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;
248 struct lu_tgt_desc *tgt;
252 ltd_foreach_tgt(ltd, tgt) {
255 LASSERT(tgt && tgt->ltd_tgt);
256 next = &tgt->ltd_tgt->dd_lu_dev;
257 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
259 CERROR("%s: error cleaning up LOD index %u: cmd %#x : rc = %d\n",
260 lod2obd(lod)->obd_name, tgt->ltd_index,
261 lcfg->lcfg_command, rc1);
265 lod_putref(lod, ltd);
269 struct lod_recovery_data {
270 struct lod_device *lrd_lod;
271 struct lod_tgt_desc *lrd_ltd;
272 struct ptlrpc_thread *lrd_thread;
278 * process update recovery record
280 * Add the update recovery recode to the update recovery list in
281 * lod_recovery_data. Then the recovery thread (target_recovery_thread)
282 * will redo these updates.
284 * \param[in]env execution environment
285 * \param[in]llh log handle of update record
286 * \param[in]rec update record to be replayed
287 * \param[in]data update recovery data which holds the necessary
288 * arguments for recovery (see struct lod_recovery_data)
290 * \retval 0 if the record is processed successfully.
291 * \retval negative errno if the record processing fails.
293 static int lod_process_recovery_updates(const struct lu_env *env,
294 struct llog_handle *llh,
295 struct llog_rec_hdr *rec,
298 struct lod_recovery_data *lrd = data;
299 struct llog_cookie *cookie = &lod_env_info(env)->lti_cookie;
300 struct lu_target *lut;
308 rc = lodname2mdt_index(lod2obd(lrd->lrd_lod)->obd_name, &index);
312 index = lrd->lrd_ltd->ltd_index;
316 llog_update_record_size((struct llog_update_record *)rec)) {
317 CERROR("%s: broken update record! index %u "DFID".%u: rc = %d\n",
318 lod2obd(lrd->lrd_lod)->obd_name, index,
319 PFID(&llh->lgh_id.lgl_oi.oi_fid), rec->lrh_index, -EIO);
323 cookie->lgc_lgl = llh->lgh_id;
324 cookie->lgc_index = rec->lrh_index;
325 cookie->lgc_subsys = LLOG_UPDATELOG_ORIG_CTXT;
327 CDEBUG(D_HA, "%s: process recovery updates "DFID".%u\n",
328 lod2obd(lrd->lrd_lod)->obd_name,
329 PFID(&llh->lgh_id.lgl_oi.oi_fid), rec->lrh_index);
330 lut = lod2lu_dev(lrd->lrd_lod)->ld_site->ls_tgt;
332 if (lut->lut_obd->obd_stopping ||
333 lut->lut_obd->obd_abort_recovery)
336 return insert_update_records_to_replay_list(lut->lut_tdtd,
337 (struct llog_update_record *)rec,
342 * recovery thread for update log
344 * Start recovery thread and prepare the sub llog, then it will retrieve
345 * the update records from the correpondent MDT and do recovery.
347 * \param[in] arg pointer to the recovery data
349 * \retval 0 if recovery succeeds
350 * \retval negative errno if recovery failed.
352 static int lod_sub_recovery_thread(void *arg)
354 struct lod_recovery_data *lrd = arg;
355 struct lod_device *lod = lrd->lrd_lod;
356 struct dt_device *dt;
357 struct ptlrpc_thread *thread = lrd->lrd_thread;
358 struct llog_ctxt *ctxt = NULL;
360 struct lu_target *lut;
361 struct lu_tgt_desc *mdt = NULL;
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);
384 dt = lrd->lrd_ltd->ltd_tgt;
386 start = ktime_get_real_seconds();
389 rc = lod_sub_prep_llog(&env, lod, dt, lrd->lrd_idx);
390 if (!rc && !lod->lod_child->dd_rdonly) {
391 /* Process the recovery record */
392 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
393 LLOG_UPDATELOG_ORIG_CTXT);
394 LASSERT(ctxt != NULL);
395 LASSERT(ctxt->loc_handle != NULL);
397 rc = llog_cat_process(&env, ctxt->loc_handle,
398 lod_process_recovery_updates, lrd, 0, 0);
402 struct lu_device *top_device;
404 top_device = lod->lod_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
406 * Because the remote target might failover at the same time,
409 if ((rc == -ETIMEDOUT || rc == -EAGAIN || rc == -EIO) &&
410 dt != lod->lod_child &&
411 !top_device->ld_obd->obd_abort_recovery &&
412 !top_device->ld_obd->obd_stopping) {
414 if (ctxt->loc_handle)
420 CDEBUG(D_HA, "%s get update log failed %d, retry\n",
421 dt->dd_lu_dev.ld_obd->obd_name, rc);
425 CERROR("%s get update log failed: rc = %d\n",
426 dt->dd_lu_dev.ld_obd->obd_name, rc);
429 spin_lock(&top_device->ld_obd->obd_dev_lock);
430 if (!top_device->ld_obd->obd_abort_recovery &&
431 !top_device->ld_obd->obd_stopping)
432 top_device->ld_obd->obd_abort_recovery = 1;
433 spin_unlock(&top_device->ld_obd->obd_dev_lock);
439 CDEBUG(D_HA, "%s retrieved update log, duration %lld, retries %d\n",
440 dt->dd_lu_dev.ld_obd->obd_name, ktime_get_real_seconds() - start,
443 spin_lock(&lod->lod_lock);
445 lod->lod_child_got_update_log = 1;
447 lrd->lrd_ltd->ltd_got_update_log = 1;
449 if (!lod->lod_child_got_update_log) {
450 spin_unlock(&lod->lod_lock);
454 lod_foreach_mdt(lod, mdt) {
455 if (!mdt->ltd_got_update_log) {
456 spin_unlock(&lod->lod_lock);
460 lut->lut_tdtd->tdtd_replay_ready = 1;
461 spin_unlock(&lod->lod_lock);
463 CDEBUG(D_HA, "%s got update logs from all MDTs.\n",
464 lut->lut_obd->obd_name);
465 wake_up(&lut->lut_obd->obd_next_transno_waitq);
470 thread->t_flags = SVC_STOPPED;
471 atomic_dec(&lut->lut_tdtd->tdtd_recovery_threads_count);
472 wake_up(&lut->lut_tdtd->tdtd_recovery_threads_waitq);
473 wake_up(&thread->t_ctl_waitq);
479 * finish sub llog context
481 * Stop update recovery thread for the sub device, then cleanup the
482 * correspondent llog ctxt.
484 * \param[in] env execution environment
485 * \param[in] lod lod device to do update recovery
486 * \param[in] thread recovery thread on this sub device
488 void lod_sub_fini_llog(const struct lu_env *env,
489 struct dt_device *dt, struct ptlrpc_thread *thread)
491 struct obd_device *obd;
492 struct llog_ctxt *ctxt;
496 obd = dt->dd_lu_dev.ld_obd;
497 CDEBUG(D_INFO, "%s: finish sub llog\n", obd->obd_name);
498 /* Stop recovery thread first */
499 if (thread && thread->t_flags & SVC_RUNNING) {
500 thread->t_flags = SVC_STOPPING;
501 wake_up(&thread->t_ctl_waitq);
502 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
505 ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
509 if (ctxt->loc_handle)
510 llog_cat_close(env, ctxt->loc_handle);
512 llog_cleanup(env, ctxt);
518 * Extract MDT target index from a device name.
520 * a helper function to extract index from the given device name
521 * like "fsname-MDTxxxx-mdtlov"
523 * \param[in] lodname device name
524 * \param[out] mdt_index extracted index
526 * \retval 0 on success
527 * \retval -EINVAL if the name is invalid
529 int lodname2mdt_index(char *lodname, u32 *mdt_index)
535 /* 1.8 configs don't have "-MDT0000" at the end */
536 ptr = strstr(lodname, "-MDT");
542 ptr = strrchr(lodname, '-');
545 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
549 if (strncmp(ptr, "-mdtlov", 7) != 0) {
551 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
555 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
557 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
561 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
563 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
567 index = simple_strtol(ptr - 4, &tmp, 16);
568 if (*tmp != '-' || index > INT_MAX) {
570 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
578 * Init sub llog context
580 * Setup update llog ctxt for update recovery threads, then start the
581 * recovery thread (lod_sub_recovery_thread) to read update llog from
582 * the correspondent MDT to do update recovery.
584 * \param[in] env execution environment
585 * \param[in] lod lod device to do update recovery
586 * \param[in] dt sub dt device for which the recovery thread is
588 * \retval 0 if initialization succeeds.
589 * \retval negative errno if initialization fails.
591 int lod_sub_init_llog(const struct lu_env *env, struct lod_device *lod,
592 struct dt_device *dt)
594 struct obd_device *obd;
595 struct lod_recovery_data *lrd = NULL;
596 struct ptlrpc_thread *thread;
597 struct task_struct *task;
598 struct lod_tgt_desc *subtgt = NULL;
605 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &master_index);
613 if (lod->lod_child == dt) {
614 thread = &lod->lod_child_recovery_thread;
615 index = master_index;
617 struct lu_tgt_desc *mdt;
619 lod_foreach_mdt(lod, mdt) {
620 if (mdt->ltd_tgt == dt) {
621 index = mdt->ltd_index;
626 LASSERT(subtgt != NULL);
627 OBD_ALLOC_PTR(subtgt->ltd_recovery_thread);
628 if (!subtgt->ltd_recovery_thread)
629 GOTO(free_lrd, rc = -ENOMEM);
631 thread = subtgt->ltd_recovery_thread;
634 CDEBUG(D_INFO, "%s init sub log %s\n", lod2obd(lod)->obd_name,
635 dt->dd_lu_dev.ld_obd->obd_name);
637 lrd->lrd_ltd = subtgt;
638 lrd->lrd_thread = thread;
639 lrd->lrd_idx = index;
640 init_waitqueue_head(&thread->t_ctl_waitq);
642 obd = dt->dd_lu_dev.ld_obd;
643 obd->obd_lvfs_ctxt.dt = dt;
644 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_UPDATELOG_ORIG_CTXT,
645 NULL, &llog_common_cat_ops);
647 CERROR("%s: cannot setup updatelog llog: rc = %d\n",
649 GOTO(free_thread, rc);
652 /* Start the recovery thread */
653 task = kthread_run(lod_sub_recovery_thread, lrd, "lod%04x_rec%04x",
654 master_index, index);
657 CERROR("%s: cannot start recovery thread: rc = %d\n",
662 wait_event_idle(thread->t_ctl_waitq, thread->t_flags & SVC_RUNNING ||
663 thread->t_flags & SVC_STOPPED);
667 lod_sub_fini_llog(env, dt, thread);
669 if (lod->lod_child != dt) {
670 OBD_FREE_PTR(subtgt->ltd_recovery_thread);
671 subtgt->ltd_recovery_thread = NULL;
679 * Stop sub recovery thread
681 * Stop sub recovery thread on all subs.
683 * \param[in] env execution environment
684 * \param[in] lod lod device to do update recovery
686 static void lod_sub_stop_recovery_threads(const struct lu_env *env,
687 struct lod_device *lod)
689 struct ptlrpc_thread *thread;
690 struct lu_tgt_desc *mdt;
693 * Stop the update log commit cancel threads and finish master
696 thread = &lod->lod_child_recovery_thread;
697 /* Stop recovery thread first */
698 if (thread && thread->t_flags & SVC_RUNNING) {
699 thread->t_flags = SVC_STOPPING;
700 wake_up(&thread->t_ctl_waitq);
701 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
704 lod_getref(&lod->lod_mdt_descs);
705 lod_foreach_mdt(lod, mdt) {
706 thread = mdt->ltd_recovery_thread;
707 if (thread && thread->t_flags & SVC_RUNNING) {
708 thread->t_flags = SVC_STOPPING;
709 wake_up(&thread->t_ctl_waitq);
710 wait_event(thread->t_ctl_waitq,
711 thread->t_flags & SVC_STOPPED);
712 OBD_FREE_PTR(mdt->ltd_recovery_thread);
713 mdt->ltd_recovery_thread = NULL;
716 lod_putref(lod, &lod->lod_mdt_descs);
720 * finish all sub llog
722 * cleanup all of sub llog ctxt on the LOD.
724 * \param[in] env execution environment
725 * \param[in] lod lod device to do update recovery
727 static void lod_sub_fini_all_llogs(const struct lu_env *env,
728 struct lod_device *lod)
730 struct lu_tgt_desc *mdt;
733 * Stop the update log commit cancel threads and finish master
736 lod_sub_fini_llog(env, lod->lod_child,
737 &lod->lod_child_recovery_thread);
738 lod_getref(&lod->lod_mdt_descs);
739 lod_foreach_mdt(lod, mdt)
740 lod_sub_fini_llog(env, mdt->ltd_tgt,
741 mdt->ltd_recovery_thread);
742 lod_putref(lod, &lod->lod_mdt_descs);
745 static char *lod_show_update_logs_retrievers(void *data, int *size, int *count)
747 struct lod_device *lod = (struct lod_device *)data;
748 struct lu_target *lut = lod2lu_dev(lod)->ld_site->ls_tgt;
749 struct lu_tgt_desc *mdt = NULL;
755 *count = atomic_read(&lut->lut_tdtd->tdtd_recovery_threads_count);
761 *size = 5 * *count + 1;
762 OBD_ALLOC(buf, *size);
767 memset(buf, 0, *size);
769 if (!lod->lod_child_got_update_log) {
770 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &i);
771 LASSERTF(rc == 0, "Fail to parse target index: rc = %d\n", rc);
773 rc = snprintf(buf + len, *size - len, " %04x", i);
780 lod_foreach_mdt(lod, mdt) {
781 if (!mdt->ltd_got_update_log) {
782 rc = snprintf(buf + len, *size - len, " %04x",
784 if (unlikely(rc <= 0))
796 * Prepare distribute txn
798 * Prepare distribute txn structure for LOD
800 * \param[in] env execution environment
801 * \param[in] lod_device LOD device
803 * \retval 0 if preparation succeeds.
804 * \retval negative errno if preparation fails.
806 static int lod_prepare_distribute_txn(const struct lu_env *env,
807 struct lod_device *lod)
809 struct target_distribute_txn_data *tdtd;
810 struct lu_target *lut;
815 /* Init update recovery data */
820 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
821 tdtd->tdtd_dt = &lod->lod_dt_dev;
822 rc = distribute_txn_init(env, lut, tdtd,
823 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id);
826 CERROR("%s: cannot init distribute txn: rc = %d\n",
827 lod2obd(lod)->obd_name, rc);
832 tdtd->tdtd_show_update_logs_retrievers =
833 lod_show_update_logs_retrievers;
834 tdtd->tdtd_show_retrievers_cbdata = lod;
836 lut->lut_tdtd = tdtd;
842 * Finish distribute txn
844 * Release the resource holding by distribute txn, i.e. stop distribute
847 * \param[in] env execution environment
848 * \param[in] lod lod device
850 static void lod_fini_distribute_txn(const struct lu_env *env,
851 struct lod_device *lod)
853 struct lu_target *lut;
855 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
856 target_recovery_fini(lut->lut_obd);
860 distribute_txn_fini(env, lut->lut_tdtd);
862 OBD_FREE_PTR(lut->lut_tdtd);
863 lut->lut_tdtd = NULL;
867 * Implementation of lu_device_operations::ldo_process_config() for LOD
869 * The method is called by the configuration subsystem during setup,
870 * cleanup and when the configuration changes. The method processes
871 * few specific commands like adding/removing the targets, changing
872 * the runtime parameters.
874 * \param[in] env LU environment provided by the caller
875 * \param[in] dev lod device
876 * \param[in] lcfg configuration command to apply
878 * \retval 0 on success
879 * \retval negative negated errno on error
881 * The examples are below.
883 * Add osc config log:
884 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
885 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
886 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
887 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
888 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
889 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
891 * Add mdc config log:
892 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
893 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
894 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
895 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
896 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
897 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
899 static int lod_process_config(const struct lu_env *env,
900 struct lu_device *dev,
901 struct lustre_cfg *lcfg)
903 struct lod_device *lod = lu2lod_dev(dev);
904 struct lu_device *next = &lod->lod_child->dd_lu_dev;
910 switch (lcfg->lcfg_command) {
911 case LCFG_LOV_DEL_OBD:
912 case LCFG_LOV_ADD_INA:
913 case LCFG_LOV_ADD_OBD:
919 * lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
920 * modify_mdc_tgts add 0:lustre-MDT0001
921 * 1:lustre-MDT0001-mdc0002
924 arg1 = lustre_cfg_string(lcfg, 1);
926 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
927 GOTO(out, rc = -EINVAL);
928 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
929 GOTO(out, rc = -EINVAL);
931 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
934 rc = lodname2mdt_index(lustre_cfg_string(lcfg, 0),
939 rc = lod_add_device(env, lod, arg1, index, gen,
940 mdt_index, LUSTRE_OSC_NAME, 1);
941 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
943 rc = lod_add_device(env, lod, arg1, index, gen,
944 mdt_index, LUSTRE_MDC_NAME, 1);
945 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
946 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
948 rc = lod_add_device(env, lod, arg1, index, gen,
949 mdt_index, LUSTRE_OSC_NAME, 0);
951 rc = lod_del_device(env, lod, &lod->lod_ost_descs,
959 struct obd_device *obd;
964 * Check if it is activate/deactivate mdc
965 * lustre-MDTXXXX-osp-MDTXXXX.active=1
967 param = lustre_cfg_buf(lcfg, 1);
968 if (strstr(param, "osp") && strstr(param, ".active=")) {
969 struct lod_tgt_desc *sub_tgt = NULL;
970 struct lu_tgt_desc *mdt;
974 ptr = strstr(param, ".");
976 obd = class_name2obd(param);
978 CERROR("%s: can not find %s: rc = %d\n",
979 lod2obd(lod)->obd_name, param, -EINVAL);
984 lod_foreach_mdt(lod, mdt) {
985 if (mdt->ltd_tgt->dd_lu_dev.ld_obd == obd) {
992 CERROR("%s: can not find %s: rc = %d\n",
993 lod2obd(lod)->obd_name, param, -EINVAL);
999 tmp = strstr(param, "=");
1002 struct llog_ctxt *ctxt;
1004 obd = sub_tgt->ltd_tgt->dd_lu_dev.ld_obd;
1005 ctxt = llog_get_context(obd,
1006 LLOG_UPDATELOG_ORIG_CTXT);
1008 rc = llog_setup(env, obd, &obd->obd_olg,
1009 LLOG_UPDATELOG_ORIG_CTXT,
1010 NULL, &llog_common_cat_ops);
1014 llog_ctxt_put(ctxt);
1016 rc = lod_sub_prep_llog(env, lod,
1018 sub_tgt->ltd_index);
1020 sub_tgt->ltd_active = 1;
1022 lod_sub_fini_llog(env, sub_tgt->ltd_tgt,
1024 sub_tgt->ltd_active = 0;
1030 if (strstr(param, PARAM_LOD) != NULL)
1031 count = class_modify_config(lcfg, PARAM_LOD,
1032 &lod->lod_dt_dev.dd_kobj);
1034 count = class_modify_config(lcfg, PARAM_LOV,
1035 &lod->lod_dt_dev.dd_kobj);
1036 rc = count > 0 ? 0 : count;
1039 case LCFG_PRE_CLEANUP: {
1040 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1041 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1042 next = &lod->lod_child->dd_lu_dev;
1043 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1045 CDEBUG(D_HA, "%s: can't process %u: %d\n",
1046 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1048 lod_sub_stop_recovery_threads(env, lod);
1049 lod_fini_distribute_txn(env, lod);
1050 lod_sub_fini_all_llogs(env, lod);
1053 case LCFG_CLEANUP: {
1054 if (lod->lod_md_root) {
1055 dt_object_put(env, &lod->lod_md_root->ldo_obj);
1056 lod->lod_md_root = NULL;
1060 * do cleanup on underlying storage only when
1061 * all OSPs are cleaned up, as they use that OSD as well
1063 lu_dev_del_linkage(dev->ld_site, dev);
1064 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1065 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1066 next = &lod->lod_child->dd_lu_dev;
1067 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1069 CERROR("%s: can't process %u: rc = %d\n",
1070 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1072 rc = obd_disconnect(lod->lod_child_exp);
1074 CERROR("error in disconnect from storage: rc = %d\n",
1079 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
1080 lcfg->lcfg_command);
1090 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
1092 * The method is called once the recovery is complete. This implementation
1093 * distributes the notification to all the known targets.
1095 * see include/lu_object.h for the details
1097 static int lod_recovery_complete(const struct lu_env *env,
1098 struct lu_device *dev)
1100 struct lod_device *lod = lu2lod_dev(dev);
1101 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1102 struct lod_tgt_desc *tgt;
1107 LASSERT(lod->lod_recovery_completed == 0);
1108 lod->lod_recovery_completed = 1;
1110 rc = next->ld_ops->ldo_recovery_complete(env, next);
1112 lod_getref(&lod->lod_ost_descs);
1113 if (lod->lod_ost_descs.ltd_tgts_size > 0) {
1114 lod_foreach_ost(lod, tgt) {
1115 LASSERT(tgt && tgt->ltd_tgt);
1116 next = &tgt->ltd_tgt->dd_lu_dev;
1117 rc = next->ld_ops->ldo_recovery_complete(env, next);
1119 CERROR("%s: can't complete recovery on #%d: rc = %d\n",
1120 lod2obd(lod)->obd_name, tgt->ltd_index,
1124 lod_putref(lod, &lod->lod_ost_descs);
1129 * Init update logs on all sub device
1131 * LOD initialize update logs on all of sub devices. Because the initialization
1132 * process might need FLD lookup, see llog_osd_open()->dt_locate()->...->
1133 * lod_object_init(), this API has to be called after LOD is initialized.
1134 * \param[in] env execution environment
1135 * \param[in] lod lod device
1137 * \retval 0 if update log is initialized successfully.
1138 * \retval negative errno if initialization fails.
1140 static int lod_sub_init_llogs(const struct lu_env *env, struct lod_device *lod)
1142 struct lu_tgt_desc *mdt;
1148 * llog must be setup after LOD is initialized, because llog
1149 * initialization include FLD lookup
1151 LASSERT(lod->lod_initialized);
1153 /* Init the llog in its own stack */
1154 rc = lod_sub_init_llog(env, lod, lod->lod_child);
1158 lod_foreach_mdt(lod, mdt) {
1159 rc = lod_sub_init_llog(env, lod, mdt->ltd_tgt);
1168 * Implementation of lu_device_operations::ldo_prepare() for LOD
1170 * see include/lu_object.h for the details.
1172 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
1173 struct lu_device *cdev)
1175 struct lod_device *lod = lu2lod_dev(cdev);
1176 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1177 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1179 struct dt_object *root;
1180 struct dt_object *dto;
1185 rc = next->ld_ops->ldo_prepare(env, pdev, next);
1187 CERROR("%s: prepare bottom error: rc = %d\n",
1188 lod2obd(lod)->obd_name, rc);
1192 lod->lod_initialized = 1;
1194 rc = dt_root_get(env, lod->lod_child, fid);
1198 root = dt_locate(env, lod->lod_child, fid);
1200 RETURN(PTR_ERR(root));
1202 /* Create update log object */
1203 index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1204 lu_update_log_fid(fid, index);
1206 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1208 lod_update_log_name,
1211 GOTO(out_put, rc = PTR_ERR(dto));
1213 dt_object_put(env, dto);
1215 /* Create update log dir */
1216 lu_update_log_dir_fid(fid, index);
1217 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1219 lod_update_log_dir_name,
1222 GOTO(out_put, rc = PTR_ERR(dto));
1224 dt_object_put(env, dto);
1226 rc = lod_prepare_distribute_txn(env, lod);
1230 rc = lod_sub_init_llogs(env, lod);
1235 dt_object_put(env, root);
1240 const struct lu_device_operations lod_lu_ops = {
1241 .ldo_object_alloc = lod_object_alloc,
1242 .ldo_process_config = lod_process_config,
1243 .ldo_recovery_complete = lod_recovery_complete,
1244 .ldo_prepare = lod_prepare,
1248 * Implementation of dt_device_operations::dt_root_get() for LOD
1250 * see include/dt_object.h for the details.
1252 static int lod_root_get(const struct lu_env *env,
1253 struct dt_device *dev, struct lu_fid *f)
1255 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
1258 static void lod_statfs_sum(struct obd_statfs *sfs,
1259 struct obd_statfs *ost_sfs, int *bs)
1261 while (ost_sfs->os_bsize < *bs) {
1263 sfs->os_bsize >>= 1;
1264 sfs->os_bavail <<= 1;
1265 sfs->os_blocks <<= 1;
1266 sfs->os_bfree <<= 1;
1267 sfs->os_granted <<= 1;
1269 while (ost_sfs->os_bsize > *bs) {
1270 ost_sfs->os_bsize >>= 1;
1271 ost_sfs->os_bavail <<= 1;
1272 ost_sfs->os_blocks <<= 1;
1273 ost_sfs->os_bfree <<= 1;
1274 ost_sfs->os_granted <<= 1;
1276 sfs->os_bavail += ost_sfs->os_bavail;
1277 sfs->os_blocks += ost_sfs->os_blocks;
1278 sfs->os_bfree += ost_sfs->os_bfree;
1279 sfs->os_granted += ost_sfs->os_granted;
1283 * Implementation of dt_device_operations::dt_statfs() for LOD
1285 * see include/dt_object.h for the details.
1287 static int lod_statfs(const struct lu_env *env, struct dt_device *dev,
1288 struct obd_statfs *sfs, struct obd_statfs_info *info)
1290 struct lod_device *lod = dt2lod_dev(dev);
1291 struct lu_tgt_desc *tgt;
1292 struct obd_statfs ost_sfs;
1297 rc = dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
1306 sfs->os_granted = 0;
1308 lod_getref(&lod->lod_mdt_descs);
1309 lod_foreach_mdt(lod, tgt) {
1310 rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
1314 sfs->os_files += ost_sfs.os_files;
1315 sfs->os_ffree += ost_sfs.os_ffree;
1316 lod_statfs_sum(sfs, &ost_sfs, &bs);
1318 lod_putref(lod, &lod->lod_mdt_descs);
1321 * at some point we can check whether DoM is enabled and
1322 * decide how to account MDT space. for simplicity let's
1323 * just fallback to pre-DoM policy if any OST is alive
1325 lod_getref(&lod->lod_ost_descs);
1326 lod_foreach_ost(lod, tgt) {
1327 rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
1329 if (rc || ost_sfs.os_bsize == 0)
1333 * if only MDTs with DoM then report only MDT blocks,
1334 * otherwise show only OST blocks, and DoM is "free"
1339 sfs->os_granted = 0;
1341 ost_files += ost_sfs.os_files;
1342 ost_ffree += ost_sfs.os_ffree;
1343 ost_sfs.os_bavail += ost_sfs.os_granted;
1344 lod_statfs_sum(sfs, &ost_sfs, &bs);
1345 LASSERTF(bs == ost_sfs.os_bsize, "%d != %d\n",
1346 (int)sfs->os_bsize, (int)ost_sfs.os_bsize);
1348 lod_putref(lod, &lod->lod_ost_descs);
1349 sfs->os_state |= OS_STATE_SUM;
1351 /* If we have _some_ OSTs, but don't have as many free objects on the
1352 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1353 * to compensate, so that the "inodes in use" number is correct.
1354 * This should be kept in sync with ll_statfs_internal().
1356 if (ost_files && ost_ffree < sfs->os_ffree) {
1357 sfs->os_files = (sfs->os_files - sfs->os_ffree) + ost_ffree;
1358 sfs->os_ffree = ost_ffree;
1361 /* a single successful statfs should be enough */
1369 * Implementation of dt_device_operations::dt_trans_create() for LOD
1371 * Creates a transaction using local (to this node) OSD.
1373 * see include/dt_object.h for the details.
1375 static struct thandle *lod_trans_create(const struct lu_env *env,
1376 struct dt_device *dt)
1380 th = top_trans_create(env, dt2lod_dev(dt)->lod_child);
1390 * Implementation of dt_device_operations::dt_trans_start() for LOD
1392 * Starts the set of local transactions using the targets involved
1393 * in declare phase. Initial support for the distributed transactions.
1395 * see include/dt_object.h for the details.
1397 static int lod_trans_start(const struct lu_env *env, struct dt_device *dt,
1400 return top_trans_start(env, dt2lod_dev(dt)->lod_child, th);
1403 static int lod_trans_cb_add(struct thandle *th,
1404 struct dt_txn_commit_cb *dcb)
1406 struct top_thandle *top_th = container_of(th, struct top_thandle,
1408 return dt_trans_cb_add(top_th->tt_master_sub_thandle, dcb);
1412 * add noop update to the update records
1414 * Add noop updates to the update records, which is only used in
1417 * \param[in] env execution environment
1418 * \param[in] dt dt device of lod
1419 * \param[in] th thandle
1420 * \param[in] count the count of update records to be added.
1422 * \retval 0 if adding succeeds.
1423 * \retval negative errno if adding fails.
1425 static int lod_add_noop_records(const struct lu_env *env,
1426 struct dt_device *dt, struct thandle *th,
1429 struct top_thandle *top_th;
1430 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1434 top_th = container_of(th, struct top_thandle, tt_super);
1435 if (!top_th->tt_multiple_thandle)
1439 for (i = 0; i < count; i++) {
1440 rc = update_record_pack(noop, th, fid);
1448 * Implementation of dt_device_operations::dt_trans_stop() for LOD
1450 * Stops the set of local transactions using the targets involved
1451 * in declare phase. Initial support for the distributed transactions.
1453 * see include/dt_object.h for the details.
1455 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
1458 if (OBD_FAIL_CHECK(OBD_FAIL_SPLIT_UPDATE_REC)) {
1461 rc = lod_add_noop_records(env, dt, th, 5000);
1465 return top_trans_stop(env, dt2lod_dev(dt)->lod_child, th);
1469 * Implementation of dt_device_operations::dt_conf_get() for LOD
1471 * Currently returns the configuration provided by the local OSD.
1473 * see include/dt_object.h for the details.
1475 static void lod_conf_get(const struct lu_env *env,
1476 const struct dt_device *dev,
1477 struct dt_device_param *param)
1479 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
1483 * Implementation of dt_device_operations::dt_sync() for LOD
1485 * Syncs all known OST targets. Very very expensive and used
1486 * rarely by LFSCK now. Should not be used in general.
1488 * see include/dt_object.h for the details.
1490 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
1492 struct lod_device *lod = dt2lod_dev(dev);
1493 struct lu_tgt_desc *tgt;
1498 lod_getref(&lod->lod_ost_descs);
1499 lod_foreach_ost(lod, tgt) {
1500 if (!tgt->ltd_active)
1502 rc = dt_sync(env, tgt->ltd_tgt);
1504 if (rc != -ENOTCONN) {
1505 CERROR("%s: can't sync ost %u: rc = %d\n",
1506 lod2obd(lod)->obd_name, tgt->ltd_index,
1513 lod_putref(lod, &lod->lod_ost_descs);
1518 lod_getref(&lod->lod_mdt_descs);
1519 lod_foreach_mdt(lod, tgt) {
1520 if (!tgt->ltd_active)
1522 rc = dt_sync(env, tgt->ltd_tgt);
1524 if (rc != -ENOTCONN) {
1525 CERROR("%s: can't sync mdt %u: rc = %d\n",
1526 lod2obd(lod)->obd_name, tgt->ltd_index,
1533 lod_putref(lod, &lod->lod_mdt_descs);
1536 rc = dt_sync(env, lod->lod_child);
1542 * Implementation of dt_device_operations::dt_ro() for LOD
1544 * Turns local OSD read-only, used for the testing only.
1546 * see include/dt_object.h for the details.
1548 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
1550 return dt_ro(env, dt2lod_dev(dev)->lod_child);
1554 * Implementation of dt_device_operations::dt_commit_async() for LOD
1556 * Asks local OSD to commit sooner.
1558 * see include/dt_object.h for the details.
1560 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
1562 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
1565 static const struct dt_device_operations lod_dt_ops = {
1566 .dt_root_get = lod_root_get,
1567 .dt_statfs = lod_statfs,
1568 .dt_trans_create = lod_trans_create,
1569 .dt_trans_start = lod_trans_start,
1570 .dt_trans_stop = lod_trans_stop,
1571 .dt_conf_get = lod_conf_get,
1572 .dt_sync = lod_sync,
1574 .dt_commit_async = lod_commit_async,
1575 .dt_trans_cb_add = lod_trans_cb_add,
1579 * Connect to a local OSD.
1581 * Used to connect to the local OSD at mount. OSD name is taken from the
1582 * configuration command passed. This connection is used to identify LU
1583 * site and pin the OSD from early removal.
1585 * \param[in] env LU environment provided by the caller
1586 * \param[in] lod lod device
1587 * \param[in] cfg configuration command to apply
1589 * \retval 0 on success
1590 * \retval negative negated errno on error
1592 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
1593 struct lustre_cfg *cfg)
1595 struct obd_connect_data *data = NULL;
1596 struct obd_device *obd;
1597 char *nextdev = NULL, *p, *s;
1603 LASSERT(lod->lod_child_exp == NULL);
1606 * compatibility hack: we still use old config logs
1607 * which specify LOV, but we need to learn underlying
1608 * OSD device, which is supposed to be:
1609 * <fsname>-MDTxxxx-osd
1611 * 2.x MGS generates lines like the following:
1612 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
1613 * 1.8 MGS generates lines like the following:
1614 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
1616 * we use "-MDT" to differentiate 2.x from 1.8
1618 p = lustre_cfg_string(cfg, 0);
1619 if (p && strstr(p, "-mdtlov")) {
1620 len = strlen(p) + 6;
1621 OBD_ALLOC(nextdev, len);
1623 GOTO(out, rc = -ENOMEM);
1626 s = strstr(nextdev, "-mdtlov");
1628 CERROR("%s: unable to parse device name: rc = %d\n",
1629 lustre_cfg_string(cfg, 0), -EINVAL);
1630 GOTO(out, rc = -EINVAL);
1633 if (strstr(nextdev, "-MDT")) {
1638 strcpy(s, "-MDT0000-osd");
1641 CERROR("%s: unable to parse device name: rc = %d\n",
1642 lustre_cfg_string(cfg, 0), -EINVAL);
1643 GOTO(out, rc = -EINVAL);
1646 OBD_ALLOC_PTR(data);
1648 GOTO(out, rc = -ENOMEM);
1650 obd = class_name2obd(nextdev);
1652 CERROR("%s: can not locate next device: rc = %d\n",
1653 nextdev, -ENOTCONN);
1654 GOTO(out, rc = -ENOTCONN);
1657 data->ocd_connect_flags = OBD_CONNECT_VERSION;
1658 data->ocd_version = LUSTRE_VERSION_CODE;
1660 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
1663 CERROR("%s: cannot connect to next dev: rc = %d\n",
1668 lod->lod_dt_dev.dd_lu_dev.ld_site =
1669 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
1670 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
1671 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
1677 OBD_FREE(nextdev, len);
1682 * Initialize LOD device at setup.
1684 * Initializes the given LOD device using the original configuration command.
1685 * The function initiates a connection to the local OSD and initializes few
1686 * internal structures like pools, target tables, etc.
1688 * \param[in] env LU environment provided by the caller
1689 * \param[in] lod lod device
1690 * \param[in] ldt not used
1691 * \param[in] cfg configuration command
1693 * \retval 0 on success
1694 * \retval negative negated errno on error
1696 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
1697 struct lu_device_type *ldt, struct lustre_cfg *cfg)
1699 struct dt_device_param ddp;
1700 struct obd_device *obd;
1705 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1708 CERROR("Cannot find obd with name '%s': rc = %d\n",
1709 lustre_cfg_string(cfg, 0), rc);
1713 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1714 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
1715 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
1716 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
1718 rc = lod_connect_to_osd(env, lod, cfg);
1722 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
1723 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
1724 lod->lod_dom_max_stripesize = (1ULL << 20); /* 1Mb as default value */
1726 /* setup obd to be used with old lov code */
1727 rc = lod_pools_init(lod, cfg);
1729 GOTO(out_disconnect, rc);
1731 rc = lod_procfs_init(lod);
1733 GOTO(out_pools, rc);
1735 spin_lock_init(&lod->lod_lock);
1736 spin_lock_init(&lod->lod_connects_lock);
1737 lu_tgt_descs_init(&lod->lod_mdt_descs, true);
1738 lu_tgt_descs_init(&lod->lod_ost_descs, false);
1743 lod_pools_fini(lod);
1745 obd_disconnect(lod->lod_child_exp);
1750 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
1752 * Releases the memory allocated for LOD device.
1754 * see include/lu_object.h for the details.
1756 static struct lu_device *lod_device_free(const struct lu_env *env,
1757 struct lu_device *lu)
1759 struct lod_device *lod = lu2lod_dev(lu);
1760 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1764 if (atomic_read(&lu->ld_ref) > 0 &&
1765 !cfs_hash_is_empty(lu->ld_site->ls_obj_hash)) {
1766 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1767 lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
1769 LASSERTF(atomic_read(&lu->ld_ref) == 0, "lu is %p\n", lu);
1770 dt_device_fini(&lod->lod_dt_dev);
1776 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
1778 * Allocates LOD device and calls the helpers to initialize it.
1780 * see include/lu_object.h for the details.
1782 static struct lu_device *lod_device_alloc(const struct lu_env *env,
1783 struct lu_device_type *type,
1784 struct lustre_cfg *lcfg)
1786 struct lod_device *lod;
1787 struct lu_device *lu_dev;
1791 lu_dev = ERR_PTR(-ENOMEM);
1795 lu_dev = lod2lu_dev(lod);
1796 dt_device_init(&lod->lod_dt_dev, type);
1797 rc = lod_init0(env, lod, type, lcfg);
1799 lod_device_free(env, lu_dev);
1800 lu_dev = ERR_PTR(rc);
1807 static void lod_avoid_guide_fini(struct lod_avoid_guide *lag)
1809 if (lag->lag_oss_avoid_array)
1810 OBD_FREE(lag->lag_oss_avoid_array,
1811 sizeof(u32) * lag->lag_oaa_size);
1812 if (lag->lag_ost_avoid_bitmap)
1813 CFS_FREE_BITMAP(lag->lag_ost_avoid_bitmap);
1817 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
1819 * Releases the internal resources used by LOD device.
1821 * see include/lu_object.h for the details.
1823 static struct lu_device *lod_device_fini(const struct lu_env *env,
1824 struct lu_device *d)
1826 struct lod_device *lod = lu2lod_dev(d);
1831 lod_pools_fini(lod);
1833 lod_procfs_fini(lod);
1835 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs);
1837 CERROR("%s: can not fini ost descriptors: rc = %d\n",
1838 lod2obd(lod)->obd_name, rc);
1840 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs);
1842 CERROR("%s: can not fini mdt descriptors: rc = %d\n",
1843 lod2obd(lod)->obd_name, rc);
1849 * Implementation of obd_ops::o_connect() for LOD
1851 * Used to track all the users of this specific LOD device,
1852 * so the device stays up until the last user disconnected.
1854 * \param[in] env LU environment provided by the caller
1855 * \param[out] exp export the caller will be using to access LOD
1856 * \param[in] obd OBD device representing LOD device
1857 * \param[in] cluuid unique identifier of the caller
1858 * \param[in] data not used
1859 * \param[in] localdata not used
1861 * \retval 0 on success
1862 * \retval negative negated errno on error
1864 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
1865 struct obd_device *obd, struct obd_uuid *cluuid,
1866 struct obd_connect_data *data, void *localdata)
1868 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1869 struct lustre_handle conn;
1874 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1876 rc = class_connect(&conn, obd, cluuid);
1880 *exp = class_conn2export(&conn);
1882 spin_lock(&lod->lod_connects_lock);
1883 lod->lod_connects++;
1884 /* at the moment we expect the only user */
1885 LASSERT(lod->lod_connects == 1);
1886 spin_unlock(&lod->lod_connects_lock);
1893 * Implementation of obd_ops::o_disconnect() for LOD
1895 * When the caller doesn't need to use this LOD instance, it calls
1896 * obd_disconnect() and LOD releases corresponding export/reference count.
1897 * Once all the users gone, LOD device is released.
1899 * \param[in] exp export provided to the caller in obd_connect()
1901 * \retval 0 on success
1902 * \retval negative negated errno on error
1904 static int lod_obd_disconnect(struct obd_export *exp)
1906 struct obd_device *obd = exp->exp_obd;
1907 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1908 int rc, release = 0;
1912 /* Only disconnect the underlying layers on the final disconnect. */
1913 spin_lock(&lod->lod_connects_lock);
1914 lod->lod_connects--;
1915 if (lod->lod_connects != 0) {
1916 /* why should there be more than 1 connect? */
1917 spin_unlock(&lod->lod_connects_lock);
1918 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
1922 spin_unlock(&lod->lod_connects_lock);
1924 /* the last user of lod has gone, let's release the device */
1928 rc = class_disconnect(exp); /* bz 9811 */
1930 if (rc == 0 && release)
1931 class_manual_cleanup(obd);
1935 LU_KEY_INIT(lod, struct lod_thread_info);
1937 static void lod_key_fini(const struct lu_context *ctx,
1938 struct lu_context_key *key, void *data)
1940 struct lod_thread_info *info = data;
1941 struct lod_layout_component *lds =
1942 info->lti_def_striping.lds_def_comp_entries;
1945 * allocated in lod_get_lov_ea
1946 * XXX: this is overload, a tread may have such store but used only
1947 * once. Probably better would be pool of such stores per LOD.
1949 if (info->lti_ea_store) {
1950 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
1951 info->lti_ea_store = NULL;
1952 info->lti_ea_store_size = 0;
1954 lu_buf_free(&info->lti_linkea_buf);
1957 lod_free_def_comp_entries(&info->lti_def_striping);
1959 if (info->lti_comp_size > 0)
1960 OBD_FREE(info->lti_comp_idx,
1961 info->lti_comp_size * sizeof(u32));
1963 lod_avoid_guide_fini(&info->lti_avoid);
1968 /* context key: lod_thread_key */
1969 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
1971 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
1973 static struct lu_device_type_operations lod_device_type_ops = {
1974 .ldto_init = lod_type_init,
1975 .ldto_fini = lod_type_fini,
1977 .ldto_start = lod_type_start,
1978 .ldto_stop = lod_type_stop,
1980 .ldto_device_alloc = lod_device_alloc,
1981 .ldto_device_free = lod_device_free,
1983 .ldto_device_fini = lod_device_fini
1986 static struct lu_device_type lod_device_type = {
1987 .ldt_tags = LU_DEVICE_DT,
1988 .ldt_name = LUSTRE_LOD_NAME,
1989 .ldt_ops = &lod_device_type_ops,
1990 .ldt_ctx_tags = LCT_MD_THREAD,
1994 * Implementation of obd_ops::o_get_info() for LOD
1996 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
1997 * the caller binary status whether LOD has seen connection to any OST target.
1998 * It will also check if the MDT update log context being initialized (if
2001 * \param[in] env LU environment provided by the caller
2002 * \param[in] exp export of the caller
2003 * \param[in] keylen len of the key
2004 * \param[in] key the key
2005 * \param[in] vallen not used
2006 * \param[in] val not used
2008 * \retval 0 if a connection was seen
2009 * \retval -EAGAIN if LOD isn't running yet or no
2010 * connection has been seen yet
2011 * \retval -EINVAL if not supported key is requested
2013 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
2014 u32 keylen, void *key, u32 *vallen, void *val)
2018 if (KEY_IS(KEY_OSP_CONNECTED)) {
2019 struct obd_device *obd = exp->exp_obd;
2020 struct lod_device *d;
2021 struct lod_tgt_desc *tgt;
2024 if (!obd->obd_set_up || obd->obd_stopping)
2027 d = lu2lod_dev(obd->obd_lu_dev);
2028 lod_getref(&d->lod_ost_descs);
2029 lod_foreach_ost(d, tgt) {
2030 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2032 /* one healthy device is enough */
2036 lod_putref(d, &d->lod_ost_descs);
2038 lod_getref(&d->lod_mdt_descs);
2039 lod_foreach_mdt(d, tgt) {
2040 struct llog_ctxt *ctxt;
2042 if (!tgt->ltd_active)
2045 ctxt = llog_get_context(tgt->ltd_tgt->dd_lu_dev.ld_obd,
2046 LLOG_UPDATELOG_ORIG_CTXT);
2048 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2050 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2054 if (!ctxt->loc_handle) {
2055 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2057 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2059 llog_ctxt_put(ctxt);
2062 llog_ctxt_put(ctxt);
2064 lod_putref(d, &d->lod_mdt_descs);
2072 static int lod_obd_set_info_async(const struct lu_env *env,
2073 struct obd_export *exp,
2074 u32 keylen, void *key,
2075 u32 vallen, void *val,
2076 struct ptlrpc_request_set *set)
2078 struct obd_device *obd = class_exp2obd(exp);
2079 struct lod_device *d;
2080 struct lod_tgt_desc *tgt;
2088 set = ptlrpc_prep_set();
2093 d = lu2lod_dev(obd->obd_lu_dev);
2094 lod_getref(&d->lod_ost_descs);
2095 lod_foreach_ost(d, tgt) {
2096 if (!tgt->ltd_active)
2099 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2101 if (rc2 != 0 && rc == 0)
2104 lod_putref(d, &d->lod_ost_descs);
2106 lod_getref(&d->lod_mdt_descs);
2107 lod_foreach_mdt(d, tgt) {
2108 if (!tgt->ltd_active)
2110 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2112 if (rc2 != 0 && rc == 0)
2115 lod_putref(d, &d->lod_mdt_descs);
2119 rc2 = ptlrpc_set_wait(env, set);
2120 if (rc2 == 0 && rc == 0)
2122 ptlrpc_set_destroy(set);
2127 static struct obd_ops lod_obd_device_ops = {
2128 .o_owner = THIS_MODULE,
2129 .o_connect = lod_obd_connect,
2130 .o_disconnect = lod_obd_disconnect,
2131 .o_get_info = lod_obd_get_info,
2132 .o_set_info_async = lod_obd_set_info_async,
2133 .o_pool_new = lod_pool_new,
2134 .o_pool_rem = lod_pool_remove,
2135 .o_pool_add = lod_pool_add,
2136 .o_pool_del = lod_pool_del,
2139 static int __init lod_init(void)
2141 struct obd_type *sym;
2144 rc = lu_kmem_init(lod_caches);
2148 rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
2149 LUSTRE_LOD_NAME, &lod_device_type);
2151 lu_kmem_fini(lod_caches);
2155 /* create "lov" entry for compatibility purposes */
2156 sym = class_add_symlinks(LUSTRE_LOV_NAME, true);
2159 /* does real "lov" already exist ? */
2167 static void __exit lod_exit(void)
2169 struct obd_type *sym = class_search_type(LUSTRE_LOV_NAME);
2171 /* if this was never fully initialized by the lov layer
2172 * then we are responsible for freeing this obd_type
2175 /* final put if we manage this obd type */
2176 if (sym->typ_sym_filter)
2177 kobject_put(&sym->typ_kobj);
2178 /* put reference taken by class_search_type */
2179 kobject_put(&sym->typ_kobj);
2182 class_unregister_type(LUSTRE_LOD_NAME);
2183 lu_kmem_fini(lod_caches);
2186 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2187 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
2188 MODULE_VERSION(LUSTRE_VERSION_STRING);
2189 MODULE_LICENSE("GPL");
2191 module_init(lod_init);
2192 module_exit(lod_exit);