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 lod_tgt_descs *ltd = &lod->lod_mdt_descs;
362 struct lod_tgt_desc *tgt = NULL;
369 thread->t_flags = SVC_RUNNING;
370 wake_up(&thread->t_ctl_waitq);
372 rc = lu_env_init(&env, LCT_LOCAL | LCT_MD_THREAD);
375 CERROR("%s: can't initialize env: rc = %d\n",
376 lod2obd(lod)->obd_name, rc);
380 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
381 atomic_inc(&lut->lut_tdtd->tdtd_recovery_threads_count);
385 dt = lrd->lrd_ltd->ltd_tgt;
387 start = ktime_get_real_seconds();
390 rc = lod_sub_prep_llog(&env, lod, dt, lrd->lrd_idx);
391 if (!rc && !lod->lod_child->dd_rdonly) {
392 /* Process the recovery record */
393 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
394 LLOG_UPDATELOG_ORIG_CTXT);
395 LASSERT(ctxt != NULL);
396 LASSERT(ctxt->loc_handle != NULL);
398 rc = llog_cat_process(&env, ctxt->loc_handle,
399 lod_process_recovery_updates, lrd, 0, 0);
403 struct lu_device *top_device;
405 top_device = lod->lod_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
407 * Because the remote target might failover at the same time,
410 if ((rc == -ETIMEDOUT || rc == -EAGAIN || rc == -EIO) &&
411 dt != lod->lod_child &&
412 !top_device->ld_obd->obd_abort_recovery &&
413 !top_device->ld_obd->obd_stopping) {
415 if (ctxt->loc_handle)
421 CDEBUG(D_HA, "%s get update log failed %d, retry\n",
422 dt->dd_lu_dev.ld_obd->obd_name, rc);
426 CERROR("%s get update log failed: rc = %d\n",
427 dt->dd_lu_dev.ld_obd->obd_name, rc);
430 spin_lock(&top_device->ld_obd->obd_dev_lock);
431 if (!top_device->ld_obd->obd_abort_recovery &&
432 !top_device->ld_obd->obd_stopping)
433 top_device->ld_obd->obd_abort_recovery = 1;
434 spin_unlock(&top_device->ld_obd->obd_dev_lock);
440 CDEBUG(D_HA, "%s retrieved update log, duration %lld, retries %d\n",
441 dt->dd_lu_dev.ld_obd->obd_name, ktime_get_real_seconds() - start,
444 spin_lock(&lod->lod_lock);
446 lod->lod_child_got_update_log = 1;
448 lrd->lrd_ltd->ltd_got_update_log = 1;
450 if (!lod->lod_child_got_update_log) {
451 spin_unlock(&lod->lod_lock);
455 ltd_foreach_tgt(ltd, tgt) {
456 if (!tgt->ltd_got_update_log) {
457 spin_unlock(&lod->lod_lock);
461 lut->lut_tdtd->tdtd_replay_ready = 1;
462 spin_unlock(&lod->lod_lock);
464 CDEBUG(D_HA, "%s got update logs from all MDTs.\n",
465 lut->lut_obd->obd_name);
466 wake_up(&lut->lut_obd->obd_next_transno_waitq);
471 thread->t_flags = SVC_STOPPED;
472 atomic_dec(&lut->lut_tdtd->tdtd_recovery_threads_count);
473 wake_up(&lut->lut_tdtd->tdtd_recovery_threads_waitq);
474 wake_up(&thread->t_ctl_waitq);
480 * finish sub llog context
482 * Stop update recovery thread for the sub device, then cleanup the
483 * correspondent llog ctxt.
485 * \param[in] env execution environment
486 * \param[in] lod lod device to do update recovery
487 * \param[in] thread recovery thread on this sub device
489 void lod_sub_fini_llog(const struct lu_env *env,
490 struct dt_device *dt, struct ptlrpc_thread *thread)
492 struct obd_device *obd;
493 struct llog_ctxt *ctxt;
497 obd = dt->dd_lu_dev.ld_obd;
498 CDEBUG(D_INFO, "%s: finish sub llog\n", obd->obd_name);
499 /* Stop recovery thread first */
500 if (thread && thread->t_flags & SVC_RUNNING) {
501 thread->t_flags = SVC_STOPPING;
502 wake_up(&thread->t_ctl_waitq);
503 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
506 ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
510 if (ctxt->loc_handle)
511 llog_cat_close(env, ctxt->loc_handle);
513 llog_cleanup(env, ctxt);
519 * Extract MDT target index from a device name.
521 * a helper function to extract index from the given device name
522 * like "fsname-MDTxxxx-mdtlov"
524 * \param[in] lodname device name
525 * \param[out] mdt_index extracted index
527 * \retval 0 on success
528 * \retval -EINVAL if the name is invalid
530 int lodname2mdt_index(char *lodname, u32 *mdt_index)
536 /* 1.8 configs don't have "-MDT0000" at the end */
537 ptr = strstr(lodname, "-MDT");
543 ptr = strrchr(lodname, '-');
546 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
550 if (strncmp(ptr, "-mdtlov", 7) != 0) {
552 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
556 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
558 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
562 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
564 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
568 index = simple_strtol(ptr - 4, &tmp, 16);
569 if (*tmp != '-' || index > INT_MAX) {
571 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
579 * Init sub llog context
581 * Setup update llog ctxt for update recovery threads, then start the
582 * recovery thread (lod_sub_recovery_thread) to read update llog from
583 * the correspondent MDT to do update recovery.
585 * \param[in] env execution environment
586 * \param[in] lod lod device to do update recovery
587 * \param[in] dt sub dt device for which the recovery thread is
589 * \retval 0 if initialization succeeds.
590 * \retval negative errno if initialization fails.
592 int lod_sub_init_llog(const struct lu_env *env, struct lod_device *lod,
593 struct dt_device *dt)
595 struct obd_device *obd;
596 struct lod_recovery_data *lrd = NULL;
597 struct ptlrpc_thread *thread;
598 struct task_struct *task;
599 struct l_wait_info lwi = { 0 };
600 struct lod_tgt_desc *subtgt = NULL;
607 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &master_index);
615 if (lod->lod_child == dt) {
616 thread = &lod->lod_child_recovery_thread;
617 index = master_index;
619 struct lu_tgt_desc *tgt;
621 ltd_foreach_tgt(&lod->lod_mdt_descs, tgt) {
622 if (tgt->ltd_tgt == dt) {
623 index = tgt->ltd_index;
628 LASSERT(subtgt != NULL);
629 OBD_ALLOC_PTR(subtgt->ltd_recovery_thread);
630 if (!subtgt->ltd_recovery_thread)
631 GOTO(free_lrd, rc = -ENOMEM);
633 thread = subtgt->ltd_recovery_thread;
636 CDEBUG(D_INFO, "%s init sub log %s\n", lod2obd(lod)->obd_name,
637 dt->dd_lu_dev.ld_obd->obd_name);
639 lrd->lrd_ltd = subtgt;
640 lrd->lrd_thread = thread;
641 lrd->lrd_idx = index;
642 init_waitqueue_head(&thread->t_ctl_waitq);
644 obd = dt->dd_lu_dev.ld_obd;
645 obd->obd_lvfs_ctxt.dt = dt;
646 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_UPDATELOG_ORIG_CTXT,
647 NULL, &llog_common_cat_ops);
649 CERROR("%s: cannot setup updatelog llog: rc = %d\n",
651 GOTO(free_thread, rc);
654 /* Start the recovery thread */
655 task = kthread_run(lod_sub_recovery_thread, lrd, "lod%04x_rec%04x",
656 master_index, index);
659 CERROR("%s: cannot start recovery thread: rc = %d\n",
664 l_wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_RUNNING ||
665 thread->t_flags & SVC_STOPPED, &lwi);
669 lod_sub_fini_llog(env, dt, thread);
671 if (lod->lod_child != dt) {
672 OBD_FREE_PTR(subtgt->ltd_recovery_thread);
673 subtgt->ltd_recovery_thread = NULL;
681 * Stop sub recovery thread
683 * Stop sub recovery thread on all subs.
685 * \param[in] env execution environment
686 * \param[in] lod lod device to do update recovery
688 static void lod_sub_stop_recovery_threads(const struct lu_env *env,
689 struct lod_device *lod)
691 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
692 struct ptlrpc_thread *thread;
693 struct lu_tgt_desc *tgt;
696 * Stop the update log commit cancel threads and finish master
699 thread = &lod->lod_child_recovery_thread;
700 /* Stop recovery thread first */
701 if (thread && thread->t_flags & SVC_RUNNING) {
702 thread->t_flags = SVC_STOPPING;
703 wake_up(&thread->t_ctl_waitq);
704 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
708 ltd_foreach_tgt(ltd, tgt) {
709 thread = tgt->ltd_recovery_thread;
710 if (thread && thread->t_flags & SVC_RUNNING) {
711 thread->t_flags = SVC_STOPPING;
712 wake_up(&thread->t_ctl_waitq);
713 wait_event(thread->t_ctl_waitq,
714 thread->t_flags & SVC_STOPPED);
715 OBD_FREE_PTR(tgt->ltd_recovery_thread);
716 tgt->ltd_recovery_thread = NULL;
720 lod_putref(lod, ltd);
724 * finish all sub llog
726 * cleanup all of sub llog ctxt on the LOD.
728 * \param[in] env execution environment
729 * \param[in] lod lod device to do update recovery
731 static void lod_sub_fini_all_llogs(const struct lu_env *env,
732 struct lod_device *lod)
734 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
735 struct lu_tgt_desc *tgt;
738 * Stop the update log commit cancel threads and finish master
741 lod_sub_fini_llog(env, lod->lod_child,
742 &lod->lod_child_recovery_thread);
744 ltd_foreach_tgt(ltd, tgt)
745 lod_sub_fini_llog(env, tgt->ltd_tgt,
746 tgt->ltd_recovery_thread);
747 lod_putref(lod, ltd);
750 static char *lod_show_update_logs_retrievers(void *data, int *size, int *count)
752 struct lod_device *lod = (struct lod_device *)data;
753 struct lu_target *lut = lod2lu_dev(lod)->ld_site->ls_tgt;
754 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
755 struct lod_tgt_desc *tgt = NULL;
761 *count = atomic_read(&lut->lut_tdtd->tdtd_recovery_threads_count);
767 *size = 5 * *count + 1;
768 OBD_ALLOC(buf, *size);
773 memset(buf, 0, *size);
775 if (!lod->lod_child_got_update_log) {
776 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &i);
777 LASSERTF(rc == 0, "Fail to parse target index: rc = %d\n", rc);
779 rc = snprintf(buf + len, *size - len, " %04x", i);
786 ltd_foreach_tgt(ltd, tgt) {
787 if (!tgt->ltd_got_update_log) {
788 rc = snprintf(buf + len, *size - len, " %04x",
790 if (unlikely(rc <= 0))
802 * Prepare distribute txn
804 * Prepare distribute txn structure for LOD
806 * \param[in] env execution environment
807 * \param[in] lod_device LOD device
809 * \retval 0 if preparation succeeds.
810 * \retval negative errno if preparation fails.
812 static int lod_prepare_distribute_txn(const struct lu_env *env,
813 struct lod_device *lod)
815 struct target_distribute_txn_data *tdtd;
816 struct lu_target *lut;
821 /* Init update recovery data */
826 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
827 tdtd->tdtd_dt = &lod->lod_dt_dev;
828 rc = distribute_txn_init(env, lut, tdtd,
829 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id);
832 CERROR("%s: cannot init distribute txn: rc = %d\n",
833 lod2obd(lod)->obd_name, rc);
838 tdtd->tdtd_show_update_logs_retrievers =
839 lod_show_update_logs_retrievers;
840 tdtd->tdtd_show_retrievers_cbdata = lod;
842 lut->lut_tdtd = tdtd;
848 * Finish distribute txn
850 * Release the resource holding by distribute txn, i.e. stop distribute
853 * \param[in] env execution environment
854 * \param[in] lod lod device
856 static void lod_fini_distribute_txn(const struct lu_env *env,
857 struct lod_device *lod)
859 struct lu_target *lut;
861 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
862 target_recovery_fini(lut->lut_obd);
866 distribute_txn_fini(env, lut->lut_tdtd);
868 OBD_FREE_PTR(lut->lut_tdtd);
869 lut->lut_tdtd = NULL;
873 * Implementation of lu_device_operations::ldo_process_config() for LOD
875 * The method is called by the configuration subsystem during setup,
876 * cleanup and when the configuration changes. The method processes
877 * few specific commands like adding/removing the targets, changing
878 * the runtime parameters.
880 * \param[in] env LU environment provided by the caller
881 * \param[in] dev lod device
882 * \param[in] lcfg configuration command to apply
884 * \retval 0 on success
885 * \retval negative negated errno on error
887 * The examples are below.
889 * Add osc config log:
890 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
891 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
892 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
893 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
894 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
895 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
897 * Add mdc config log:
898 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
899 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
900 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
901 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
902 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
903 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
905 static int lod_process_config(const struct lu_env *env,
906 struct lu_device *dev,
907 struct lustre_cfg *lcfg)
909 struct lod_device *lod = lu2lod_dev(dev);
910 struct lu_device *next = &lod->lod_child->dd_lu_dev;
916 switch (lcfg->lcfg_command) {
917 case LCFG_LOV_DEL_OBD:
918 case LCFG_LOV_ADD_INA:
919 case LCFG_LOV_ADD_OBD:
925 * lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
926 * modify_mdc_tgts add 0:lustre-MDT0001
927 * 1:lustre-MDT0001-mdc0002
930 arg1 = lustre_cfg_string(lcfg, 1);
932 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
933 GOTO(out, rc = -EINVAL);
934 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
935 GOTO(out, rc = -EINVAL);
937 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
940 rc = lodname2mdt_index(lustre_cfg_string(lcfg, 0),
945 rc = lod_add_device(env, lod, arg1, index, gen,
946 mdt_index, LUSTRE_OSC_NAME, 1);
947 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
949 rc = lod_add_device(env, lod, arg1, index, gen,
950 mdt_index, LUSTRE_MDC_NAME, 1);
951 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
952 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
954 rc = lod_add_device(env, lod, arg1, index, gen,
955 mdt_index, LUSTRE_OSC_NAME, 0);
957 rc = lod_del_device(env, lod,
959 arg1, index, gen, true);
966 struct obd_device *obd;
971 * Check if it is activate/deactivate mdc
972 * lustre-MDTXXXX-osp-MDTXXXX.active=1
974 param = lustre_cfg_buf(lcfg, 1);
975 if (strstr(param, "osp") && strstr(param, ".active=")) {
976 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
977 struct lod_tgt_desc *sub_tgt = NULL;
978 struct lu_tgt_desc *tgt;
982 ptr = strstr(param, ".");
984 obd = class_name2obd(param);
986 CERROR("%s: can not find %s: rc = %d\n",
987 lod2obd(lod)->obd_name, param, -EINVAL);
992 ltd_foreach_tgt(ltd, tgt) {
993 if (tgt->ltd_tgt->dd_lu_dev.ld_obd == obd) {
1000 CERROR("%s: can not find %s: rc = %d\n",
1001 lod2obd(lod)->obd_name, param, -EINVAL);
1007 tmp = strstr(param, "=");
1010 struct llog_ctxt *ctxt;
1012 obd = sub_tgt->ltd_tgt->dd_lu_dev.ld_obd;
1013 ctxt = llog_get_context(obd,
1014 LLOG_UPDATELOG_ORIG_CTXT);
1016 rc = llog_setup(env, obd, &obd->obd_olg,
1017 LLOG_UPDATELOG_ORIG_CTXT,
1018 NULL, &llog_common_cat_ops);
1022 llog_ctxt_put(ctxt);
1024 rc = lod_sub_prep_llog(env, lod,
1026 sub_tgt->ltd_index);
1028 sub_tgt->ltd_active = 1;
1030 lod_sub_fini_llog(env, sub_tgt->ltd_tgt,
1032 sub_tgt->ltd_active = 0;
1038 if (strstr(param, PARAM_LOD) != NULL)
1039 count = class_modify_config(lcfg, PARAM_LOD,
1040 &lod->lod_dt_dev.dd_kobj);
1042 count = class_modify_config(lcfg, PARAM_LOV,
1043 &lod->lod_dt_dev.dd_kobj);
1044 rc = count > 0 ? 0 : count;
1047 case LCFG_PRE_CLEANUP: {
1048 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1049 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1050 next = &lod->lod_child->dd_lu_dev;
1051 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1053 CDEBUG(D_HA, "%s: can't process %u: %d\n",
1054 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1056 lod_sub_stop_recovery_threads(env, lod);
1057 lod_fini_distribute_txn(env, lod);
1058 lod_sub_fini_all_llogs(env, lod);
1061 case LCFG_CLEANUP: {
1062 if (lod->lod_md_root) {
1063 dt_object_put(env, &lod->lod_md_root->ldo_obj);
1064 lod->lod_md_root = NULL;
1068 * do cleanup on underlying storage only when
1069 * all OSPs are cleaned up, as they use that OSD as well
1071 lu_dev_del_linkage(dev->ld_site, dev);
1072 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1073 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1074 next = &lod->lod_child->dd_lu_dev;
1075 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1077 CERROR("%s: can't process %u: rc = %d\n",
1078 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1080 rc = obd_disconnect(lod->lod_child_exp);
1082 CERROR("error in disconnect from storage: rc = %d\n",
1087 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
1088 lcfg->lcfg_command);
1098 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
1100 * The method is called once the recovery is complete. This implementation
1101 * distributes the notification to all the known targets.
1103 * see include/lu_object.h for the details
1105 static int lod_recovery_complete(const struct lu_env *env,
1106 struct lu_device *dev)
1108 struct lod_device *lod = lu2lod_dev(dev);
1109 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1115 LASSERT(lod->lod_recovery_completed == 0);
1116 lod->lod_recovery_completed = 1;
1118 rc = next->ld_ops->ldo_recovery_complete(env, next);
1120 lod_getref(&lod->lod_ost_descs);
1121 if (lod->lod_osts_size > 0) {
1122 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
1123 struct lod_tgt_desc *tgt;
1125 tgt = OST_TGT(lod, i);
1126 LASSERT(tgt && tgt->ltd_tgt);
1127 next = &tgt->ltd_ost->dd_lu_dev;
1128 rc = next->ld_ops->ldo_recovery_complete(env, next);
1130 CERROR("%s: can't complete recovery on #%d: rc = %d\n",
1131 lod2obd(lod)->obd_name, i, rc);
1134 lod_putref(lod, &lod->lod_ost_descs);
1139 * Init update logs on all sub device
1141 * LOD initialize update logs on all of sub devices. Because the initialization
1142 * process might need FLD lookup, see llog_osd_open()->dt_locate()->...->
1143 * lod_object_init(), this API has to be called after LOD is initialized.
1144 * \param[in] env execution environment
1145 * \param[in] lod lod device
1147 * \retval 0 if update log is initialized successfully.
1148 * \retval negative errno if initialization fails.
1150 static int lod_sub_init_llogs(const struct lu_env *env, struct lod_device *lod)
1152 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1153 struct lu_tgt_desc *tgt;
1159 * llog must be setup after LOD is initialized, because llog
1160 * initialization include FLD lookup
1162 LASSERT(lod->lod_initialized);
1164 /* Init the llog in its own stack */
1165 rc = lod_sub_init_llog(env, lod, lod->lod_child);
1169 ltd_foreach_tgt(ltd, tgt) {
1170 rc = lod_sub_init_llog(env, lod, tgt->ltd_tgt);
1179 * Implementation of lu_device_operations::ldo_prepare() for LOD
1181 * see include/lu_object.h for the details.
1183 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
1184 struct lu_device *cdev)
1186 struct lod_device *lod = lu2lod_dev(cdev);
1187 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1188 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1190 struct dt_object *root;
1191 struct dt_object *dto;
1196 rc = next->ld_ops->ldo_prepare(env, pdev, next);
1198 CERROR("%s: prepare bottom error: rc = %d\n",
1199 lod2obd(lod)->obd_name, rc);
1203 lod->lod_initialized = 1;
1205 rc = dt_root_get(env, lod->lod_child, fid);
1209 root = dt_locate(env, lod->lod_child, fid);
1211 RETURN(PTR_ERR(root));
1213 /* Create update log object */
1214 index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1215 lu_update_log_fid(fid, index);
1217 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1219 lod_update_log_name,
1222 GOTO(out_put, rc = PTR_ERR(dto));
1224 dt_object_put(env, dto);
1226 /* Create update log dir */
1227 lu_update_log_dir_fid(fid, index);
1228 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1230 lod_update_log_dir_name,
1233 GOTO(out_put, rc = PTR_ERR(dto));
1235 dt_object_put(env, dto);
1237 rc = lod_prepare_distribute_txn(env, lod);
1241 rc = lod_sub_init_llogs(env, lod);
1246 dt_object_put(env, root);
1251 const struct lu_device_operations lod_lu_ops = {
1252 .ldo_object_alloc = lod_object_alloc,
1253 .ldo_process_config = lod_process_config,
1254 .ldo_recovery_complete = lod_recovery_complete,
1255 .ldo_prepare = lod_prepare,
1259 * Implementation of dt_device_operations::dt_root_get() for LOD
1261 * see include/dt_object.h for the details.
1263 static int lod_root_get(const struct lu_env *env,
1264 struct dt_device *dev, struct lu_fid *f)
1266 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
1269 static void lod_statfs_sum(struct obd_statfs *sfs,
1270 struct obd_statfs *ost_sfs, int *bs)
1272 while (ost_sfs->os_bsize < *bs) {
1274 sfs->os_bsize >>= 1;
1275 sfs->os_bavail <<= 1;
1276 sfs->os_blocks <<= 1;
1277 sfs->os_bfree <<= 1;
1278 sfs->os_granted <<= 1;
1280 while (ost_sfs->os_bsize > *bs) {
1281 ost_sfs->os_bsize >>= 1;
1282 ost_sfs->os_bavail <<= 1;
1283 ost_sfs->os_blocks <<= 1;
1284 ost_sfs->os_bfree <<= 1;
1285 ost_sfs->os_granted <<= 1;
1287 sfs->os_bavail += ost_sfs->os_bavail;
1288 sfs->os_blocks += ost_sfs->os_blocks;
1289 sfs->os_bfree += ost_sfs->os_bfree;
1290 sfs->os_granted += ost_sfs->os_granted;
1294 * Implementation of dt_device_operations::dt_statfs() for LOD
1296 * see include/dt_object.h for the details.
1298 static int lod_statfs(const struct lu_env *env, struct dt_device *dev,
1299 struct obd_statfs *sfs, struct obd_statfs_info *info)
1301 struct lod_device *lod = dt2lod_dev(dev);
1302 struct lod_ost_desc *ost;
1303 struct lod_mdt_desc *mdt;
1304 struct obd_statfs ost_sfs;
1309 rc = dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
1318 sfs->os_granted = 0;
1320 lod_getref(&lod->lod_mdt_descs);
1321 lod_foreach_mdt(lod, i) {
1322 mdt = MDT_TGT(lod, i);
1323 LASSERT(mdt && mdt->ltd_mdt);
1324 rc = dt_statfs(env, mdt->ltd_mdt, &ost_sfs);
1328 sfs->os_files += ost_sfs.os_files;
1329 sfs->os_ffree += ost_sfs.os_ffree;
1330 lod_statfs_sum(sfs, &ost_sfs, &bs);
1332 lod_putref(lod, &lod->lod_mdt_descs);
1335 * at some point we can check whether DoM is enabled and
1336 * decide how to account MDT space. for simplicity let's
1337 * just fallback to pre-DoM policy if any OST is alive
1339 lod_getref(&lod->lod_ost_descs);
1340 lod_foreach_ost(lod, i) {
1341 ost = OST_TGT(lod, i);
1342 LASSERT(ost && ost->ltd_ost);
1343 rc = dt_statfs(env, ost->ltd_ost, &ost_sfs);
1345 if (rc || ost_sfs.os_bsize == 0)
1349 * if only MDTs with DoM then report only MDT blocks,
1350 * otherwise show only OST blocks, and DoM is "free"
1355 sfs->os_granted = 0;
1357 ost_files += ost_sfs.os_files;
1358 ost_ffree += ost_sfs.os_ffree;
1359 ost_sfs.os_bavail += ost_sfs.os_granted;
1360 lod_statfs_sum(sfs, &ost_sfs, &bs);
1361 LASSERTF(bs == ost_sfs.os_bsize, "%d != %d\n",
1362 (int)sfs->os_bsize, (int)ost_sfs.os_bsize);
1364 lod_putref(lod, &lod->lod_ost_descs);
1365 sfs->os_state |= OS_STATE_SUM;
1367 /* If we have _some_ OSTs, but don't have as many free objects on the
1368 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1369 * to compensate, so that the "inodes in use" number is correct.
1370 * This should be kept in sync with ll_statfs_internal().
1372 if (ost_files && ost_ffree < sfs->os_ffree) {
1373 sfs->os_files = (sfs->os_files - sfs->os_ffree) + ost_ffree;
1374 sfs->os_ffree = ost_ffree;
1377 /* a single successful statfs should be enough */
1385 * Implementation of dt_device_operations::dt_trans_create() for LOD
1387 * Creates a transaction using local (to this node) OSD.
1389 * see include/dt_object.h for the details.
1391 static struct thandle *lod_trans_create(const struct lu_env *env,
1392 struct dt_device *dt)
1396 th = top_trans_create(env, dt2lod_dev(dt)->lod_child);
1406 * Implementation of dt_device_operations::dt_trans_start() for LOD
1408 * Starts the set of local transactions using the targets involved
1409 * in declare phase. Initial support for the distributed transactions.
1411 * see include/dt_object.h for the details.
1413 static int lod_trans_start(const struct lu_env *env, struct dt_device *dt,
1416 return top_trans_start(env, dt2lod_dev(dt)->lod_child, th);
1419 static int lod_trans_cb_add(struct thandle *th,
1420 struct dt_txn_commit_cb *dcb)
1422 struct top_thandle *top_th = container_of(th, struct top_thandle,
1424 return dt_trans_cb_add(top_th->tt_master_sub_thandle, dcb);
1428 * add noop update to the update records
1430 * Add noop updates to the update records, which is only used in
1433 * \param[in] env execution environment
1434 * \param[in] dt dt device of lod
1435 * \param[in] th thandle
1436 * \param[in] count the count of update records to be added.
1438 * \retval 0 if adding succeeds.
1439 * \retval negative errno if adding fails.
1441 static int lod_add_noop_records(const struct lu_env *env,
1442 struct dt_device *dt, struct thandle *th,
1445 struct top_thandle *top_th;
1446 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1450 top_th = container_of(th, struct top_thandle, tt_super);
1451 if (!top_th->tt_multiple_thandle)
1455 for (i = 0; i < count; i++) {
1456 rc = update_record_pack(noop, th, fid);
1464 * Implementation of dt_device_operations::dt_trans_stop() for LOD
1466 * Stops the set of local transactions using the targets involved
1467 * in declare phase. Initial support for the distributed transactions.
1469 * see include/dt_object.h for the details.
1471 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
1474 if (OBD_FAIL_CHECK(OBD_FAIL_SPLIT_UPDATE_REC)) {
1477 rc = lod_add_noop_records(env, dt, th, 5000);
1481 return top_trans_stop(env, dt2lod_dev(dt)->lod_child, th);
1485 * Implementation of dt_device_operations::dt_conf_get() for LOD
1487 * Currently returns the configuration provided by the local OSD.
1489 * see include/dt_object.h for the details.
1491 static void lod_conf_get(const struct lu_env *env,
1492 const struct dt_device *dev,
1493 struct dt_device_param *param)
1495 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
1499 * Implementation of dt_device_operations::dt_sync() for LOD
1501 * Syncs all known OST targets. Very very expensive and used
1502 * rarely by LFSCK now. Should not be used in general.
1504 * see include/dt_object.h for the details.
1506 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
1508 struct lod_device *lod = dt2lod_dev(dev);
1509 struct lod_ost_desc *ost;
1510 struct lod_mdt_desc *mdt;
1516 lod_getref(&lod->lod_ost_descs);
1517 lod_foreach_ost(lod, i) {
1518 ost = OST_TGT(lod, i);
1519 LASSERT(ost && ost->ltd_ost);
1520 if (!ost->ltd_active)
1522 rc = dt_sync(env, ost->ltd_ost);
1524 if (rc != -ENOTCONN) {
1525 CERROR("%s: can't sync ost %u: rc = %d\n",
1526 lod2obd(lod)->obd_name, i, rc);
1532 lod_putref(lod, &lod->lod_ost_descs);
1537 lod_getref(&lod->lod_mdt_descs);
1538 lod_foreach_mdt(lod, i) {
1539 mdt = MDT_TGT(lod, i);
1540 LASSERT(mdt && mdt->ltd_mdt);
1541 if (!mdt->ltd_active)
1543 rc = dt_sync(env, mdt->ltd_mdt);
1545 if (rc != -ENOTCONN) {
1546 CERROR("%s: can't sync mdt %u: rc = %d\n",
1547 lod2obd(lod)->obd_name, i, rc);
1553 lod_putref(lod, &lod->lod_mdt_descs);
1556 rc = dt_sync(env, lod->lod_child);
1562 * Implementation of dt_device_operations::dt_ro() for LOD
1564 * Turns local OSD read-only, used for the testing only.
1566 * see include/dt_object.h for the details.
1568 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
1570 return dt_ro(env, dt2lod_dev(dev)->lod_child);
1574 * Implementation of dt_device_operations::dt_commit_async() for LOD
1576 * Asks local OSD to commit sooner.
1578 * see include/dt_object.h for the details.
1580 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
1582 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
1585 static const struct dt_device_operations lod_dt_ops = {
1586 .dt_root_get = lod_root_get,
1587 .dt_statfs = lod_statfs,
1588 .dt_trans_create = lod_trans_create,
1589 .dt_trans_start = lod_trans_start,
1590 .dt_trans_stop = lod_trans_stop,
1591 .dt_conf_get = lod_conf_get,
1592 .dt_sync = lod_sync,
1594 .dt_commit_async = lod_commit_async,
1595 .dt_trans_cb_add = lod_trans_cb_add,
1599 * Connect to a local OSD.
1601 * Used to connect to the local OSD at mount. OSD name is taken from the
1602 * configuration command passed. This connection is used to identify LU
1603 * site and pin the OSD from early removal.
1605 * \param[in] env LU environment provided by the caller
1606 * \param[in] lod lod device
1607 * \param[in] cfg configuration command to apply
1609 * \retval 0 on success
1610 * \retval negative negated errno on error
1612 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
1613 struct lustre_cfg *cfg)
1615 struct obd_connect_data *data = NULL;
1616 struct obd_device *obd;
1617 char *nextdev = NULL, *p, *s;
1623 LASSERT(lod->lod_child_exp == NULL);
1626 * compatibility hack: we still use old config logs
1627 * which specify LOV, but we need to learn underlying
1628 * OSD device, which is supposed to be:
1629 * <fsname>-MDTxxxx-osd
1631 * 2.x MGS generates lines like the following:
1632 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
1633 * 1.8 MGS generates lines like the following:
1634 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
1636 * we use "-MDT" to differentiate 2.x from 1.8
1638 p = lustre_cfg_string(cfg, 0);
1639 if (p && strstr(p, "-mdtlov")) {
1640 len = strlen(p) + 6;
1641 OBD_ALLOC(nextdev, len);
1643 GOTO(out, rc = -ENOMEM);
1646 s = strstr(nextdev, "-mdtlov");
1648 CERROR("%s: unable to parse device name: rc = %d\n",
1649 lustre_cfg_string(cfg, 0), -EINVAL);
1650 GOTO(out, rc = -EINVAL);
1653 if (strstr(nextdev, "-MDT")) {
1658 strcpy(s, "-MDT0000-osd");
1661 CERROR("%s: unable to parse device name: rc = %d\n",
1662 lustre_cfg_string(cfg, 0), -EINVAL);
1663 GOTO(out, rc = -EINVAL);
1666 OBD_ALLOC_PTR(data);
1668 GOTO(out, rc = -ENOMEM);
1670 obd = class_name2obd(nextdev);
1672 CERROR("%s: can not locate next device: rc = %d\n",
1673 nextdev, -ENOTCONN);
1674 GOTO(out, rc = -ENOTCONN);
1677 data->ocd_connect_flags = OBD_CONNECT_VERSION;
1678 data->ocd_version = LUSTRE_VERSION_CODE;
1680 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
1683 CERROR("%s: cannot connect to next dev: rc = %d\n",
1688 lod->lod_dt_dev.dd_lu_dev.ld_site =
1689 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
1690 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
1691 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
1697 OBD_FREE(nextdev, len);
1702 * Initialize LOD device at setup.
1704 * Initializes the given LOD device using the original configuration command.
1705 * The function initiates a connection to the local OSD and initializes few
1706 * internal structures like pools, target tables, etc.
1708 * \param[in] env LU environment provided by the caller
1709 * \param[in] lod lod device
1710 * \param[in] ldt not used
1711 * \param[in] cfg configuration command
1713 * \retval 0 on success
1714 * \retval negative negated errno on error
1716 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
1717 struct lu_device_type *ldt, struct lustre_cfg *cfg)
1719 struct dt_device_param ddp;
1720 struct obd_device *obd;
1725 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1728 CERROR("Cannot find obd with name '%s': rc = %d\n",
1729 lustre_cfg_string(cfg, 0), rc);
1733 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1734 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
1735 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
1736 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
1738 rc = lod_connect_to_osd(env, lod, cfg);
1742 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
1743 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
1744 lod->lod_dom_max_stripesize = (1ULL << 20); /* 1Mb as default value */
1746 /* setup obd to be used with old lov code */
1747 rc = lod_pools_init(lod, cfg);
1749 GOTO(out_disconnect, rc);
1751 rc = lod_procfs_init(lod);
1753 GOTO(out_pools, rc);
1755 spin_lock_init(&lod->lod_lock);
1756 spin_lock_init(&lod->lod_connects_lock);
1757 lu_tgt_descs_init(&lod->lod_mdt_descs);
1758 lu_tgt_descs_init(&lod->lod_ost_descs);
1763 lod_pools_fini(lod);
1765 obd_disconnect(lod->lod_child_exp);
1770 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
1772 * Releases the memory allocated for LOD device.
1774 * see include/lu_object.h for the details.
1776 static struct lu_device *lod_device_free(const struct lu_env *env,
1777 struct lu_device *lu)
1779 struct lod_device *lod = lu2lod_dev(lu);
1780 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1784 if (atomic_read(&lu->ld_ref) > 0 &&
1785 !cfs_hash_is_empty(lu->ld_site->ls_obj_hash)) {
1786 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1787 lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
1789 LASSERTF(atomic_read(&lu->ld_ref) == 0, "lu is %p\n", lu);
1790 dt_device_fini(&lod->lod_dt_dev);
1796 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
1798 * Allocates LOD device and calls the helpers to initialize it.
1800 * see include/lu_object.h for the details.
1802 static struct lu_device *lod_device_alloc(const struct lu_env *env,
1803 struct lu_device_type *type,
1804 struct lustre_cfg *lcfg)
1806 struct lod_device *lod;
1807 struct lu_device *lu_dev;
1811 lu_dev = ERR_PTR(-ENOMEM);
1815 lu_dev = lod2lu_dev(lod);
1816 dt_device_init(&lod->lod_dt_dev, type);
1817 rc = lod_init0(env, lod, type, lcfg);
1819 lod_device_free(env, lu_dev);
1820 lu_dev = ERR_PTR(rc);
1827 static void lod_avoid_guide_fini(struct lod_avoid_guide *lag)
1829 if (lag->lag_oss_avoid_array)
1830 OBD_FREE(lag->lag_oss_avoid_array,
1831 sizeof(u32) * lag->lag_oaa_size);
1832 if (lag->lag_ost_avoid_bitmap)
1833 CFS_FREE_BITMAP(lag->lag_ost_avoid_bitmap);
1837 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
1839 * Releases the internal resources used by LOD device.
1841 * see include/lu_object.h for the details.
1843 static struct lu_device *lod_device_fini(const struct lu_env *env,
1844 struct lu_device *d)
1846 struct lod_device *lod = lu2lod_dev(d);
1851 lod_pools_fini(lod);
1853 lod_procfs_fini(lod);
1855 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs, true);
1857 CERROR("%s: can not fini ost descriptors: rc = %d\n",
1858 lod2obd(lod)->obd_name, rc);
1860 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs, false);
1862 CERROR("%s: can not fini mdt descriptors: rc = %d\n",
1863 lod2obd(lod)->obd_name, rc);
1869 * Implementation of obd_ops::o_connect() for LOD
1871 * Used to track all the users of this specific LOD device,
1872 * so the device stays up until the last user disconnected.
1874 * \param[in] env LU environment provided by the caller
1875 * \param[out] exp export the caller will be using to access LOD
1876 * \param[in] obd OBD device representing LOD device
1877 * \param[in] cluuid unique identifier of the caller
1878 * \param[in] data not used
1879 * \param[in] localdata not used
1881 * \retval 0 on success
1882 * \retval negative negated errno on error
1884 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
1885 struct obd_device *obd, struct obd_uuid *cluuid,
1886 struct obd_connect_data *data, void *localdata)
1888 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1889 struct lustre_handle conn;
1894 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1896 rc = class_connect(&conn, obd, cluuid);
1900 *exp = class_conn2export(&conn);
1902 spin_lock(&lod->lod_connects_lock);
1903 lod->lod_connects++;
1904 /* at the moment we expect the only user */
1905 LASSERT(lod->lod_connects == 1);
1906 spin_unlock(&lod->lod_connects_lock);
1913 * Implementation of obd_ops::o_disconnect() for LOD
1915 * When the caller doesn't need to use this LOD instance, it calls
1916 * obd_disconnect() and LOD releases corresponding export/reference count.
1917 * Once all the users gone, LOD device is released.
1919 * \param[in] exp export provided to the caller in obd_connect()
1921 * \retval 0 on success
1922 * \retval negative negated errno on error
1924 static int lod_obd_disconnect(struct obd_export *exp)
1926 struct obd_device *obd = exp->exp_obd;
1927 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1928 int rc, release = 0;
1932 /* Only disconnect the underlying layers on the final disconnect. */
1933 spin_lock(&lod->lod_connects_lock);
1934 lod->lod_connects--;
1935 if (lod->lod_connects != 0) {
1936 /* why should there be more than 1 connect? */
1937 spin_unlock(&lod->lod_connects_lock);
1938 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
1942 spin_unlock(&lod->lod_connects_lock);
1944 /* the last user of lod has gone, let's release the device */
1948 rc = class_disconnect(exp); /* bz 9811 */
1950 if (rc == 0 && release)
1951 class_manual_cleanup(obd);
1955 LU_KEY_INIT(lod, struct lod_thread_info);
1957 static void lod_key_fini(const struct lu_context *ctx,
1958 struct lu_context_key *key, void *data)
1960 struct lod_thread_info *info = data;
1961 struct lod_layout_component *lds =
1962 info->lti_def_striping.lds_def_comp_entries;
1965 * allocated in lod_get_lov_ea
1966 * XXX: this is overload, a tread may have such store but used only
1967 * once. Probably better would be pool of such stores per LOD.
1969 if (info->lti_ea_store) {
1970 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
1971 info->lti_ea_store = NULL;
1972 info->lti_ea_store_size = 0;
1974 lu_buf_free(&info->lti_linkea_buf);
1977 lod_free_def_comp_entries(&info->lti_def_striping);
1979 if (info->lti_comp_size > 0)
1980 OBD_FREE(info->lti_comp_idx,
1981 info->lti_comp_size * sizeof(u32));
1983 lod_avoid_guide_fini(&info->lti_avoid);
1988 /* context key: lod_thread_key */
1989 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
1991 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
1993 static struct lu_device_type_operations lod_device_type_ops = {
1994 .ldto_init = lod_type_init,
1995 .ldto_fini = lod_type_fini,
1997 .ldto_start = lod_type_start,
1998 .ldto_stop = lod_type_stop,
2000 .ldto_device_alloc = lod_device_alloc,
2001 .ldto_device_free = lod_device_free,
2003 .ldto_device_fini = lod_device_fini
2006 static struct lu_device_type lod_device_type = {
2007 .ldt_tags = LU_DEVICE_DT,
2008 .ldt_name = LUSTRE_LOD_NAME,
2009 .ldt_ops = &lod_device_type_ops,
2010 .ldt_ctx_tags = LCT_MD_THREAD,
2014 * Implementation of obd_ops::o_get_info() for LOD
2016 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
2017 * the caller binary status whether LOD has seen connection to any OST target.
2018 * It will also check if the MDT update log context being initialized (if
2021 * \param[in] env LU environment provided by the caller
2022 * \param[in] exp export of the caller
2023 * \param[in] keylen len of the key
2024 * \param[in] key the key
2025 * \param[in] vallen not used
2026 * \param[in] val not used
2028 * \retval 0 if a connection was seen
2029 * \retval -EAGAIN if LOD isn't running yet or no
2030 * connection has been seen yet
2031 * \retval -EINVAL if not supported key is requested
2033 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
2034 u32 keylen, void *key, u32 *vallen, void *val)
2038 if (KEY_IS(KEY_OSP_CONNECTED)) {
2039 struct obd_device *obd = exp->exp_obd;
2040 struct lod_device *d;
2041 struct lod_tgt_desc *tgt;
2045 if (!obd->obd_set_up || obd->obd_stopping)
2048 d = lu2lod_dev(obd->obd_lu_dev);
2049 lod_getref(&d->lod_ost_descs);
2050 lod_foreach_ost(d, i) {
2051 tgt = OST_TGT(d, i);
2052 LASSERT(tgt && tgt->ltd_tgt);
2053 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2055 /* one healthy device is enough */
2059 lod_putref(d, &d->lod_ost_descs);
2061 lod_getref(&d->lod_mdt_descs);
2062 lod_foreach_mdt(d, i) {
2063 struct llog_ctxt *ctxt;
2065 tgt = MDT_TGT(d, i);
2066 LASSERT(tgt != NULL);
2067 LASSERT(tgt->ltd_tgt != NULL);
2068 if (!tgt->ltd_active)
2071 ctxt = llog_get_context(tgt->ltd_tgt->dd_lu_dev.ld_obd,
2072 LLOG_UPDATELOG_ORIG_CTXT);
2074 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2076 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2080 if (!ctxt->loc_handle) {
2081 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2083 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2085 llog_ctxt_put(ctxt);
2088 llog_ctxt_put(ctxt);
2090 lod_putref(d, &d->lod_mdt_descs);
2098 static int lod_obd_set_info_async(const struct lu_env *env,
2099 struct obd_export *exp,
2100 u32 keylen, void *key,
2101 u32 vallen, void *val,
2102 struct ptlrpc_request_set *set)
2104 struct obd_device *obd = class_exp2obd(exp);
2105 struct lod_device *d;
2106 struct lod_tgt_desc *tgt;
2114 set = ptlrpc_prep_set();
2119 d = lu2lod_dev(obd->obd_lu_dev);
2120 lod_getref(&d->lod_ost_descs);
2121 lod_foreach_ost(d, i) {
2122 tgt = OST_TGT(d, i);
2123 LASSERT(tgt && tgt->ltd_tgt);
2124 if (!tgt->ltd_active)
2127 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2129 if (rc2 != 0 && rc == 0)
2132 lod_putref(d, &d->lod_ost_descs);
2134 lod_getref(&d->lod_mdt_descs);
2135 lod_foreach_mdt(d, i) {
2136 tgt = MDT_TGT(d, i);
2137 LASSERT(tgt && tgt->ltd_tgt);
2138 if (!tgt->ltd_active)
2140 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2142 if (rc2 != 0 && rc == 0)
2145 lod_putref(d, &d->lod_mdt_descs);
2149 rc2 = ptlrpc_set_wait(env, set);
2150 if (rc2 == 0 && rc == 0)
2152 ptlrpc_set_destroy(set);
2157 static struct obd_ops lod_obd_device_ops = {
2158 .o_owner = THIS_MODULE,
2159 .o_connect = lod_obd_connect,
2160 .o_disconnect = lod_obd_disconnect,
2161 .o_get_info = lod_obd_get_info,
2162 .o_set_info_async = lod_obd_set_info_async,
2163 .o_pool_new = lod_pool_new,
2164 .o_pool_rem = lod_pool_remove,
2165 .o_pool_add = lod_pool_add,
2166 .o_pool_del = lod_pool_del,
2169 static struct obd_type *sym;
2171 static int __init lod_init(void)
2175 rc = lu_kmem_init(lod_caches);
2179 rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
2180 LUSTRE_LOD_NAME, &lod_device_type);
2182 lu_kmem_fini(lod_caches);
2186 /* create "lov" entry for compatibility purposes */
2187 sym = class_add_symlinks(LUSTRE_LOV_NAME, true);
2190 /* does real "lov" already exist ? */
2198 static void __exit lod_exit(void)
2200 if (!IS_ERR_OR_NULL(sym))
2201 kobject_put(&sym->typ_kobj);
2203 class_unregister_type(LUSTRE_LOD_NAME);
2204 lu_kmem_fini(lod_caches);
2207 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2208 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
2209 MODULE_VERSION(LUSTRE_VERSION_STRING);
2210 MODULE_LICENSE("GPL");
2212 module_init(lod_init);
2213 module_exit(lod_exit);