sub_fini_all_llogs(env, lod);
1083 case LCFG_CLEANUP: {
1084 if (lod->lod_md_root) {
1085 dt_object_put(env, &lod->lod_md_root->ldo_obj);
1086 lod->lod_md_root = NULL;
1090 * do cleanup on underlying storage only when
1091 * all OSPs are cleaned up, as they use that OSD as well
1093 lu_dev_del_linkage(dev->ld_site, dev);
1094 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1095 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1096 next = &lod->lod_child->dd_lu_dev;
1097 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1099 CERROR("%s: can't process %u: rc = %d\n",
1100 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1102 rc = obd_disconnect(lod->lod_child_exp);
1104 CERROR("error in disconnect from storage: rc = %d\n",
1109 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
1110 lcfg->lcfg_command);
1120 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
1122 * The method is called once the recovery is complete. This implementation
1123 * distributes the notification to all the known targets.
1125 * see include/lu_object.h for the details
1127 static int lod_recovery_complete(const struct lu_env *env,
1128 struct lu_device *dev)
1130 struct lod_device *lod = lu2lod_dev(dev);
1131 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1137 LASSERT(lod->lod_recovery_completed == 0);
1138 lod->lod_recovery_completed = 1;
1140 rc = next->ld_ops->ldo_recovery_complete(env, next);
1142 lod_getref(&lod->lod_ost_descs);
1143 if (lod->lod_osts_size > 0) {
1144 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
1145 struct lod_tgt_desc *tgt;
1147 tgt = OST_TGT(lod, i);
1148 LASSERT(tgt && tgt->ltd_tgt);
1149 next = &tgt->ltd_ost->dd_lu_dev;
1150 rc = next->ld_ops->ldo_recovery_complete(env, next);
1152 CERROR("%s: can't complete recovery on #%d: rc = %d\n",
1153 lod2obd(lod)->obd_name, i, rc);
1156 lod_putref(lod, &lod->lod_ost_descs);
1161 * Init update logs on all sub device
1163 * LOD initialize update logs on all of sub devices. Because the initialization
1164 * process might need FLD lookup, see llog_osd_open()->dt_locate()->...->
1165 * lod_object_init(), this API has to be called after LOD is initialized.
1166 * \param[in] env execution environment
1167 * \param[in] lod lod device
1169 * \retval 0 if update log is initialized successfully.
1170 * \retval negative errno if initialization fails.
1172 static int lod_sub_init_llogs(const struct lu_env *env, struct lod_device *lod)
1174 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1181 * llog must be setup after LOD is initialized, because llog
1182 * initialization include FLD lookup
1184 LASSERT(lod->lod_initialized);
1186 /* Init the llog in its own stack */
1187 rc = lod_sub_init_llog(env, lod, lod->lod_child);
1191 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
1192 struct lod_tgt_desc *tgt;
1194 tgt = LTD_TGT(ltd, i);
1195 rc = lod_sub_init_llog(env, lod, tgt->ltd_tgt);
1204 * Implementation of lu_device_operations::ldo_prepare() for LOD
1206 * see include/lu_object.h for the details.
1208 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
1209 struct lu_device *cdev)
1211 struct lod_device *lod = lu2lod_dev(cdev);
1212 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1213 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1215 struct dt_object *root;
1216 struct dt_object *dto;
1221 rc = next->ld_ops->ldo_prepare(env, pdev, next);
1223 CERROR("%s: prepare bottom error: rc = %d\n",
1224 lod2obd(lod)->obd_name, rc);
1228 lod->lod_initialized = 1;
1230 rc = dt_root_get(env, lod->lod_child, fid);
1234 root = dt_locate(env, lod->lod_child, fid);
1236 RETURN(PTR_ERR(root));
1238 /* Create update log object */
1239 index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1240 lu_update_log_fid(fid, index);
1242 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1244 lod_update_log_name,
1247 GOTO(out_put, rc = PTR_ERR(dto));
1249 dt_object_put(env, dto);
1251 /* Create update log dir */
1252 lu_update_log_dir_fid(fid, index);
1253 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1255 lod_update_log_dir_name,
1258 GOTO(out_put, rc = PTR_ERR(dto));
1260 dt_object_put(env, dto);
1262 rc = lod_prepare_distribute_txn(env, lod);
1266 rc = lod_sub_init_llogs(env, lod);
1271 dt_object_put(env, root);
1276 const struct lu_device_operations lod_lu_ops = {
1277 .ldo_object_alloc = lod_object_alloc,
1278 .ldo_process_config = lod_process_config,
1279 .ldo_recovery_complete = lod_recovery_complete,
1280 .ldo_prepare = lod_prepare,
1284 * Implementation of dt_device_operations::dt_root_get() for LOD
1286 * see include/dt_object.h for the details.
1288 static int lod_root_get(const struct lu_env *env,
1289 struct dt_device *dev, struct lu_fid *f)
1291 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
1294 static void lod_statfs_sum(struct obd_statfs *sfs,
1295 struct obd_statfs *ost_sfs, int *bs)
1297 while (ost_sfs->os_bsize < *bs) {
1299 sfs->os_bsize >>= 1;
1300 sfs->os_bavail <<= 1;
1301 sfs->os_blocks <<= 1;
1302 sfs->os_bfree <<= 1;
1303 sfs->os_granted <<= 1;
1305 while (ost_sfs->os_bsize > *bs) {
1306 ost_sfs->os_bsize >>= 1;
1307 ost_sfs->os_bavail <<= 1;
1308 ost_sfs->os_blocks <<= 1;
1309 ost_sfs->os_bfree <<= 1;
1310 ost_sfs->os_granted <<= 1;
1312 sfs->os_bavail += ost_sfs->os_bavail;
1313 sfs->os_blocks += ost_sfs->os_blocks;
1314 sfs->os_bfree += ost_sfs->os_bfree;
1315 sfs->os_granted += ost_sfs->os_granted;
1319 * Implementation of dt_device_operations::dt_statfs() for LOD
1321 * see include/dt_object.h for the details.
1323 static int lod_statfs(const struct lu_env *env,
1324 struct dt_device *dev, struct obd_statfs *sfs)
1326 struct lod_device *lod = dt2lod_dev(dev);
1327 struct lod_ost_desc *ost;
1328 struct lod_mdt_desc *mdt;
1329 struct obd_statfs ost_sfs;
1333 rc = dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
1342 sfs->os_granted = 0;
1344 lod_getref(&lod->lod_mdt_descs);
1345 lod_foreach_mdt(lod, i) {
1346 mdt = MDT_TGT(lod, i);
1347 LASSERT(mdt && mdt->ltd_mdt);
1348 rc = dt_statfs(env, mdt->ltd_mdt, &ost_sfs);
1352 sfs->os_files += ost_sfs.os_files;
1353 sfs->os_ffree += ost_sfs.os_ffree;
1354 lod_statfs_sum(sfs, &ost_sfs, &bs);
1356 lod_putref(lod, &lod->lod_mdt_descs);
1359 * at some point we can check whether DoM is enabled and
1360 * decide how to account MDT space. for simplicity let's
1361 * just fallback to pre-DoM policy if any OST is alive
1365 lod_getref(&lod->lod_ost_descs);
1366 lod_foreach_ost(lod, i) {
1367 ost = OST_TGT(lod, i);
1368 LASSERT(ost && ost->ltd_ost);
1369 rc = dt_statfs(env, ost->ltd_ost, &ost_sfs);
1371 if (rc || ost_sfs.os_bsize == 0)
1375 * if only MDTs and DoM report MDT space,
1376 * otherwise only OST space
1381 sfs->os_granted = 0;
1384 ost_sfs.os_bavail += ost_sfs.os_granted;
1385 lod_statfs_sum(sfs, &ost_sfs, &bs);
1386 LASSERTF(bs == ost_sfs.os_bsize, "%d != %d\n",
1387 (int)sfs->os_bsize, (int)ost_sfs.os_bsize);
1389 lod_putref(lod, &lod->lod_ost_descs);
1390 sfs->os_state |= OS_STATE_SUM;
1392 /* a single successful statfs should be enough */
1400 * Implementation of dt_device_operations::dt_trans_create() for LOD
1402 * Creates a transaction using local (to this node) OSD.
1404 * see include/dt_object.h for the details.
1406 static struct thandle *lod_trans_create(const struct lu_env *env,
1407 struct dt_device *dt)
1411 th = top_trans_create(env, dt2lod_dev(dt)->lod_child);
1421 * Implementation of dt_device_operations::dt_trans_start() for LOD
1423 * Starts the set of local transactions using the targets involved
1424 * in declare phase. Initial support for the distributed transactions.
1426 * see include/dt_object.h for the details.
1428 static int lod_trans_start(const struct lu_env *env, struct dt_device *dt,
1431 return top_trans_start(env, dt2lod_dev(dt)->lod_child, th);
1434 static int lod_trans_cb_add(struct thandle *th,
1435 struct dt_txn_commit_cb *dcb)
1437 struct top_thandle *top_th = container_of(th, struct top_thandle,
1439 return dt_trans_cb_add(top_th->tt_master_sub_thandle, dcb);
1443 * add noop update to the update records
1445 * Add noop updates to the update records, which is only used in
1448 * \param[in] env execution environment
1449 * \param[in] dt dt device of lod
1450 * \param[in] th thandle
1451 * \param[in] count the count of update records to be added.
1453 * \retval 0 if adding succeeds.
1454 * \retval negative errno if adding fails.
1456 static int lod_add_noop_records(const struct lu_env *env,
1457 struct dt_device *dt, struct thandle *th,
1460 struct top_thandle *top_th;
1461 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1465 top_th = container_of(th, struct top_thandle, tt_super);
1466 if (!top_th->tt_multiple_thandle)
1470 for (i = 0; i < count; i++) {
1471 rc = update_record_pack(noop, th, fid);
1479 * Implementation of dt_device_operations::dt_trans_stop() for LOD
1481 * Stops the set of local transactions using the targets involved
1482 * in declare phase. Initial support for the distributed transactions.
1484 * see include/dt_object.h for the details.
1486 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
1489 if (OBD_FAIL_CHECK(OBD_FAIL_SPLIT_UPDATE_REC)) {
1492 rc = lod_add_noop_records(env, dt, th, 5000);
1496 return top_trans_stop(env, dt2lod_dev(dt)->lod_child, th);
1500 * Implementation of dt_device_operations::dt_conf_get() for LOD
1502 * Currently returns the configuration provided by the local OSD.
1504 * see include/dt_object.h for the details.
1506 static void lod_conf_get(const struct lu_env *env,
1507 const struct dt_device *dev,
1508 struct dt_device_param *param)
1510 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
1514 * Implementation of dt_device_operations::dt_sync() for LOD
1516 * Syncs all known OST targets. Very very expensive and used
1517 * rarely by LFSCK now. Should not be used in general.
1519 * see include/dt_object.h for the details.
1521 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
1523 struct lod_device *lod = dt2lod_dev(dev);
1524 struct lod_ost_desc *ost;
1525 struct lod_mdt_desc *mdt;
1531 lod_getref(&lod->lod_ost_descs);
1532 lod_foreach_ost(lod, i) {
1533 ost = OST_TGT(lod, i);
1534 LASSERT(ost && ost->ltd_ost);
1535 if (!ost->ltd_active)
1537 rc = dt_sync(env, ost->ltd_ost);
1539 if (rc != -ENOTCONN) {
1540 CERROR("%s: can't sync ost %u: rc = %d\n",
1541 lod2obd(lod)->obd_name, i, rc);
1547 lod_putref(lod, &lod->lod_ost_descs);
1552 lod_getref(&lod->lod_mdt_descs);
1553 lod_foreach_mdt(lod, i) {
1554 mdt = MDT_TGT(lod, i);
1555 LASSERT(mdt && mdt->ltd_mdt);
1556 if (!mdt->ltd_active)
1558 rc = dt_sync(env, mdt->ltd_mdt);
1560 if (rc != -ENOTCONN) {
1561 CERROR("%s: can't sync mdt %u: rc = %d\n",
1562 lod2obd(lod)->obd_name, i, rc);
1568 lod_putref(lod, &lod->lod_mdt_descs);
1571 rc = dt_sync(env, lod->lod_child);
1577 * Implementation of dt_device_operations::dt_ro() for LOD
1579 * Turns local OSD read-only, used for the testing only.
1581 * see include/dt_object.h for the details.
1583 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
1585 return dt_ro(env, dt2lod_dev(dev)->lod_child);
1589 * Implementation of dt_device_operations::dt_commit_async() for LOD
1591 * Asks local OSD to commit sooner.
1593 * see include/dt_object.h for the details.
1595 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
1597 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
1600 static const struct dt_device_operations lod_dt_ops = {
1601 .dt_root_get = lod_root_get,
1602 .dt_statfs = lod_statfs,
1603 .dt_trans_create = lod_trans_create,
1604 .dt_trans_start = lod_trans_start,
1605 .dt_trans_stop = lod_trans_stop,
1606 .dt_conf_get = lod_conf_get,
1607 .dt_sync = lod_sync,
1609 .dt_commit_async = lod_commit_async,
1610 .dt_trans_cb_add = lod_trans_cb_add,
1614 * Connect to a local OSD.
1616 * Used to connect to the local OSD at mount. OSD name is taken from the
1617 * configuration command passed. This connection is used to identify LU
1618 * site and pin the OSD from early removal.
1620 * \param[in] env LU environment provided by the caller
1621 * \param[in] lod lod device
1622 * \param[in] cfg configuration command to apply
1624 * \retval 0 on success
1625 * \retval negative negated errno on error
1627 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
1628 struct lustre_cfg *cfg)
1630 struct obd_connect_data *data = NULL;
1631 struct obd_device *obd;
1632 char *nextdev = NULL, *p, *s;
1638 LASSERT(lod->lod_child_exp == NULL);
1641 * compatibility hack: we still use old config logs
1642 * which specify LOV, but we need to learn underlying
1643 * OSD device, which is supposed to be:
1644 * <fsname>-MDTxxxx-osd
1646 * 2.x MGS generates lines like the following:
1647 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
1648 * 1.8 MGS generates lines like the following:
1649 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
1651 * we use "-MDT" to differentiate 2.x from 1.8
1653 p = lustre_cfg_string(cfg, 0);
1654 if (p && strstr(p, "-mdtlov")) {
1655 len = strlen(p) + 6;
1656 OBD_ALLOC(nextdev, len);
1658 GOTO(out, rc = -ENOMEM);
1661 s = strstr(nextdev, "-mdtlov");
1663 CERROR("%s: unable to parse device name: rc = %d\n",
1664 lustre_cfg_string(cfg, 0), -EINVAL);
1665 GOTO(out, rc = -EINVAL);
1668 if (strstr(nextdev, "-MDT")) {
1673 strcpy(s, "-MDT0000-osd");
1676 CERROR("%s: unable to parse device name: rc = %d\n",
1677 lustre_cfg_string(cfg, 0), -EINVAL);
1678 GOTO(out, rc = -EINVAL);
1681 OBD_ALLOC_PTR(data);
1683 GOTO(out, rc = -ENOMEM);
1685 obd = class_name2obd(nextdev);
1687 CERROR("%s: can not locate next device: rc = %d\n",
1688 nextdev, -ENOTCONN);
1689 GOTO(out, rc = -ENOTCONN);
1692 data->ocd_connect_flags = OBD_CONNECT_VERSION;
1693 data->ocd_version = LUSTRE_VERSION_CODE;
1695 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
1698 CERROR("%s: cannot connect to next dev: rc = %d\n",
1703 lod->lod_dt_dev.dd_lu_dev.ld_site =
1704 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
1705 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
1706 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
1712 OBD_FREE(nextdev, len);
1717 * Allocate and initialize target table.
1719 * A helper function to initialize the target table and allocate
1720 * a bitmap of the available targets.
1722 * \param[in] ltd target's table to initialize
1724 * \retval 0 on success
1725 * \retval negative negated errno on error
1727 static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
1729 mutex_init(<d->ltd_mutex);
1730 init_rwsem(<d->ltd_rw_sem);
1733 * the OST array and bitmap are allocated/grown dynamically as OSTs are
1734 * added to the LOD, see lod_add_device()
1736 ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
1737 if (!ltd->ltd_tgt_bitmap)
1740 ltd->ltd_tgts_size = 32;
1743 ltd->ltd_death_row = 0;
1744 ltd->ltd_refcount = 0;
1749 * Initialize LOD device at setup.
1751 * Initializes the given LOD device using the original configuration command.
1752 * The function initiates a connection to the local OSD and initializes few
1753 * internal structures like pools, target tables, etc.
1755 * \param[in] env LU environment provided by the caller
1756 * \param[in] lod lod device
1757 * \param[in] ldt not used
1758 * \param[in] cfg configuration command
1760 * \retval 0 on success
1761 * \retval negative negated errno on error
1763 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
1764 struct lu_device_type *ldt, struct lustre_cfg *cfg)
1766 struct dt_device_param ddp;
1767 struct obd_device *obd;
1772 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1775 CERROR("Cannot find obd with name '%s': rc = %d\n",
1776 lustre_cfg_string(cfg, 0), rc);
1780 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1781 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
1782 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
1783 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
1785 rc = lod_connect_to_osd(env, lod, cfg);
1789 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
1790 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
1791 lod->lod_dom_max_stripesize = (1ULL << 20); /* 1Mb as default value */
1793 /* setup obd to be used with old lov code */
1794 rc = lod_pools_init(lod, cfg);
1796 GOTO(out_disconnect, rc);
1798 rc = lod_procfs_init(lod);
1800 GOTO(out_pools, rc);
1802 spin_lock_init(&lod->lod_lock);
1803 spin_lock_init(&lod->lod_connects_lock);
1804 lod_tgt_desc_init(&lod->lod_mdt_descs);
1805 lod_tgt_desc_init(&lod->lod_ost_descs);
1810 lod_pools_fini(lod);
1812 obd_disconnect(lod->lod_child_exp);
1817 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
1819 * Releases the memory allocated for LOD device.
1821 * see include/lu_object.h for the details.
1823 static struct lu_device *lod_device_free(const struct lu_env *env,
1824 struct lu_device *lu)
1826 struct lod_device *lod = lu2lod_dev(lu);
1827 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1831 if (atomic_read(&lu->ld_ref) > 0 &&
1832 !cfs_hash_is_empty(lu->ld_site->ls_obj_hash)) {
1833 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1834 lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
1836 LASSERTF(atomic_read(&lu->ld_ref) == 0, "lu is %p\n", lu);
1837 dt_device_fini(&lod->lod_dt_dev);
1843 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
1845 * Allocates LOD device and calls the helpers to initialize it.
1847 * see include/lu_object.h for the details.
1849 static struct lu_device *lod_device_alloc(const struct lu_env *env,
1850 struct lu_device_type *type,
1851 struct lustre_cfg *lcfg)
1853 struct lod_device *lod;
1854 struct lu_device *lu_dev;
1858 lu_dev = ERR_PTR(-ENOMEM);
1862 lu_dev = lod2lu_dev(lod);
1863 dt_device_init(&lod->lod_dt_dev, type);
1864 rc = lod_init0(env, lod, type, lcfg);
1866 lod_device_free(env, lu_dev);
1867 lu_dev = ERR_PTR(rc);
1874 static void lod_avoid_guide_fini(struct lod_avoid_guide *lag)
1876 if (lag->lag_oss_avoid_array)
1877 OBD_FREE(lag->lag_oss_avoid_array,
1878 sizeof(u32) * lag->lag_oaa_size);
1879 if (lag->lag_ost_avoid_bitmap)
1880 CFS_FREE_BITMAP(lag->lag_ost_avoid_bitmap);
1884 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
1886 * Releases the internal resources used by LOD device.
1888 * see include/lu_object.h for the details.
1890 static struct lu_device *lod_device_fini(const struct lu_env *env,
1891 struct lu_device *d)
1893 struct lod_device *lod = lu2lod_dev(d);
1898 lod_pools_fini(lod);
1900 lod_procfs_fini(lod);
1902 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs, true);
1904 CERROR("%s: can not fini ost descriptors: rc = %d\n",
1905 lod2obd(lod)->obd_name, rc);
1907 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs, false);
1909 CERROR("%s: can not fini mdt descriptors: rc = %d\n",
1910 lod2obd(lod)->obd_name, rc);
1916 * Implementation of obd_ops::o_connect() for LOD
1918 * Used to track all the users of this specific LOD device,
1919 * so the device stays up until the last user disconnected.
1921 * \param[in] env LU environment provided by the caller
1922 * \param[out] exp export the caller will be using to access LOD
1923 * \param[in] obd OBD device representing LOD device
1924 * \param[in] cluuid unique identifier of the caller
1925 * \param[in] data not used
1926 * \param[in] localdata not used
1928 * \retval 0 on success
1929 * \retval negative negated errno on error
1931 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
1932 struct obd_device *obd, struct obd_uuid *cluuid,
1933 struct obd_connect_data *data, void *localdata)
1935 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1936 struct lustre_handle conn;
1941 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1943 rc = class_connect(&conn, obd, cluuid);
1947 *exp = class_conn2export(&conn);
1949 spin_lock(&lod->lod_connects_lock);
1950 lod->lod_connects++;
1951 /* at the moment we expect the only user */
1952 LASSERT(lod->lod_connects == 1);
1953 spin_unlock(&lod->lod_connects_lock);
1960 * Implementation of obd_ops::o_disconnect() for LOD
1962 * When the caller doesn't need to use this LOD instance, it calls
1963 * obd_disconnect() and LOD releases corresponding export/reference count.
1964 * Once all the users gone, LOD device is released.
1966 * \param[in] exp export provided to the caller in obd_connect()
1968 * \retval 0 on success
1969 * \retval negative negated errno on error
1971 static int lod_obd_disconnect(struct obd_export *exp)
1973 struct obd_device *obd = exp->exp_obd;
1974 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1975 int rc, release = 0;
1979 /* Only disconnect the underlying layers on the final disconnect. */
1980 spin_lock(&lod->lod_connects_lock);
1981 lod->lod_connects--;
1982 if (lod->lod_connects != 0) {
1983 /* why should there be more than 1 connect? */
1984 spin_unlock(&lod->lod_connects_lock);
1985 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
1989 spin_unlock(&lod->lod_connects_lock);
1991 /* the last user of lod has gone, let's release the device */
1995 rc = class_disconnect(exp); /* bz 9811 */
1997 if (rc == 0 && release)
1998 class_manual_cleanup(obd);
2002 LU_KEY_INIT(lod, struct lod_thread_info);
2004 static void lod_key_fini(const struct lu_context *ctx,
2005 struct lu_context_key *key, void *data)
2007 struct lod_thread_info *info = data;
2008 struct lod_layout_component *lds =
2009 info->lti_def_striping.lds_def_comp_entries;
2012 * allocated in lod_get_lov_ea
2013 * XXX: this is overload, a tread may have such store but used only
2014 * once. Probably better would be pool of such stores per LOD.
2016 if (info->lti_ea_store) {
2017 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
2018 info->lti_ea_store = NULL;
2019 info->lti_ea_store_size = 0;
2021 lu_buf_free(&info->lti_linkea_buf);
2024 lod_free_def_comp_entries(&info->lti_def_striping);
2026 if (info->lti_comp_size > 0)
2027 OBD_FREE(info->lti_comp_idx,
2028 info->lti_comp_size * sizeof(u32));
2030 lod_avoid_guide_fini(&info->lti_avoid);
2035 /* context key: lod_thread_key */
2036 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
2038 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
2040 static struct lu_device_type_operations lod_device_type_ops = {
2041 .ldto_init = lod_type_init,
2042 .ldto_fini = lod_type_fini,
2044 .ldto_start = lod_type_start,
2045 .ldto_stop = lod_type_stop,
2047 .ldto_device_alloc = lod_device_alloc,
2048 .ldto_device_free = lod_device_free,
2050 .ldto_device_fini = lod_device_fini
2053 static struct lu_device_type lod_device_type = {
2054 .ldt_tags = LU_DEVICE_DT,
2055 .ldt_name = LUSTRE_LOD_NAME,
2056 .ldt_ops = &lod_device_type_ops,
2057 .ldt_ctx_tags = LCT_MD_THREAD,
2061 * Implementation of obd_ops::o_get_info() for LOD
2063 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
2064 * the caller binary status whether LOD has seen connection to any OST target.
2065 * It will also check if the MDT update log context being initialized (if
2068 * \param[in] env LU environment provided by the caller
2069 * \param[in] exp export of the caller
2070 * \param[in] keylen len of the key
2071 * \param[in] key the key
2072 * \param[in] vallen not used
2073 * \param[in] val not used
2075 * \retval 0 if a connection was seen
2076 * \retval -EAGAIN if LOD isn't running yet or no
2077 * connection has been seen yet
2078 * \retval -EINVAL if not supported key is requested
2080 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
2081 u32 keylen, void *key, u32 *vallen, void *val)
2085 if (KEY_IS(KEY_OSP_CONNECTED)) {
2086 struct obd_device *obd = exp->exp_obd;
2087 struct lod_device *d;
2088 struct lod_tgt_desc *tgt;
2092 if (!obd->obd_set_up || obd->obd_stopping)
2095 d = lu2lod_dev(obd->obd_lu_dev);
2096 lod_getref(&d->lod_ost_descs);
2097 lod_foreach_ost(d, i) {
2098 tgt = OST_TGT(d, i);
2099 LASSERT(tgt && tgt->ltd_tgt);
2100 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2102 /* one healthy device is enough */
2106 lod_putref(d, &d->lod_ost_descs);
2108 lod_getref(&d->lod_mdt_descs);
2109 lod_foreach_mdt(d, i) {
2110 struct llog_ctxt *ctxt;
2112 tgt = MDT_TGT(d, i);
2113 LASSERT(tgt != NULL);
2114 LASSERT(tgt->ltd_tgt != NULL);
2115 if (!tgt->ltd_active)
2118 ctxt = llog_get_context(tgt->ltd_tgt->dd_lu_dev.ld_obd,
2119 LLOG_UPDATELOG_ORIG_CTXT);
2121 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2123 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2127 if (!ctxt->loc_handle) {
2128 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2130 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2132 llog_ctxt_put(ctxt);
2135 llog_ctxt_put(ctxt);
2137 lod_putref(d, &d->lod_mdt_descs);
2145 static int lod_obd_set_info_async(const struct lu_env *env,
2146 struct obd_export *exp,
2147 u32 keylen, void *key,
2148 u32 vallen, void *val,
2149 struct ptlrpc_request_set *set)
2151 struct obd_device *obd = class_exp2obd(exp);
2152 struct lod_device *d;
2153 struct lod_tgt_desc *tgt;
2161 set = ptlrpc_prep_set();
2166 d = lu2lod_dev(obd->obd_lu_dev);
2167 lod_getref(&d->lod_ost_descs);
2168 lod_foreach_ost(d, i) {
2169 tgt = OST_TGT(d, i);
2170 LASSERT(tgt && tgt->ltd_tgt);
2171 if (!tgt->ltd_active)
2174 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2176 if (rc2 != 0 && rc == 0)
2179 lod_putref(d, &d->lod_ost_descs);
2181 lod_getref(&d->lod_mdt_descs);
2182 lod_foreach_mdt(d, i) {
2183 tgt = MDT_TGT(d, i);
2184 LASSERT(tgt && tgt->ltd_tgt);
2185 if (!tgt->ltd_active)
2187 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2189 if (rc2 != 0 && rc == 0)
2192 lod_putref(d, &d->lod_mdt_descs);
2196 rc2 = ptlrpc_set_wait(env, set);
2197 if (rc2 == 0 && rc == 0)
2199 ptlrpc_set_destroy(set);
2204 static struct obd_ops lod_obd_device_ops = {
2205 .o_owner = THIS_MODULE,
2206 .o_connect = lod_obd_connect,
2207 .o_disconnect = lod_obd_disconnect,
2208 .o_get_info = lod_obd_get_info,
2209 .o_set_info_async = lod_obd_set_info_async,
2210 .o_pool_new = lod_pool_new,
2211 .o_pool_rem = lod_pool_remove,
2212 .o_pool_add = lod_pool_add,
2213 .o_pool_del = lod_pool_del,
2216 static struct obd_type sym;
2218 static int __init lod_init(void)
2220 struct dentry *symlink;
2221 struct obd_type *type;
2222 struct kobject *kobj;
2226 rc = lu_kmem_init(lod_caches);
2230 rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
2231 LUSTRE_LOD_NAME, &lod_device_type);
2233 lu_kmem_fini(lod_caches);
2237 /* create "lov" entry for compatibility purposes */
2239 dname.len = strlen(dname.name);
2240 dname.hash = ll_full_name_hash(debugfs_lustre_root, dname.name,
2242 symlink = d_lookup(debugfs_lustre_root, &dname);
2244 symlink = debugfs_create_dir(dname.name, debugfs_lustre_root);
2245 if (IS_ERR_OR_NULL(symlink)) {
2246 rc = symlink ? PTR_ERR(symlink) : -ENOMEM;
2249 sym.typ_debugfs_entry = symlink;
2254 kobj = kset_find_obj(lustre_kset, dname.name);
2260 kobj = class_setup_tunables(dname.name);
2263 if (sym.typ_debugfs_entry)
2264 ldebugfs_remove(&sym.typ_debugfs_entry);
2267 sym.typ_kobj = kobj;
2270 type = class_search_type(LUSTRE_LOV_NAME);
2271 if (type && type->typ_procroot)
2274 type = class_search_type(LUSTRE_LOD_NAME);
2275 type->typ_procsym = lprocfs_register("lov", proc_lustre_root,
2277 if (IS_ERR(type->typ_procsym)) {
2278 CERROR("lod: can't create compat entry \"lov\": %d\n",
2279 (int)PTR_ERR(type->typ_procsym));
2280 type->typ_procsym = NULL;
2286 static void __exit lod_exit(void)
2288 ldebugfs_remove(&sym.typ_debugfs_entry);
2289 kobject_put(sym.typ_kobj);
2290 class_unregister_type(LUSTRE_LOD_NAME);
2291 lu_kmem_fini(lod_caches);
2294 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2295 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
2296 MODULE_VERSION(LUSTRE_VERSION_STRING);
2297 MODULE_LICENSE("GPL");
2299 module_init(lod_init);
2300 module_exit(lod_exit);