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 l_wait_info lwi = { 0 };
599 struct lod_tgt_desc *subtgt = NULL;
606 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &master_index);
614 if (lod->lod_child == dt) {
615 thread = &lod->lod_child_recovery_thread;
616 index = master_index;
618 struct lu_tgt_desc *mdt;
620 lod_foreach_mdt(lod, mdt) {
621 if (mdt->ltd_tgt == dt) {
622 index = mdt->ltd_index;
627 LASSERT(subtgt != NULL);
628 OBD_ALLOC_PTR(subtgt->ltd_recovery_thread);
629 if (!subtgt->ltd_recovery_thread)
630 GOTO(free_lrd, rc = -ENOMEM);
632 thread = subtgt->ltd_recovery_thread;
635 CDEBUG(D_INFO, "%s init sub log %s\n", lod2obd(lod)->obd_name,
636 dt->dd_lu_dev.ld_obd->obd_name);
638 lrd->lrd_ltd = subtgt;
639 lrd->lrd_thread = thread;
640 lrd->lrd_idx = index;
641 init_waitqueue_head(&thread->t_ctl_waitq);
643 obd = dt->dd_lu_dev.ld_obd;
644 obd->obd_lvfs_ctxt.dt = dt;
645 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_UPDATELOG_ORIG_CTXT,
646 NULL, &llog_common_cat_ops);
648 CERROR("%s: cannot setup updatelog llog: rc = %d\n",
650 GOTO(free_thread, rc);
653 /* Start the recovery thread */
654 task = kthread_run(lod_sub_recovery_thread, lrd, "lod%04x_rec%04x",
655 master_index, index);
658 CERROR("%s: cannot start recovery thread: rc = %d\n",
663 l_wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_RUNNING ||
664 thread->t_flags & SVC_STOPPED, &lwi);
668 lod_sub_fini_llog(env, dt, thread);
670 if (lod->lod_child != dt) {
671 OBD_FREE_PTR(subtgt->ltd_recovery_thread);
672 subtgt->ltd_recovery_thread = NULL;
680 * Stop sub recovery thread
682 * Stop sub recovery thread on all subs.
684 * \param[in] env execution environment
685 * \param[in] lod lod device to do update recovery
687 static void lod_sub_stop_recovery_threads(const struct lu_env *env,
688 struct lod_device *lod)
690 struct ptlrpc_thread *thread;
691 struct lu_tgt_desc *mdt;
694 * Stop the update log commit cancel threads and finish master
697 thread = &lod->lod_child_recovery_thread;
698 /* Stop recovery thread first */
699 if (thread && thread->t_flags & SVC_RUNNING) {
700 thread->t_flags = SVC_STOPPING;
701 wake_up(&thread->t_ctl_waitq);
702 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
705 lod_getref(&lod->lod_mdt_descs);
706 lod_foreach_mdt(lod, mdt) {
707 thread = mdt->ltd_recovery_thread;
708 if (thread && thread->t_flags & SVC_RUNNING) {
709 thread->t_flags = SVC_STOPPING;
710 wake_up(&thread->t_ctl_waitq);
711 wait_event(thread->t_ctl_waitq,
712 thread->t_flags & SVC_STOPPED);
713 OBD_FREE_PTR(mdt->ltd_recovery_thread);
714 mdt->ltd_recovery_thread = NULL;
717 lod_putref(lod, &lod->lod_mdt_descs);
721 * finish all sub llog
723 * cleanup all of sub llog ctxt on the LOD.
725 * \param[in] env execution environment
726 * \param[in] lod lod device to do update recovery
728 static void lod_sub_fini_all_llogs(const struct lu_env *env,
729 struct lod_device *lod)
731 struct lu_tgt_desc *mdt;
734 * Stop the update log commit cancel threads and finish master
737 lod_sub_fini_llog(env, lod->lod_child,
738 &lod->lod_child_recovery_thread);
739 lod_getref(&lod->lod_mdt_descs);
740 lod_foreach_mdt(lod, mdt)
741 lod_sub_fini_llog(env, mdt->ltd_tgt,
742 mdt->ltd_recovery_thread);
743 lod_putref(lod, &lod->lod_mdt_descs);
746 static char *lod_show_update_logs_retrievers(void *data, int *size, int *count)
748 struct lod_device *lod = (struct lod_device *)data;
749 struct lu_target *lut = lod2lu_dev(lod)->ld_site->ls_tgt;
750 struct lu_tgt_desc *mdt = NULL;
756 *count = atomic_read(&lut->lut_tdtd->tdtd_recovery_threads_count);
762 *size = 5 * *count + 1;
763 OBD_ALLOC(buf, *size);
768 memset(buf, 0, *size);
770 if (!lod->lod_child_got_update_log) {
771 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &i);
772 LASSERTF(rc == 0, "Fail to parse target index: rc = %d\n", rc);
774 rc = snprintf(buf + len, *size - len, " %04x", i);
781 lod_foreach_mdt(lod, mdt) {
782 if (!mdt->ltd_got_update_log) {
783 rc = snprintf(buf + len, *size - len, " %04x",
785 if (unlikely(rc <= 0))
797 * Prepare distribute txn
799 * Prepare distribute txn structure for LOD
801 * \param[in] env execution environment
802 * \param[in] lod_device LOD device
804 * \retval 0 if preparation succeeds.
805 * \retval negative errno if preparation fails.
807 static int lod_prepare_distribute_txn(const struct lu_env *env,
808 struct lod_device *lod)
810 struct target_distribute_txn_data *tdtd;
811 struct lu_target *lut;
816 /* Init update recovery data */
821 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
822 tdtd->tdtd_dt = &lod->lod_dt_dev;
823 rc = distribute_txn_init(env, lut, tdtd,
824 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id);
827 CERROR("%s: cannot init distribute txn: rc = %d\n",
828 lod2obd(lod)->obd_name, rc);
833 tdtd->tdtd_show_update_logs_retrievers =
834 lod_show_update_logs_retrievers;
835 tdtd->tdtd_show_retrievers_cbdata = lod;
837 lut->lut_tdtd = tdtd;
843 * Finish distribute txn
845 * Release the resource holding by distribute txn, i.e. stop distribute
848 * \param[in] env execution environment
849 * \param[in] lod lod device
851 static void lod_fini_distribute_txn(const struct lu_env *env,
852 struct lod_device *lod)
854 struct lu_target *lut;
856 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
857 target_recovery_fini(lut->lut_obd);
861 distribute_txn_fini(env, lut->lut_tdtd);
863 OBD_FREE_PTR(lut->lut_tdtd);
864 lut->lut_tdtd = NULL;
868 * Implementation of lu_device_operations::ldo_process_config() for LOD
870 * The method is called by the configuration subsystem during setup,
871 * cleanup and when the configuration changes. The method processes
872 * few specific commands like adding/removing the targets, changing
873 * the runtime parameters.
875 * \param[in] env LU environment provided by the caller
876 * \param[in] dev lod device
877 * \param[in] lcfg configuration command to apply
879 * \retval 0 on success
880 * \retval negative negated errno on error
882 * The examples are below.
884 * Add osc config log:
885 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
886 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
887 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
888 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
889 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
890 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
892 * Add mdc config log:
893 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
894 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
895 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
896 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
897 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
898 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
900 static int lod_process_config(const struct lu_env *env,
901 struct lu_device *dev,
902 struct lustre_cfg *lcfg)
904 struct lod_device *lod = lu2lod_dev(dev);
905 struct lu_device *next = &lod->lod_child->dd_lu_dev;
911 switch (lcfg->lcfg_command) {
912 case LCFG_LOV_DEL_OBD:
913 case LCFG_LOV_ADD_INA:
914 case LCFG_LOV_ADD_OBD:
920 * lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
921 * modify_mdc_tgts add 0:lustre-MDT0001
922 * 1:lustre-MDT0001-mdc0002
925 arg1 = lustre_cfg_string(lcfg, 1);
927 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
928 GOTO(out, rc = -EINVAL);
929 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
930 GOTO(out, rc = -EINVAL);
932 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
935 rc = lodname2mdt_index(lustre_cfg_string(lcfg, 0),
940 rc = lod_add_device(env, lod, arg1, index, gen,
941 mdt_index, LUSTRE_OSC_NAME, 1);
942 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
944 rc = lod_add_device(env, lod, arg1, index, gen,
945 mdt_index, LUSTRE_MDC_NAME, 1);
946 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
947 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
949 rc = lod_add_device(env, lod, arg1, index, gen,
950 mdt_index, LUSTRE_OSC_NAME, 0);
952 rc = lod_del_device(env, lod, &lod->lod_ost_descs,
960 struct obd_device *obd;
965 * Check if it is activate/deactivate mdc
966 * lustre-MDTXXXX-osp-MDTXXXX.active=1
968 param = lustre_cfg_buf(lcfg, 1);
969 if (strstr(param, "osp") && strstr(param, ".active=")) {
970 struct lod_tgt_desc *sub_tgt = NULL;
971 struct lu_tgt_desc *mdt;
975 ptr = strstr(param, ".");
977 obd = class_name2obd(param);
979 CERROR("%s: can not find %s: rc = %d\n",
980 lod2obd(lod)->obd_name, param, -EINVAL);
985 lod_foreach_mdt(lod, mdt) {
986 if (mdt->ltd_tgt->dd_lu_dev.ld_obd == obd) {
993 CERROR("%s: can not find %s: rc = %d\n",
994 lod2obd(lod)->obd_name, param, -EINVAL);
1000 tmp = strstr(param, "=");
1003 struct llog_ctxt *ctxt;
1005 obd = sub_tgt->ltd_tgt->dd_lu_dev.ld_obd;
1006 ctxt = llog_get_context(obd,
1007 LLOG_UPDATELOG_ORIG_CTXT);
1009 rc = llog_setup(env, obd, &obd->obd_olg,
1010 LLOG_UPDATELOG_ORIG_CTXT,
1011 NULL, &llog_common_cat_ops);
1015 llog_ctxt_put(ctxt);
1017 rc = lod_sub_prep_llog(env, lod,
1019 sub_tgt->ltd_index);
1021 sub_tgt->ltd_active = 1;
1023 lod_sub_fini_llog(env, sub_tgt->ltd_tgt,
1025 sub_tgt->ltd_active = 0;
1031 if (strstr(param, PARAM_LOD) != NULL)
1032 count = class_modify_config(lcfg, PARAM_LOD,
1033 &lod->lod_dt_dev.dd_kobj);
1035 count = class_modify_config(lcfg, PARAM_LOV,
1036 &lod->lod_dt_dev.dd_kobj);
1037 rc = count > 0 ? 0 : count;
1040 case LCFG_PRE_CLEANUP: {
1041 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1042 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1043 next = &lod->lod_child->dd_lu_dev;
1044 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1046 CDEBUG(D_HA, "%s: can't process %u: %d\n",
1047 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1049 lod_sub_stop_recovery_threads(env, lod);
1050 lod_fini_distribute_txn(env, lod);
1051 lod_sub_fini_all_llogs(env, lod);
1054 case LCFG_CLEANUP: {
1055 if (lod->lod_md_root) {
1056 dt_object_put(env, &lod->lod_md_root->ldo_obj);
1057 lod->lod_md_root = NULL;
1061 * do cleanup on underlying storage only when
1062 * all OSPs are cleaned up, as they use that OSD as well
1064 lu_dev_del_linkage(dev->ld_site, dev);
1065 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1066 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1067 next = &lod->lod_child->dd_lu_dev;
1068 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1070 CERROR("%s: can't process %u: rc = %d\n",
1071 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1073 rc = obd_disconnect(lod->lod_child_exp);
1075 CERROR("error in disconnect from storage: rc = %d\n",
1080 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
1081 lcfg->lcfg_command);
1091 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
1093 * The method is called once the recovery is complete. This implementation
1094 * distributes the notification to all the known targets.
1096 * see include/lu_object.h for the details
1098 static int lod_recovery_complete(const struct lu_env *env,
1099 struct lu_device *dev)
1101 struct lod_device *lod = lu2lod_dev(dev);
1102 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1103 struct lod_tgt_desc *tgt;
1108 LASSERT(lod->lod_recovery_completed == 0);
1109 lod->lod_recovery_completed = 1;
1111 rc = next->ld_ops->ldo_recovery_complete(env, next);
1113 lod_getref(&lod->lod_ost_descs);
1114 if (lod->lod_ost_descs.ltd_tgts_size > 0) {
1115 lod_foreach_ost(lod, tgt) {
1116 LASSERT(tgt && tgt->ltd_tgt);
1117 next = &tgt->ltd_tgt->dd_lu_dev;
1118 rc = next->ld_ops->ldo_recovery_complete(env, next);
1120 CERROR("%s: can't complete recovery on #%d: rc = %d\n",
1121 lod2obd(lod)->obd_name, tgt->ltd_index,
1125 lod_putref(lod, &lod->lod_ost_descs);
1130 * Init update logs on all sub device
1132 * LOD initialize update logs on all of sub devices. Because the initialization
1133 * process might need FLD lookup, see llog_osd_open()->dt_locate()->...->
1134 * lod_object_init(), this API has to be called after LOD is initialized.
1135 * \param[in] env execution environment
1136 * \param[in] lod lod device
1138 * \retval 0 if update log is initialized successfully.
1139 * \retval negative errno if initialization fails.
1141 static int lod_sub_init_llogs(const struct lu_env *env, struct lod_device *lod)
1143 struct lu_tgt_desc *mdt;
1149 * llog must be setup after LOD is initialized, because llog
1150 * initialization include FLD lookup
1152 LASSERT(lod->lod_initialized);
1154 /* Init the llog in its own stack */
1155 rc = lod_sub_init_llog(env, lod, lod->lod_child);
1159 lod_foreach_mdt(lod, mdt) {
1160 rc = lod_sub_init_llog(env, lod, mdt->ltd_tgt);
1169 * Implementation of lu_device_operations::ldo_prepare() for LOD
1171 * see include/lu_object.h for the details.
1173 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
1174 struct lu_device *cdev)
1176 struct lod_device *lod = lu2lod_dev(cdev);
1177 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1178 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1180 struct dt_object *root;
1181 struct dt_object *dto;
1186 rc = next->ld_ops->ldo_prepare(env, pdev, next);
1188 CERROR("%s: prepare bottom error: rc = %d\n",
1189 lod2obd(lod)->obd_name, rc);
1193 lod->lod_initialized = 1;
1195 rc = dt_root_get(env, lod->lod_child, fid);
1199 root = dt_locate(env, lod->lod_child, fid);
1201 RETURN(PTR_ERR(root));
1203 /* Create update log object */
1204 index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1205 lu_update_log_fid(fid, index);
1207 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1209 lod_update_log_name,
1212 GOTO(out_put, rc = PTR_ERR(dto));
1214 dt_object_put(env, dto);
1216 /* Create update log dir */
1217 lu_update_log_dir_fid(fid, index);
1218 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1220 lod_update_log_dir_name,
1223 GOTO(out_put, rc = PTR_ERR(dto));
1225 dt_object_put(env, dto);
1227 rc = lod_prepare_distribute_txn(env, lod);
1231 rc = lod_sub_init_llogs(env, lod);
1236 dt_object_put(env, root);
1241 const struct lu_device_operations lod_lu_ops = {
1242 .ldo_object_alloc = lod_object_alloc,
1243 .ldo_process_config = lod_process_config,
1244 .ldo_recovery_complete = lod_recovery_complete,
1245 .ldo_prepare = lod_prepare,
1249 * Implementation of dt_device_operations::dt_root_get() for LOD
1251 * see include/dt_object.h for the details.
1253 static int lod_root_get(const struct lu_env *env,
1254 struct dt_device *dev, struct lu_fid *f)
1256 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
1259 static void lod_statfs_sum(struct obd_statfs *sfs,
1260 struct obd_statfs *ost_sfs, int *bs)
1262 while (ost_sfs->os_bsize < *bs) {
1264 sfs->os_bsize >>= 1;
1265 sfs->os_bavail <<= 1;
1266 sfs->os_blocks <<= 1;
1267 sfs->os_bfree <<= 1;
1268 sfs->os_granted <<= 1;
1270 while (ost_sfs->os_bsize > *bs) {
1271 ost_sfs->os_bsize >>= 1;
1272 ost_sfs->os_bavail <<= 1;
1273 ost_sfs->os_blocks <<= 1;
1274 ost_sfs->os_bfree <<= 1;
1275 ost_sfs->os_granted <<= 1;
1277 sfs->os_bavail += ost_sfs->os_bavail;
1278 sfs->os_blocks += ost_sfs->os_blocks;
1279 sfs->os_bfree += ost_sfs->os_bfree;
1280 sfs->os_granted += ost_sfs->os_granted;
1284 * Implementation of dt_device_operations::dt_statfs() for LOD
1286 * see include/dt_object.h for the details.
1288 static int lod_statfs(const struct lu_env *env, struct dt_device *dev,
1289 struct obd_statfs *sfs, struct obd_statfs_info *info)
1291 struct lod_device *lod = dt2lod_dev(dev);
1292 struct lu_tgt_desc *tgt;
1293 struct obd_statfs ost_sfs;
1298 rc = dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
1307 sfs->os_granted = 0;
1309 lod_getref(&lod->lod_mdt_descs);
1310 lod_foreach_mdt(lod, tgt) {
1311 rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
1315 sfs->os_files += ost_sfs.os_files;
1316 sfs->os_ffree += ost_sfs.os_ffree;
1317 lod_statfs_sum(sfs, &ost_sfs, &bs);
1319 lod_putref(lod, &lod->lod_mdt_descs);
1322 * at some point we can check whether DoM is enabled and
1323 * decide how to account MDT space. for simplicity let's
1324 * just fallback to pre-DoM policy if any OST is alive
1326 lod_getref(&lod->lod_ost_descs);
1327 lod_foreach_ost(lod, tgt) {
1328 rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
1330 if (rc || ost_sfs.os_bsize == 0)
1334 * if only MDTs with DoM then report only MDT blocks,
1335 * otherwise show only OST blocks, and DoM is "free"
1340 sfs->os_granted = 0;
1342 ost_files += ost_sfs.os_files;
1343 ost_ffree += ost_sfs.os_ffree;
1344 ost_sfs.os_bavail += ost_sfs.os_granted;
1345 lod_statfs_sum(sfs, &ost_sfs, &bs);
1346 LASSERTF(bs == ost_sfs.os_bsize, "%d != %d\n",
1347 (int)sfs->os_bsize, (int)ost_sfs.os_bsize);
1349 lod_putref(lod, &lod->lod_ost_descs);
1350 sfs->os_state |= OS_STATE_SUM;
1352 /* If we have _some_ OSTs, but don't have as many free objects on the
1353 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1354 * to compensate, so that the "inodes in use" number is correct.
1355 * This should be kept in sync with ll_statfs_internal().
1357 if (ost_files && ost_ffree < sfs->os_ffree) {
1358 sfs->os_files = (sfs->os_files - sfs->os_ffree) + ost_ffree;
1359 sfs->os_ffree = ost_ffree;
1362 /* a single successful statfs should be enough */
1370 * Implementation of dt_device_operations::dt_trans_create() for LOD
1372 * Creates a transaction using local (to this node) OSD.
1374 * see include/dt_object.h for the details.
1376 static struct thandle *lod_trans_create(const struct lu_env *env,
1377 struct dt_device *dt)
1381 th = top_trans_create(env, dt2lod_dev(dt)->lod_child);
1391 * Implementation of dt_device_operations::dt_trans_start() for LOD
1393 * Starts the set of local transactions using the targets involved
1394 * in declare phase. Initial support for the distributed transactions.
1396 * see include/dt_object.h for the details.
1398 static int lod_trans_start(const struct lu_env *env, struct dt_device *dt,
1401 return top_trans_start(env, dt2lod_dev(dt)->lod_child, th);
1404 static int lod_trans_cb_add(struct thandle *th,
1405 struct dt_txn_commit_cb *dcb)
1407 struct top_thandle *top_th = container_of(th, struct top_thandle,
1409 return dt_trans_cb_add(top_th->tt_master_sub_thandle, dcb);
1413 * add noop update to the update records
1415 * Add noop updates to the update records, which is only used in
1418 * \param[in] env execution environment
1419 * \param[in] dt dt device of lod
1420 * \param[in] th thandle
1421 * \param[in] count the count of update records to be added.
1423 * \retval 0 if adding succeeds.
1424 * \retval negative errno if adding fails.
1426 static int lod_add_noop_records(const struct lu_env *env,
1427 struct dt_device *dt, struct thandle *th,
1430 struct top_thandle *top_th;
1431 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1435 top_th = container_of(th, struct top_thandle, tt_super);
1436 if (!top_th->tt_multiple_thandle)
1440 for (i = 0; i < count; i++) {
1441 rc = update_record_pack(noop, th, fid);
1449 * Implementation of dt_device_operations::dt_trans_stop() for LOD
1451 * Stops the set of local transactions using the targets involved
1452 * in declare phase. Initial support for the distributed transactions.
1454 * see include/dt_object.h for the details.
1456 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
1459 if (OBD_FAIL_CHECK(OBD_FAIL_SPLIT_UPDATE_REC)) {
1462 rc = lod_add_noop_records(env, dt, th, 5000);
1466 return top_trans_stop(env, dt2lod_dev(dt)->lod_child, th);
1470 * Implementation of dt_device_operations::dt_conf_get() for LOD
1472 * Currently returns the configuration provided by the local OSD.
1474 * see include/dt_object.h for the details.
1476 static void lod_conf_get(const struct lu_env *env,
1477 const struct dt_device *dev,
1478 struct dt_device_param *param)
1480 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
1484 * Implementation of dt_device_operations::dt_sync() for LOD
1486 * Syncs all known OST targets. Very very expensive and used
1487 * rarely by LFSCK now. Should not be used in general.
1489 * see include/dt_object.h for the details.
1491 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
1493 struct lod_device *lod = dt2lod_dev(dev);
1494 struct lu_tgt_desc *tgt;
1499 lod_getref(&lod->lod_ost_descs);
1500 lod_foreach_ost(lod, tgt) {
1501 if (!tgt->ltd_active)
1503 rc = dt_sync(env, tgt->ltd_tgt);
1505 if (rc != -ENOTCONN) {
1506 CERROR("%s: can't sync ost %u: rc = %d\n",
1507 lod2obd(lod)->obd_name, tgt->ltd_index,
1514 lod_putref(lod, &lod->lod_ost_descs);
1519 lod_getref(&lod->lod_mdt_descs);
1520 lod_foreach_mdt(lod, tgt) {
1521 if (!tgt->ltd_active)
1523 rc = dt_sync(env, tgt->ltd_tgt);
1525 if (rc != -ENOTCONN) {
1526 CERROR("%s: can't sync mdt %u: rc = %d\n",
1527 lod2obd(lod)->obd_name, tgt->ltd_index,
1534 lod_putref(lod, &lod->lod_mdt_descs);
1537 rc = dt_sync(env, lod->lod_child);
1543 * Implementation of dt_device_operations::dt_ro() for LOD
1545 * Turns local OSD read-only, used for the testing only.
1547 * see include/dt_object.h for the details.
1549 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
1551 return dt_ro(env, dt2lod_dev(dev)->lod_child);
1555 * Implementation of dt_device_operations::dt_commit_async() for LOD
1557 * Asks local OSD to commit sooner.
1559 * see include/dt_object.h for the details.
1561 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
1563 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
1566 static const struct dt_device_operations lod_dt_ops = {
1567 .dt_root_get = lod_root_get,
1568 .dt_statfs = lod_statfs,
1569 .dt_trans_create = lod_trans_create,
1570 .dt_trans_start = lod_trans_start,
1571 .dt_trans_stop = lod_trans_stop,
1572 .dt_conf_get = lod_conf_get,
1573 .dt_sync = lod_sync,
1575 .dt_commit_async = lod_commit_async,
1576 .dt_trans_cb_add = lod_trans_cb_add,
1580 * Connect to a local OSD.
1582 * Used to connect to the local OSD at mount. OSD name is taken from the
1583 * configuration command passed. This connection is used to identify LU
1584 * site and pin the OSD from early removal.
1586 * \param[in] env LU environment provided by the caller
1587 * \param[in] lod lod device
1588 * \param[in] cfg configuration command to apply
1590 * \retval 0 on success
1591 * \retval negative negated errno on error
1593 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
1594 struct lustre_cfg *cfg)
1596 struct obd_connect_data *data = NULL;
1597 struct obd_device *obd;
1598 char *nextdev = NULL, *p, *s;
1604 LASSERT(lod->lod_child_exp == NULL);
1607 * compatibility hack: we still use old config logs
1608 * which specify LOV, but we need to learn underlying
1609 * OSD device, which is supposed to be:
1610 * <fsname>-MDTxxxx-osd
1612 * 2.x MGS generates lines like the following:
1613 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
1614 * 1.8 MGS generates lines like the following:
1615 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
1617 * we use "-MDT" to differentiate 2.x from 1.8
1619 p = lustre_cfg_string(cfg, 0);
1620 if (p && strstr(p, "-mdtlov")) {
1621 len = strlen(p) + 6;
1622 OBD_ALLOC(nextdev, len);
1624 GOTO(out, rc = -ENOMEM);
1627 s = strstr(nextdev, "-mdtlov");
1629 CERROR("%s: unable to parse device name: rc = %d\n",
1630 lustre_cfg_string(cfg, 0), -EINVAL);
1631 GOTO(out, rc = -EINVAL);
1634 if (strstr(nextdev, "-MDT")) {
1639 strcpy(s, "-MDT0000-osd");
1642 CERROR("%s: unable to parse device name: rc = %d\n",
1643 lustre_cfg_string(cfg, 0), -EINVAL);
1644 GOTO(out, rc = -EINVAL);
1647 OBD_ALLOC_PTR(data);
1649 GOTO(out, rc = -ENOMEM);
1651 obd = class_name2obd(nextdev);
1653 CERROR("%s: can not locate next device: rc = %d\n",
1654 nextdev, -ENOTCONN);
1655 GOTO(out, rc = -ENOTCONN);
1658 data->ocd_connect_flags = OBD_CONNECT_VERSION;
1659 data->ocd_version = LUSTRE_VERSION_CODE;
1661 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
1664 CERROR("%s: cannot connect to next dev: rc = %d\n",
1669 lod->lod_dt_dev.dd_lu_dev.ld_site =
1670 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
1671 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
1672 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
1678 OBD_FREE(nextdev, len);
1683 * Initialize LOD device at setup.
1685 * Initializes the given LOD device using the original configuration command.
1686 * The function initiates a connection to the local OSD and initializes few
1687 * internal structures like pools, target tables, etc.
1689 * \param[in] env LU environment provided by the caller
1690 * \param[in] lod lod device
1691 * \param[in] ldt not used
1692 * \param[in] cfg configuration command
1694 * \retval 0 on success
1695 * \retval negative negated errno on error
1697 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
1698 struct lu_device_type *ldt, struct lustre_cfg *cfg)
1700 struct dt_device_param ddp;
1701 struct obd_device *obd;
1706 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1709 CERROR("Cannot find obd with name '%s': rc = %d\n",
1710 lustre_cfg_string(cfg, 0), rc);
1714 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1715 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
1716 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
1717 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
1719 rc = lod_connect_to_osd(env, lod, cfg);
1723 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
1724 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
1725 lod->lod_dom_max_stripesize = (1ULL << 20); /* 1Mb as default value */
1727 /* setup obd to be used with old lov code */
1728 rc = lod_pools_init(lod, cfg);
1730 GOTO(out_disconnect, rc);
1732 rc = lod_procfs_init(lod);
1734 GOTO(out_pools, rc);
1736 spin_lock_init(&lod->lod_lock);
1737 spin_lock_init(&lod->lod_connects_lock);
1738 lu_tgt_descs_init(&lod->lod_mdt_descs, true);
1739 lu_tgt_descs_init(&lod->lod_ost_descs, false);
1744 lod_pools_fini(lod);
1746 obd_disconnect(lod->lod_child_exp);
1751 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
1753 * Releases the memory allocated for LOD device.
1755 * see include/lu_object.h for the details.
1757 static struct lu_device *lod_device_free(const struct lu_env *env,
1758 struct lu_device *lu)
1760 struct lod_device *lod = lu2lod_dev(lu);
1761 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1765 if (atomic_read(&lu->ld_ref) > 0 &&
1766 !cfs_hash_is_empty(lu->ld_site->ls_obj_hash)) {
1767 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1768 lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
1770 LASSERTF(atomic_read(&lu->ld_ref) == 0, "lu is %p\n", lu);
1771 dt_device_fini(&lod->lod_dt_dev);
1777 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
1779 * Allocates LOD device and calls the helpers to initialize it.
1781 * see include/lu_object.h for the details.
1783 static struct lu_device *lod_device_alloc(const struct lu_env *env,
1784 struct lu_device_type *type,
1785 struct lustre_cfg *lcfg)
1787 struct lod_device *lod;
1788 struct lu_device *lu_dev;
1792 lu_dev = ERR_PTR(-ENOMEM);
1796 lu_dev = lod2lu_dev(lod);
1797 dt_device_init(&lod->lod_dt_dev, type);
1798 rc = lod_init0(env, lod, type, lcfg);
1800 lod_device_free(env, lu_dev);
1801 lu_dev = ERR_PTR(rc);
1808 static void lod_avoid_guide_fini(struct lod_avoid_guide *lag)
1810 if (lag->lag_oss_avoid_array)
1811 OBD_FREE(lag->lag_oss_avoid_array,
1812 sizeof(u32) * lag->lag_oaa_size);
1813 if (lag->lag_ost_avoid_bitmap)
1814 CFS_FREE_BITMAP(lag->lag_ost_avoid_bitmap);
1818 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
1820 * Releases the internal resources used by LOD device.
1822 * see include/lu_object.h for the details.
1824 static struct lu_device *lod_device_fini(const struct lu_env *env,
1825 struct lu_device *d)
1827 struct lod_device *lod = lu2lod_dev(d);
1832 lod_pools_fini(lod);
1834 lod_procfs_fini(lod);
1836 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs);
1838 CERROR("%s: can not fini ost descriptors: rc = %d\n",
1839 lod2obd(lod)->obd_name, rc);
1841 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs);
1843 CERROR("%s: can not fini mdt descriptors: rc = %d\n",
1844 lod2obd(lod)->obd_name, rc);
1850 * Implementation of obd_ops::o_connect() for LOD
1852 * Used to track all the users of this specific LOD device,
1853 * so the device stays up until the last user disconnected.
1855 * \param[in] env LU environment provided by the caller
1856 * \param[out] exp export the caller will be using to access LOD
1857 * \param[in] obd OBD device representing LOD device
1858 * \param[in] cluuid unique identifier of the caller
1859 * \param[in] data not used
1860 * \param[in] localdata not used
1862 * \retval 0 on success
1863 * \retval negative negated errno on error
1865 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
1866 struct obd_device *obd, struct obd_uuid *cluuid,
1867 struct obd_connect_data *data, void *localdata)
1869 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1870 struct lustre_handle conn;
1875 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1877 rc = class_connect(&conn, obd, cluuid);
1881 *exp = class_conn2export(&conn);
1883 spin_lock(&lod->lod_connects_lock);
1884 lod->lod_connects++;
1885 /* at the moment we expect the only user */
1886 LASSERT(lod->lod_connects == 1);
1887 spin_unlock(&lod->lod_connects_lock);
1894 * Implementation of obd_ops::o_disconnect() for LOD
1896 * When the caller doesn't need to use this LOD instance, it calls
1897 * obd_disconnect() and LOD releases corresponding export/reference count.
1898 * Once all the users gone, LOD device is released.
1900 * \param[in] exp export provided to the caller in obd_connect()
1902 * \retval 0 on success
1903 * \retval negative negated errno on error
1905 static int lod_obd_disconnect(struct obd_export *exp)
1907 struct obd_device *obd = exp->exp_obd;
1908 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1909 int rc, release = 0;
1913 /* Only disconnect the underlying layers on the final disconnect. */
1914 spin_lock(&lod->lod_connects_lock);
1915 lod->lod_connects--;
1916 if (lod->lod_connects != 0) {
1917 /* why should there be more than 1 connect? */
1918 spin_unlock(&lod->lod_connects_lock);
1919 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
1923 spin_unlock(&lod->lod_connects_lock);
1925 /* the last user of lod has gone, let's release the device */
1929 rc = class_disconnect(exp); /* bz 9811 */
1931 if (rc == 0 && release)
1932 class_manual_cleanup(obd);
1936 LU_KEY_INIT(lod, struct lod_thread_info);
1938 static void lod_key_fini(const struct lu_context *ctx,
1939 struct lu_context_key *key, void *data)
1941 struct lod_thread_info *info = data;
1942 struct lod_layout_component *lds =
1943 info->lti_def_striping.lds_def_comp_entries;
1946 * allocated in lod_get_lov_ea
1947 * XXX: this is overload, a tread may have such store but used only
1948 * once. Probably better would be pool of such stores per LOD.
1950 if (info->lti_ea_store) {
1951 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
1952 info->lti_ea_store = NULL;
1953 info->lti_ea_store_size = 0;
1955 lu_buf_free(&info->lti_linkea_buf);
1958 lod_free_def_comp_entries(&info->lti_def_striping);
1960 if (info->lti_comp_size > 0)
1961 OBD_FREE(info->lti_comp_idx,
1962 info->lti_comp_size * sizeof(u32));
1964 lod_avoid_guide_fini(&info->lti_avoid);
1969 /* context key: lod_thread_key */
1970 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
1972 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
1974 static struct lu_device_type_operations lod_device_type_ops = {
1975 .ldto_init = lod_type_init,
1976 .ldto_fini = lod_type_fini,
1978 .ldto_start = lod_type_start,
1979 .ldto_stop = lod_type_stop,
1981 .ldto_device_alloc = lod_device_alloc,
1982 .ldto_device_free = lod_device_free,
1984 .ldto_device_fini = lod_device_fini
1987 static struct lu_device_type lod_device_type = {
1988 .ldt_tags = LU_DEVICE_DT,
1989 .ldt_name = LUSTRE_LOD_NAME,
1990 .ldt_ops = &lod_device_type_ops,
1991 .ldt_ctx_tags = LCT_MD_THREAD,
1995 * Implementation of obd_ops::o_get_info() for LOD
1997 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
1998 * the caller binary status whether LOD has seen connection to any OST target.
1999 * It will also check if the MDT update log context being initialized (if
2002 * \param[in] env LU environment provided by the caller
2003 * \param[in] exp export of the caller
2004 * \param[in] keylen len of the key
2005 * \param[in] key the key
2006 * \param[in] vallen not used
2007 * \param[in] val not used
2009 * \retval 0 if a connection was seen
2010 * \retval -EAGAIN if LOD isn't running yet or no
2011 * connection has been seen yet
2012 * \retval -EINVAL if not supported key is requested
2014 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
2015 u32 keylen, void *key, u32 *vallen, void *val)
2019 if (KEY_IS(KEY_OSP_CONNECTED)) {
2020 struct obd_device *obd = exp->exp_obd;
2021 struct lod_device *d;
2022 struct lod_tgt_desc *tgt;
2025 if (!obd->obd_set_up || obd->obd_stopping)
2028 d = lu2lod_dev(obd->obd_lu_dev);
2029 lod_getref(&d->lod_ost_descs);
2030 lod_foreach_ost(d, tgt) {
2031 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2033 /* one healthy device is enough */
2037 lod_putref(d, &d->lod_ost_descs);
2039 lod_getref(&d->lod_mdt_descs);
2040 lod_foreach_mdt(d, tgt) {
2041 struct llog_ctxt *ctxt;
2043 if (!tgt->ltd_active)
2046 ctxt = llog_get_context(tgt->ltd_tgt->dd_lu_dev.ld_obd,
2047 LLOG_UPDATELOG_ORIG_CTXT);
2049 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2051 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2055 if (!ctxt->loc_handle) {
2056 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2058 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2060 llog_ctxt_put(ctxt);
2063 llog_ctxt_put(ctxt);
2065 lod_putref(d, &d->lod_mdt_descs);
2073 static int lod_obd_set_info_async(const struct lu_env *env,
2074 struct obd_export *exp,
2075 u32 keylen, void *key,
2076 u32 vallen, void *val,
2077 struct ptlrpc_request_set *set)
2079 struct obd_device *obd = class_exp2obd(exp);
2080 struct lod_device *d;
2081 struct lod_tgt_desc *tgt;
2089 set = ptlrpc_prep_set();
2094 d = lu2lod_dev(obd->obd_lu_dev);
2095 lod_getref(&d->lod_ost_descs);
2096 lod_foreach_ost(d, tgt) {
2097 if (!tgt->ltd_active)
2100 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2102 if (rc2 != 0 && rc == 0)
2105 lod_putref(d, &d->lod_ost_descs);
2107 lod_getref(&d->lod_mdt_descs);
2108 lod_foreach_mdt(d, tgt) {
2109 if (!tgt->ltd_active)
2111 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2113 if (rc2 != 0 && rc == 0)
2116 lod_putref(d, &d->lod_mdt_descs);
2120 rc2 = ptlrpc_set_wait(env, set);
2121 if (rc2 == 0 && rc == 0)
2123 ptlrpc_set_destroy(set);
2128 static struct obd_ops lod_obd_device_ops = {
2129 .o_owner = THIS_MODULE,
2130 .o_connect = lod_obd_connect,
2131 .o_disconnect = lod_obd_disconnect,
2132 .o_get_info = lod_obd_get_info,
2133 .o_set_info_async = lod_obd_set_info_async,
2134 .o_pool_new = lod_pool_new,
2135 .o_pool_rem = lod_pool_remove,
2136 .o_pool_add = lod_pool_add,
2137 .o_pool_del = lod_pool_del,
2140 static int __init lod_init(void)
2142 struct obd_type *sym;
2145 rc = lu_kmem_init(lod_caches);
2149 rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
2150 LUSTRE_LOD_NAME, &lod_device_type);
2152 lu_kmem_fini(lod_caches);
2156 /* create "lov" entry for compatibility purposes */
2157 sym = class_add_symlinks(LUSTRE_LOV_NAME, true);
2160 /* does real "lov" already exist ? */
2168 static void __exit lod_exit(void)
2170 struct obd_type *sym = class_search_type(LUSTRE_LOV_NAME);
2172 /* if this was never fully initialized by the lov layer
2173 * then we are responsible for freeing this obd_type
2176 /* final put if we manage this obd type */
2177 if (sym->typ_sym_filter)
2178 kobject_put(&sym->typ_kobj);
2179 /* put reference taken by class_search_type */
2180 kobject_put(&sym->typ_kobj);
2183 class_unregister_type(LUSTRE_LOD_NAME);
2184 lu_kmem_fini(lod_caches);
2187 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2188 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
2189 MODULE_VERSION(LUSTRE_VERSION_STRING);
2190 MODULE_LICENSE("GPL");
2192 module_init(lod_init);
2193 module_exit(lod_exit);