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 task_struct **lrd_task;
274 struct lu_env lrd_env;
275 struct completion *lrd_started;
280 * process update recovery record
282 * Add the update recovery recode to the update recovery list in
283 * lod_recovery_data. Then the recovery thread (target_recovery_thread)
284 * will redo these updates.
286 * \param[in]env execution environment
287 * \param[in]llh log handle of update record
288 * \param[in]rec update record to be replayed
289 * \param[in]data update recovery data which holds the necessary
290 * arguments for recovery (see struct lod_recovery_data)
292 * \retval 0 if the record is processed successfully.
293 * \retval negative errno if the record processing fails.
295 static int lod_process_recovery_updates(const struct lu_env *env,
296 struct llog_handle *llh,
297 struct llog_rec_hdr *rec,
300 struct lod_recovery_data *lrd = data;
301 struct llog_cookie *cookie = &lod_env_info(env)->lti_cookie;
302 struct lu_target *lut;
310 rc = lodname2mdt_index(lod2obd(lrd->lrd_lod)->obd_name, &index);
314 index = lrd->lrd_ltd->ltd_index;
318 llog_update_record_size((struct llog_update_record *)rec)) {
319 CERROR("%s: broken update record! index %u "DFID".%u: rc = %d\n",
320 lod2obd(lrd->lrd_lod)->obd_name, index,
321 PFID(&llh->lgh_id.lgl_oi.oi_fid), rec->lrh_index, -EIO);
325 cookie->lgc_lgl = llh->lgh_id;
326 cookie->lgc_index = rec->lrh_index;
327 cookie->lgc_subsys = LLOG_UPDATELOG_ORIG_CTXT;
329 CDEBUG(D_HA, "%s: process recovery updates "DFID".%u\n",
330 lod2obd(lrd->lrd_lod)->obd_name,
331 PFID(&llh->lgh_id.lgl_oi.oi_fid), rec->lrh_index);
332 lut = lod2lu_dev(lrd->lrd_lod)->ld_site->ls_tgt;
334 if (lut->lut_obd->obd_stopping ||
335 lut->lut_obd->obd_abort_recovery)
338 return insert_update_records_to_replay_list(lut->lut_tdtd,
339 (struct llog_update_record *)rec,
344 * recovery thread for update log
346 * Start recovery thread and prepare the sub llog, then it will retrieve
347 * the update records from the correpondent MDT and do recovery.
349 * \param[in] arg pointer to the recovery data
351 * \retval 0 if recovery succeeds
352 * \retval negative errno if recovery failed.
354 static int lod_sub_recovery_thread(void *arg)
356 struct lod_recovery_data *lrd = arg;
357 struct lod_device *lod = lrd->lrd_lod;
358 struct dt_device *dt;
359 struct llog_ctxt *ctxt = NULL;
360 struct lu_env *env = &lrd->lrd_env;
361 struct lu_target *lut;
362 struct lu_tgt_desc *mdt = NULL;
369 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
370 atomic_inc(&lut->lut_tdtd->tdtd_recovery_threads_count);
374 dt = lrd->lrd_ltd->ltd_tgt;
376 start = ktime_get_real_seconds();
377 complete(lrd->lrd_started);
381 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_TGT_RECOVERY_CONNECT)) &&
383 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_RECOVERY_CONNECT, cfs_fail_val);
386 rc = lod_sub_prep_llog(env, lod, dt, lrd->lrd_idx);
388 if (!rc && !lod->lod_child->dd_rdonly) {
389 /* Process the recovery record */
390 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
391 LLOG_UPDATELOG_ORIG_CTXT);
392 LASSERT(ctxt != NULL);
393 LASSERT(ctxt->loc_handle != NULL);
395 rc = llog_cat_process(env, ctxt->loc_handle,
396 lod_process_recovery_updates, lrd, 0, 0);
400 struct lu_device *top_device;
402 top_device = lod->lod_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
404 * Because the remote target might failover at the same time,
407 if ((rc == -ETIMEDOUT || rc == -EAGAIN || rc == -EIO) &&
408 dt != lod->lod_child &&
409 !top_device->ld_obd->obd_abort_recovery &&
410 !top_device->ld_obd->obd_stopping) {
412 if (ctxt->loc_handle)
418 CDEBUG(D_HA, "%s get update log failed %d, retry\n",
419 dt->dd_lu_dev.ld_obd->obd_name, rc);
423 CERROR("%s get update log failed: rc = %d\n",
424 dt->dd_lu_dev.ld_obd->obd_name, rc);
427 spin_lock(&top_device->ld_obd->obd_dev_lock);
428 if (!top_device->ld_obd->obd_abort_recovery &&
429 !top_device->ld_obd->obd_stopping)
430 top_device->ld_obd->obd_abort_recovery = 1;
431 spin_unlock(&top_device->ld_obd->obd_dev_lock);
437 CDEBUG(D_HA, "%s retrieved update log, duration %lld, retries %d\n",
438 dt->dd_lu_dev.ld_obd->obd_name, ktime_get_real_seconds() - start,
441 spin_lock(&lod->lod_lock);
443 lod->lod_child_got_update_log = 1;
445 lrd->lrd_ltd->ltd_got_update_log = 1;
447 if (!lod->lod_child_got_update_log) {
448 spin_unlock(&lod->lod_lock);
452 lod_foreach_mdt(lod, mdt) {
453 if (!mdt->ltd_got_update_log) {
454 spin_unlock(&lod->lod_lock);
458 lut->lut_tdtd->tdtd_replay_ready = 1;
459 spin_unlock(&lod->lod_lock);
461 CDEBUG(D_HA, "%s got update logs from all MDTs.\n",
462 lut->lut_obd->obd_name);
463 wake_up(&lut->lut_obd->obd_next_transno_waitq);
467 atomic_dec(&lut->lut_tdtd->tdtd_recovery_threads_count);
468 wake_up(&lut->lut_tdtd->tdtd_recovery_threads_waitq);
469 if (xchg(lrd->lrd_task, NULL) == NULL)
470 /* Someone is waiting for us to finish, need
471 * to synchronize cleanly.
473 wait_var_event(lrd, kthread_should_stop());
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 task_struct **thread)
492 struct obd_device *obd;
493 struct llog_ctxt *ctxt;
494 struct task_struct *task = NULL;
498 obd = dt->dd_lu_dev.ld_obd;
499 CDEBUG(D_INFO, "%s: finish sub llog\n", obd->obd_name);
500 /* Wait for recovery thread to complete */
502 task = xchg(thread, NULL);
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)
533 const char *ptr, *tmp;
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 rc = target_name2index(ptr - 7, &index, &tmp);
569 if (rc < 0 || rc & LDD_F_SV_ALL || *tmp != '-') {
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 DECLARE_COMPLETION_ONSTACK(started);
598 struct task_struct **taskp;
599 struct task_struct *task;
600 struct lod_tgt_desc *subtgt = NULL;
607 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &master_index);
615 if (lod->lod_child == dt) {
616 taskp = &lod->lod_child_recovery_task;
617 index = master_index;
619 struct lu_tgt_desc *mdt;
621 lod_foreach_mdt(lod, mdt) {
622 if (mdt->ltd_tgt == dt) {
623 index = mdt->ltd_index;
628 LASSERT(subtgt != NULL);
629 taskp = &subtgt->ltd_recovery_task;
632 CDEBUG(D_INFO, "%s init sub log %s\n", lod2obd(lod)->obd_name,
633 dt->dd_lu_dev.ld_obd->obd_name);
635 lrd->lrd_ltd = subtgt;
636 lrd->lrd_task = taskp;
637 lrd->lrd_idx = index;
638 lrd->lrd_started = &started;
640 obd = dt->dd_lu_dev.ld_obd;
641 obd->obd_lvfs_ctxt.dt = dt;
642 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_UPDATELOG_ORIG_CTXT,
643 NULL, &llog_common_cat_ops);
645 CERROR("%s: cannot setup updatelog llog: rc = %d\n",
650 rc = lu_env_init(&lrd->lrd_env, LCT_LOCAL | LCT_MD_THREAD);
652 CERROR("%s: can't initialize env: rc = %d\n",
653 lod2obd(lod)->obd_name, rc);
657 /* Start the recovery thread */
658 task = kthread_create(lod_sub_recovery_thread, lrd, "lod%04x_rec%04x",
659 master_index, index);
662 CERROR("%s: cannot start recovery thread: rc = %d\n",
664 lu_env_fini(&lrd->lrd_env);
668 wake_up_process(task);
669 wait_for_completion(&started);
673 lod_sub_fini_llog(env, dt, taskp);
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 task_struct *task;
691 struct lu_tgt_desc *mdt;
694 * Stop the update log commit cancel threads and finish master
697 task = xchg(&lod->lod_child_recovery_task, NULL);
701 lod_getref(&lod->lod_mdt_descs);
702 lod_foreach_mdt(lod, mdt) {
703 task = xchg(&mdt->ltd_recovery_task, NULL);
707 lod_putref(lod, &lod->lod_mdt_descs);
711 * finish all sub llog
713 * cleanup all of sub llog ctxt on the LOD.
715 * \param[in] env execution environment
716 * \param[in] lod lod device to do update recovery
718 static void lod_sub_fini_all_llogs(const struct lu_env *env,
719 struct lod_device *lod)
721 struct lu_tgt_desc *mdt;
724 * Stop the update log commit cancel threads and finish master
727 lod_sub_fini_llog(env, lod->lod_child,
728 &lod->lod_child_recovery_task);
729 lod_getref(&lod->lod_mdt_descs);
730 lod_foreach_mdt(lod, mdt)
731 lod_sub_fini_llog(env, mdt->ltd_tgt,
732 &mdt->ltd_recovery_task);
733 lod_putref(lod, &lod->lod_mdt_descs);
736 static char *lod_show_update_logs_retrievers(void *data, int *size, int *count)
738 struct lod_device *lod = (struct lod_device *)data;
739 struct lu_target *lut = lod2lu_dev(lod)->ld_site->ls_tgt;
740 struct lu_tgt_desc *mdt = NULL;
746 *count = atomic_read(&lut->lut_tdtd->tdtd_recovery_threads_count);
752 *size = 5 * *count + 1;
753 OBD_ALLOC(buf, *size);
758 memset(buf, 0, *size);
760 if (!lod->lod_child_got_update_log) {
761 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &i);
762 LASSERTF(rc == 0, "Fail to parse target index: rc = %d\n", rc);
764 rc = scnprintf(buf + len, *size - len, " %04x", i);
771 lod_foreach_mdt(lod, mdt) {
772 if (!mdt->ltd_got_update_log) {
773 rc = scnprintf(buf + len, *size - len, " %04x",
775 if (unlikely(rc <= 0))
787 * Prepare distribute txn
789 * Prepare distribute txn structure for LOD
791 * \param[in] env execution environment
792 * \param[in] lod_device LOD device
794 * \retval 0 if preparation succeeds.
795 * \retval negative errno if preparation fails.
797 static int lod_prepare_distribute_txn(const struct lu_env *env,
798 struct lod_device *lod)
800 struct target_distribute_txn_data *tdtd;
801 struct lu_target *lut;
806 /* Init update recovery data */
811 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
812 tdtd->tdtd_dt = &lod->lod_dt_dev;
813 rc = distribute_txn_init(env, lut, tdtd,
814 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id);
817 CERROR("%s: cannot init distribute txn: rc = %d\n",
818 lod2obd(lod)->obd_name, rc);
823 tdtd->tdtd_show_update_logs_retrievers =
824 lod_show_update_logs_retrievers;
825 tdtd->tdtd_show_retrievers_cbdata = lod;
827 lut->lut_tdtd = tdtd;
833 * Finish distribute txn
835 * Release the resource holding by distribute txn, i.e. stop distribute
838 * \param[in] env execution environment
839 * \param[in] lod lod device
841 static void lod_fini_distribute_txn(const struct lu_env *env,
842 struct lod_device *lod)
844 struct lu_target *lut;
846 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
847 target_recovery_fini(lut->lut_obd);
851 distribute_txn_fini(env, lut->lut_tdtd);
853 OBD_FREE_PTR(lut->lut_tdtd);
854 lut->lut_tdtd = NULL;
858 * Implementation of lu_device_operations::ldo_process_config() for LOD
860 * The method is called by the configuration subsystem during setup,
861 * cleanup and when the configuration changes. The method processes
862 * few specific commands like adding/removing the targets, changing
863 * the runtime parameters.
865 * \param[in] env LU environment provided by the caller
866 * \param[in] dev lod device
867 * \param[in] lcfg configuration command to apply
869 * \retval 0 on success
870 * \retval negative negated errno on error
872 * The examples are below.
874 * Add osc config log:
875 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
876 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
877 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
878 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
879 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
880 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
882 * Add mdc config log:
883 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
884 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
885 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
886 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
887 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
888 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
890 static int lod_process_config(const struct lu_env *env,
891 struct lu_device *dev,
892 struct lustre_cfg *lcfg)
894 struct lod_device *lod = lu2lod_dev(dev);
895 struct lu_device *next = &lod->lod_child->dd_lu_dev;
901 switch (lcfg->lcfg_command) {
902 case LCFG_LOV_DEL_OBD:
903 case LCFG_LOV_ADD_INA:
904 case LCFG_LOV_ADD_OBD:
910 * lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
911 * modify_mdc_tgts add 0:lustre-MDT0001
912 * 1:lustre-MDT0001-mdc0002
915 arg1 = lustre_cfg_string(lcfg, 1);
917 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
918 GOTO(out, rc = -EINVAL);
919 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
920 GOTO(out, rc = -EINVAL);
922 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
925 rc = lodname2mdt_index(lustre_cfg_string(lcfg, 0),
930 rc = lod_add_device(env, lod, arg1, index, gen,
931 mdt_index, LUSTRE_OSC_NAME, 1);
932 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
934 rc = lod_add_device(env, lod, arg1, index, gen,
935 mdt_index, LUSTRE_MDC_NAME, 1);
936 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
937 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
939 rc = lod_add_device(env, lod, arg1, index, gen,
940 mdt_index, LUSTRE_OSC_NAME, 0);
942 rc = lod_del_device(env, lod, &lod->lod_ost_descs,
950 struct obd_device *obd;
955 * Check if it is activate/deactivate mdc
956 * lustre-MDTXXXX-osp-MDTXXXX.active=1
958 param = lustre_cfg_buf(lcfg, 1);
959 if (strstr(param, "osp") && strstr(param, ".active=")) {
960 struct lod_tgt_desc *sub_tgt = NULL;
961 struct lu_tgt_desc *mdt;
965 ptr = strstr(param, ".");
967 obd = class_name2obd(param);
969 CERROR("%s: can not find %s: rc = %d\n",
970 lod2obd(lod)->obd_name, param, -EINVAL);
975 lod_foreach_mdt(lod, mdt) {
976 if (mdt->ltd_tgt->dd_lu_dev.ld_obd == obd) {
983 CERROR("%s: can not find %s: rc = %d\n",
984 lod2obd(lod)->obd_name, param, -EINVAL);
990 tmp = strstr(param, "=");
993 struct llog_ctxt *ctxt;
995 obd = sub_tgt->ltd_tgt->dd_lu_dev.ld_obd;
996 ctxt = llog_get_context(obd,
997 LLOG_UPDATELOG_ORIG_CTXT);
999 rc = llog_setup(env, obd, &obd->obd_olg,
1000 LLOG_UPDATELOG_ORIG_CTXT,
1001 NULL, &llog_common_cat_ops);
1005 llog_ctxt_put(ctxt);
1007 rc = lod_sub_prep_llog(env, lod,
1009 sub_tgt->ltd_index);
1011 sub_tgt->ltd_active = 1;
1013 lod_sub_fini_llog(env, sub_tgt->ltd_tgt,
1015 sub_tgt->ltd_active = 0;
1021 if (strstr(param, PARAM_LOD) != NULL)
1022 count = class_modify_config(lcfg, PARAM_LOD,
1023 &lod->lod_dt_dev.dd_kobj);
1025 count = class_modify_config(lcfg, PARAM_LOV,
1026 &lod->lod_dt_dev.dd_kobj);
1027 rc = count > 0 ? 0 : count;
1030 case LCFG_PRE_CLEANUP: {
1031 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1032 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1033 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_RECOVERY_CONNECT, cfs_fail_val * 2);
1034 next = &lod->lod_child->dd_lu_dev;
1035 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1037 CDEBUG(D_HA, "%s: can't process %u: %d\n",
1038 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1040 lod_sub_stop_recovery_threads(env, lod);
1041 lod_fini_distribute_txn(env, lod);
1042 lod_sub_fini_all_llogs(env, lod);
1045 case LCFG_CLEANUP: {
1046 if (lod->lod_md_root) {
1047 dt_object_put(env, &lod->lod_md_root->ldo_obj);
1048 lod->lod_md_root = NULL;
1052 * do cleanup on underlying storage only when
1053 * all OSPs are cleaned up, as they use that OSD as well
1055 lu_dev_del_linkage(dev->ld_site, dev);
1056 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1057 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1058 next = &lod->lod_child->dd_lu_dev;
1059 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1061 CERROR("%s: can't process %u: rc = %d\n",
1062 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1064 rc = obd_disconnect(lod->lod_child_exp);
1066 CERROR("error in disconnect from storage: rc = %d\n",
1071 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
1072 lcfg->lcfg_command);
1082 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
1084 * The method is called once the recovery is complete. This implementation
1085 * distributes the notification to all the known targets.
1087 * see include/lu_object.h for the details
1089 static int lod_recovery_complete(const struct lu_env *env,
1090 struct lu_device *dev)
1092 struct lod_device *lod = lu2lod_dev(dev);
1093 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1094 struct lod_tgt_desc *tgt;
1099 LASSERT(lod->lod_recovery_completed == 0);
1100 lod->lod_recovery_completed = 1;
1102 rc = next->ld_ops->ldo_recovery_complete(env, next);
1104 lod_getref(&lod->lod_ost_descs);
1105 if (lod->lod_ost_descs.ltd_tgts_size > 0) {
1106 lod_foreach_ost(lod, tgt) {
1107 LASSERT(tgt && tgt->ltd_tgt);
1108 next = &tgt->ltd_tgt->dd_lu_dev;
1109 rc = next->ld_ops->ldo_recovery_complete(env, next);
1111 CERROR("%s: can't complete recovery on #%d: rc = %d\n",
1112 lod2obd(lod)->obd_name, tgt->ltd_index,
1116 lod_putref(lod, &lod->lod_ost_descs);
1121 * Init update logs on all sub device
1123 * LOD initialize update logs on all of sub devices. Because the initialization
1124 * process might need FLD lookup, see llog_osd_open()->dt_locate()->...->
1125 * lod_object_init(), this API has to be called after LOD is initialized.
1126 * \param[in] env execution environment
1127 * \param[in] lod lod device
1129 * \retval 0 if update log is initialized successfully.
1130 * \retval negative errno if initialization fails.
1132 static int lod_sub_init_llogs(const struct lu_env *env, struct lod_device *lod)
1134 struct lu_tgt_desc *mdt;
1140 * llog must be setup after LOD is initialized, because llog
1141 * initialization include FLD lookup
1143 LASSERT(lod->lod_initialized);
1145 /* Init the llog in its own stack */
1146 rc = lod_sub_init_llog(env, lod, lod->lod_child);
1150 lod_foreach_mdt(lod, mdt) {
1151 rc = lod_sub_init_llog(env, lod, mdt->ltd_tgt);
1160 * Implementation of lu_device_operations::ldo_prepare() for LOD
1162 * see include/lu_object.h for the details.
1164 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
1165 struct lu_device *cdev)
1167 struct lod_device *lod = lu2lod_dev(cdev);
1168 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1169 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1171 struct dt_object *root;
1172 struct dt_object *dto;
1177 rc = next->ld_ops->ldo_prepare(env, pdev, next);
1179 CERROR("%s: prepare bottom error: rc = %d\n",
1180 lod2obd(lod)->obd_name, rc);
1184 lod->lod_initialized = 1;
1186 rc = dt_root_get(env, lod->lod_child, fid);
1190 root = dt_locate(env, lod->lod_child, fid);
1192 RETURN(PTR_ERR(root));
1194 /* Create update log object */
1195 index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1196 lu_update_log_fid(fid, index);
1198 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1200 lod_update_log_name,
1203 GOTO(out_put, rc = PTR_ERR(dto));
1205 dt_object_put(env, dto);
1207 /* Create update log dir */
1208 lu_update_log_dir_fid(fid, index);
1209 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1211 lod_update_log_dir_name,
1214 GOTO(out_put, rc = PTR_ERR(dto));
1216 dt_object_put(env, dto);
1218 rc = lod_prepare_distribute_txn(env, lod);
1222 rc = lod_sub_init_llogs(env, lod);
1227 dt_object_put(env, root);
1232 const struct lu_device_operations lod_lu_ops = {
1233 .ldo_object_alloc = lod_object_alloc,
1234 .ldo_process_config = lod_process_config,
1235 .ldo_recovery_complete = lod_recovery_complete,
1236 .ldo_prepare = lod_prepare,
1240 * Implementation of dt_device_operations::dt_root_get() for LOD
1242 * see include/dt_object.h for the details.
1244 static int lod_root_get(const struct lu_env *env,
1245 struct dt_device *dev, struct lu_fid *f)
1247 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
1250 static void lod_statfs_sum(struct obd_statfs *sfs,
1251 struct obd_statfs *ost_sfs, int *bs)
1253 while (ost_sfs->os_bsize < *bs) {
1255 sfs->os_bsize >>= 1;
1256 sfs->os_bavail <<= 1;
1257 sfs->os_blocks <<= 1;
1258 sfs->os_bfree <<= 1;
1259 sfs->os_granted <<= 1;
1261 while (ost_sfs->os_bsize > *bs) {
1262 ost_sfs->os_bsize >>= 1;
1263 ost_sfs->os_bavail <<= 1;
1264 ost_sfs->os_blocks <<= 1;
1265 ost_sfs->os_bfree <<= 1;
1266 ost_sfs->os_granted <<= 1;
1268 sfs->os_bavail += ost_sfs->os_bavail;
1269 sfs->os_blocks += ost_sfs->os_blocks;
1270 sfs->os_bfree += ost_sfs->os_bfree;
1271 sfs->os_granted += ost_sfs->os_granted;
1275 * Implementation of dt_device_operations::dt_statfs() for LOD
1277 * see include/dt_object.h for the details.
1279 static int lod_statfs(const struct lu_env *env, struct dt_device *dev,
1280 struct obd_statfs *sfs, struct obd_statfs_info *info)
1282 struct lod_device *lod = dt2lod_dev(dev);
1283 struct lu_tgt_desc *tgt;
1284 struct obd_statfs ost_sfs;
1289 rc = dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
1298 sfs->os_granted = 0;
1300 lod_getref(&lod->lod_mdt_descs);
1301 lod_foreach_mdt(lod, tgt) {
1302 rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
1306 sfs->os_files += ost_sfs.os_files;
1307 sfs->os_ffree += ost_sfs.os_ffree;
1308 lod_statfs_sum(sfs, &ost_sfs, &bs);
1310 lod_putref(lod, &lod->lod_mdt_descs);
1313 * at some point we can check whether DoM is enabled and
1314 * decide how to account MDT space. for simplicity let's
1315 * just fallback to pre-DoM policy if any OST is alive
1317 lod_getref(&lod->lod_ost_descs);
1318 lod_foreach_ost(lod, tgt) {
1319 rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
1321 if (rc || ost_sfs.os_bsize == 0)
1325 * if only MDTs with DoM then report only MDT blocks,
1326 * otherwise show only OST blocks, and DoM is "free"
1331 sfs->os_granted = 0;
1333 ost_files += ost_sfs.os_files;
1334 ost_ffree += ost_sfs.os_ffree;
1335 ost_sfs.os_bavail += ost_sfs.os_granted;
1336 lod_statfs_sum(sfs, &ost_sfs, &bs);
1337 LASSERTF(bs == ost_sfs.os_bsize, "%d != %d\n",
1338 (int)sfs->os_bsize, (int)ost_sfs.os_bsize);
1340 lod_putref(lod, &lod->lod_ost_descs);
1341 sfs->os_state |= OS_STATE_SUM;
1343 /* If we have _some_ OSTs, but don't have as many free objects on the
1344 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1345 * to compensate, so that the "inodes in use" number is correct.
1346 * This should be kept in sync with ll_statfs_internal().
1348 if (ost_files && ost_ffree < sfs->os_ffree) {
1349 sfs->os_files = (sfs->os_files - sfs->os_ffree) + ost_ffree;
1350 sfs->os_ffree = ost_ffree;
1353 /* a single successful statfs should be enough */
1361 * Implementation of dt_device_operations::dt_trans_create() for LOD
1363 * Creates a transaction using local (to this node) OSD.
1365 * see include/dt_object.h for the details.
1367 static struct thandle *lod_trans_create(const struct lu_env *env,
1368 struct dt_device *dt)
1372 th = top_trans_create(env, dt2lod_dev(dt)->lod_child);
1382 * Implementation of dt_device_operations::dt_trans_start() for LOD
1384 * Starts the set of local transactions using the targets involved
1385 * in declare phase. Initial support for the distributed transactions.
1387 * see include/dt_object.h for the details.
1389 static int lod_trans_start(const struct lu_env *env, struct dt_device *dt,
1392 return top_trans_start(env, dt2lod_dev(dt)->lod_child, th);
1395 static int lod_trans_cb_add(struct thandle *th,
1396 struct dt_txn_commit_cb *dcb)
1398 struct top_thandle *top_th = container_of(th, struct top_thandle,
1400 return dt_trans_cb_add(top_th->tt_master_sub_thandle, dcb);
1404 * add noop update to the update records
1406 * Add noop updates to the update records, which is only used in
1409 * \param[in] env execution environment
1410 * \param[in] dt dt device of lod
1411 * \param[in] th thandle
1412 * \param[in] count the count of update records to be added.
1414 * \retval 0 if adding succeeds.
1415 * \retval negative errno if adding fails.
1417 static int lod_add_noop_records(const struct lu_env *env,
1418 struct dt_device *dt, struct thandle *th,
1421 struct top_thandle *top_th;
1422 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1426 top_th = container_of(th, struct top_thandle, tt_super);
1427 if (!top_th->tt_multiple_thandle)
1431 for (i = 0; i < count; i++) {
1432 rc = update_record_pack(noop, th, fid);
1440 * Implementation of dt_device_operations::dt_trans_stop() for LOD
1442 * Stops the set of local transactions using the targets involved
1443 * in declare phase. Initial support for the distributed transactions.
1445 * see include/dt_object.h for the details.
1447 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
1450 if (OBD_FAIL_CHECK(OBD_FAIL_SPLIT_UPDATE_REC)) {
1453 rc = lod_add_noop_records(env, dt, th, 5000);
1457 return top_trans_stop(env, dt2lod_dev(dt)->lod_child, th);
1461 * Implementation of dt_device_operations::dt_conf_get() for LOD
1463 * Currently returns the configuration provided by the local OSD.
1465 * see include/dt_object.h for the details.
1467 static void lod_conf_get(const struct lu_env *env,
1468 const struct dt_device *dev,
1469 struct dt_device_param *param)
1471 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
1475 * Implementation of dt_device_operations::dt_sync() for LOD
1477 * Syncs all known OST targets. Very very expensive and used
1478 * rarely by LFSCK now. Should not be used in general.
1480 * see include/dt_object.h for the details.
1482 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
1484 struct lod_device *lod = dt2lod_dev(dev);
1485 struct lu_tgt_desc *tgt;
1490 lod_getref(&lod->lod_ost_descs);
1491 lod_foreach_ost(lod, tgt) {
1492 if (!tgt->ltd_active)
1494 rc = dt_sync(env, tgt->ltd_tgt);
1496 if (rc != -ENOTCONN) {
1497 CERROR("%s: can't sync ost %u: rc = %d\n",
1498 lod2obd(lod)->obd_name, tgt->ltd_index,
1505 lod_putref(lod, &lod->lod_ost_descs);
1510 lod_getref(&lod->lod_mdt_descs);
1511 lod_foreach_mdt(lod, tgt) {
1512 if (!tgt->ltd_active)
1514 rc = dt_sync(env, tgt->ltd_tgt);
1516 if (rc != -ENOTCONN) {
1517 CERROR("%s: can't sync mdt %u: rc = %d\n",
1518 lod2obd(lod)->obd_name, tgt->ltd_index,
1525 lod_putref(lod, &lod->lod_mdt_descs);
1528 rc = dt_sync(env, lod->lod_child);
1534 * Implementation of dt_device_operations::dt_ro() for LOD
1536 * Turns local OSD read-only, used for the testing only.
1538 * see include/dt_object.h for the details.
1540 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
1542 return dt_ro(env, dt2lod_dev(dev)->lod_child);
1546 * Implementation of dt_device_operations::dt_commit_async() for LOD
1548 * Asks local OSD to commit sooner.
1550 * see include/dt_object.h for the details.
1552 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
1554 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
1557 static const struct dt_device_operations lod_dt_ops = {
1558 .dt_root_get = lod_root_get,
1559 .dt_statfs = lod_statfs,
1560 .dt_trans_create = lod_trans_create,
1561 .dt_trans_start = lod_trans_start,
1562 .dt_trans_stop = lod_trans_stop,
1563 .dt_conf_get = lod_conf_get,
1564 .dt_sync = lod_sync,
1566 .dt_commit_async = lod_commit_async,
1567 .dt_trans_cb_add = lod_trans_cb_add,
1571 * Connect to a local OSD.
1573 * Used to connect to the local OSD at mount. OSD name is taken from the
1574 * configuration command passed. This connection is used to identify LU
1575 * site and pin the OSD from early removal.
1577 * \param[in] env LU environment provided by the caller
1578 * \param[in] lod lod device
1579 * \param[in] cfg configuration command to apply
1581 * \retval 0 on success
1582 * \retval negative negated errno on error
1584 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
1585 struct lustre_cfg *cfg)
1587 struct obd_connect_data *data = NULL;
1588 struct obd_device *obd;
1589 char *nextdev = NULL, *p, *s;
1595 LASSERT(lod->lod_child_exp == NULL);
1598 * compatibility hack: we still use old config logs
1599 * which specify LOV, but we need to learn underlying
1600 * OSD device, which is supposed to be:
1601 * <fsname>-MDTxxxx-osd
1603 * 2.x MGS generates lines like the following:
1604 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
1605 * 1.8 MGS generates lines like the following:
1606 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
1608 * we use "-MDT" to differentiate 2.x from 1.8
1610 p = lustre_cfg_string(cfg, 0);
1611 if (p && strstr(p, "-mdtlov")) {
1612 len = strlen(p) + 6;
1613 OBD_ALLOC(nextdev, len);
1615 GOTO(out, rc = -ENOMEM);
1618 s = strstr(nextdev, "-mdtlov");
1620 CERROR("%s: unable to parse device name: rc = %d\n",
1621 lustre_cfg_string(cfg, 0), -EINVAL);
1622 GOTO(out, rc = -EINVAL);
1625 if (strstr(nextdev, "-MDT")) {
1630 strcpy(s, "-MDT0000-osd");
1633 CERROR("%s: unable to parse device name: rc = %d\n",
1634 lustre_cfg_string(cfg, 0), -EINVAL);
1635 GOTO(out, rc = -EINVAL);
1638 OBD_ALLOC_PTR(data);
1640 GOTO(out, rc = -ENOMEM);
1642 obd = class_name2obd(nextdev);
1644 CERROR("%s: can not locate next device: rc = %d\n",
1645 nextdev, -ENOTCONN);
1646 GOTO(out, rc = -ENOTCONN);
1649 data->ocd_connect_flags = OBD_CONNECT_VERSION;
1650 data->ocd_version = LUSTRE_VERSION_CODE;
1652 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
1655 CERROR("%s: cannot connect to next dev: rc = %d\n",
1660 lod->lod_dt_dev.dd_lu_dev.ld_site =
1661 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
1662 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
1663 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
1669 OBD_FREE(nextdev, len);
1674 * Initialize LOD device at setup.
1676 * Initializes the given LOD device using the original configuration command.
1677 * The function initiates a connection to the local OSD and initializes few
1678 * internal structures like pools, target tables, etc.
1680 * \param[in] env LU environment provided by the caller
1681 * \param[in] lod lod device
1682 * \param[in] ldt not used
1683 * \param[in] cfg configuration command
1685 * \retval 0 on success
1686 * \retval negative negated errno on error
1688 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
1689 struct lu_device_type *ldt, struct lustre_cfg *cfg)
1691 struct dt_device_param ddp;
1692 struct obd_device *obd;
1697 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1700 CERROR("Cannot find obd with name '%s': rc = %d\n",
1701 lustre_cfg_string(cfg, 0), rc);
1705 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1706 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
1707 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
1708 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
1710 rc = lod_connect_to_osd(env, lod, cfg);
1714 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
1715 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
1716 lod->lod_dom_max_stripesize = (1ULL << 20); /* 1Mb as default value */
1718 /* setup obd to be used with old lov code */
1719 rc = lod_pools_init(lod, cfg);
1721 GOTO(out_disconnect, rc);
1723 rc = lod_procfs_init(lod);
1725 GOTO(out_pools, rc);
1727 spin_lock_init(&lod->lod_lock);
1728 spin_lock_init(&lod->lod_connects_lock);
1729 lu_tgt_descs_init(&lod->lod_mdt_descs, true);
1730 lu_tgt_descs_init(&lod->lod_ost_descs, false);
1735 lod_pools_fini(lod);
1737 obd_disconnect(lod->lod_child_exp);
1742 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
1744 * Releases the memory allocated for LOD device.
1746 * see include/lu_object.h for the details.
1748 static struct lu_device *lod_device_free(const struct lu_env *env,
1749 struct lu_device *lu)
1751 struct lod_device *lod = lu2lod_dev(lu);
1752 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1756 if (atomic_read(&lu->ld_ref) > 0 &&
1757 !cfs_hash_is_empty(lu->ld_site->ls_obj_hash)) {
1758 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
1759 lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
1761 LASSERTF(atomic_read(&lu->ld_ref) == 0, "lu is %p\n", lu);
1762 dt_device_fini(&lod->lod_dt_dev);
1768 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
1770 * Allocates LOD device and calls the helpers to initialize it.
1772 * see include/lu_object.h for the details.
1774 static struct lu_device *lod_device_alloc(const struct lu_env *env,
1775 struct lu_device_type *type,
1776 struct lustre_cfg *lcfg)
1778 struct lod_device *lod;
1779 struct lu_device *lu_dev;
1783 lu_dev = ERR_PTR(-ENOMEM);
1787 lu_dev = lod2lu_dev(lod);
1788 dt_device_init(&lod->lod_dt_dev, type);
1789 rc = lod_init0(env, lod, type, lcfg);
1791 lod_device_free(env, lu_dev);
1792 lu_dev = ERR_PTR(rc);
1799 static void lod_avoid_guide_fini(struct lod_avoid_guide *lag)
1801 if (lag->lag_oss_avoid_array)
1802 OBD_FREE(lag->lag_oss_avoid_array,
1803 sizeof(u32) * lag->lag_oaa_size);
1804 if (lag->lag_ost_avoid_bitmap)
1805 CFS_FREE_BITMAP(lag->lag_ost_avoid_bitmap);
1809 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
1811 * Releases the internal resources used by LOD device.
1813 * see include/lu_object.h for the details.
1815 static struct lu_device *lod_device_fini(const struct lu_env *env,
1816 struct lu_device *d)
1818 struct lod_device *lod = lu2lod_dev(d);
1823 lod_pools_fini(lod);
1825 lod_procfs_fini(lod);
1827 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs);
1829 CERROR("%s: can not fini ost descriptors: rc = %d\n",
1830 lod2obd(lod)->obd_name, rc);
1832 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs);
1834 CERROR("%s: can not fini mdt descriptors: rc = %d\n",
1835 lod2obd(lod)->obd_name, rc);
1841 * Implementation of obd_ops::o_connect() for LOD
1843 * Used to track all the users of this specific LOD device,
1844 * so the device stays up until the last user disconnected.
1846 * \param[in] env LU environment provided by the caller
1847 * \param[out] exp export the caller will be using to access LOD
1848 * \param[in] obd OBD device representing LOD device
1849 * \param[in] cluuid unique identifier of the caller
1850 * \param[in] data not used
1851 * \param[in] localdata not used
1853 * \retval 0 on success
1854 * \retval negative negated errno on error
1856 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
1857 struct obd_device *obd, struct obd_uuid *cluuid,
1858 struct obd_connect_data *data, void *localdata)
1860 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1861 struct lustre_handle conn;
1866 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1868 rc = class_connect(&conn, obd, cluuid);
1872 *exp = class_conn2export(&conn);
1874 spin_lock(&lod->lod_connects_lock);
1875 lod->lod_connects++;
1876 /* at the moment we expect the only user */
1877 LASSERT(lod->lod_connects == 1);
1878 spin_unlock(&lod->lod_connects_lock);
1885 * Implementation of obd_ops::o_disconnect() for LOD
1887 * When the caller doesn't need to use this LOD instance, it calls
1888 * obd_disconnect() and LOD releases corresponding export/reference count.
1889 * Once all the users gone, LOD device is released.
1891 * \param[in] exp export provided to the caller in obd_connect()
1893 * \retval 0 on success
1894 * \retval negative negated errno on error
1896 static int lod_obd_disconnect(struct obd_export *exp)
1898 struct obd_device *obd = exp->exp_obd;
1899 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1900 int rc, release = 0;
1904 /* Only disconnect the underlying layers on the final disconnect. */
1905 spin_lock(&lod->lod_connects_lock);
1906 lod->lod_connects--;
1907 if (lod->lod_connects != 0) {
1908 /* why should there be more than 1 connect? */
1909 spin_unlock(&lod->lod_connects_lock);
1910 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
1914 spin_unlock(&lod->lod_connects_lock);
1916 /* the last user of lod has gone, let's release the device */
1920 rc = class_disconnect(exp); /* bz 9811 */
1922 if (rc == 0 && release)
1923 class_manual_cleanup(obd);
1927 LU_KEY_INIT(lod, struct lod_thread_info);
1929 static void lod_key_fini(const struct lu_context *ctx,
1930 struct lu_context_key *key, void *data)
1932 struct lod_thread_info *info = data;
1933 struct lod_layout_component *lds =
1934 info->lti_def_striping.lds_def_comp_entries;
1937 * allocated in lod_get_lov_ea
1938 * XXX: this is overload, a tread may have such store but used only
1939 * once. Probably better would be pool of such stores per LOD.
1941 if (info->lti_ea_store) {
1942 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
1943 info->lti_ea_store = NULL;
1944 info->lti_ea_store_size = 0;
1946 lu_buf_free(&info->lti_linkea_buf);
1949 lod_free_def_comp_entries(&info->lti_def_striping);
1951 if (info->lti_comp_size > 0)
1952 OBD_FREE(info->lti_comp_idx,
1953 info->lti_comp_size * sizeof(u32));
1955 lod_avoid_guide_fini(&info->lti_avoid);
1960 /* context key: lod_thread_key */
1961 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
1963 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
1965 static struct lu_device_type_operations lod_device_type_ops = {
1966 .ldto_init = lod_type_init,
1967 .ldto_fini = lod_type_fini,
1969 .ldto_start = lod_type_start,
1970 .ldto_stop = lod_type_stop,
1972 .ldto_device_alloc = lod_device_alloc,
1973 .ldto_device_free = lod_device_free,
1975 .ldto_device_fini = lod_device_fini
1978 static struct lu_device_type lod_device_type = {
1979 .ldt_tags = LU_DEVICE_DT,
1980 .ldt_name = LUSTRE_LOD_NAME,
1981 .ldt_ops = &lod_device_type_ops,
1982 .ldt_ctx_tags = LCT_MD_THREAD,
1986 * Implementation of obd_ops::o_get_info() for LOD
1988 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
1989 * the caller binary status whether LOD has seen connection to any OST target.
1990 * It will also check if the MDT update log context being initialized (if
1993 * \param[in] env LU environment provided by the caller
1994 * \param[in] exp export of the caller
1995 * \param[in] keylen len of the key
1996 * \param[in] key the key
1997 * \param[in] vallen not used
1998 * \param[in] val not used
2000 * \retval 0 if a connection was seen
2001 * \retval -EAGAIN if LOD isn't running yet or no
2002 * connection has been seen yet
2003 * \retval -EINVAL if not supported key is requested
2005 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
2006 u32 keylen, void *key, u32 *vallen, void *val)
2010 if (KEY_IS(KEY_OSP_CONNECTED)) {
2011 struct obd_device *obd = exp->exp_obd;
2012 struct lod_device *d;
2013 struct lod_tgt_desc *tgt;
2016 if (!obd->obd_set_up || obd->obd_stopping)
2019 d = lu2lod_dev(obd->obd_lu_dev);
2020 lod_getref(&d->lod_ost_descs);
2021 lod_foreach_ost(d, tgt) {
2022 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2024 /* one healthy device is enough */
2028 lod_putref(d, &d->lod_ost_descs);
2030 lod_getref(&d->lod_mdt_descs);
2031 lod_foreach_mdt(d, tgt) {
2032 struct llog_ctxt *ctxt;
2034 if (!tgt->ltd_active)
2037 ctxt = llog_get_context(tgt->ltd_tgt->dd_lu_dev.ld_obd,
2038 LLOG_UPDATELOG_ORIG_CTXT);
2040 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2042 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2046 if (!ctxt->loc_handle) {
2047 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2049 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2051 llog_ctxt_put(ctxt);
2054 llog_ctxt_put(ctxt);
2056 lod_putref(d, &d->lod_mdt_descs);
2064 static int lod_obd_set_info_async(const struct lu_env *env,
2065 struct obd_export *exp,
2066 u32 keylen, void *key,
2067 u32 vallen, void *val,
2068 struct ptlrpc_request_set *set)
2070 struct obd_device *obd = class_exp2obd(exp);
2071 struct lod_device *d;
2072 struct lod_tgt_desc *tgt;
2080 set = ptlrpc_prep_set();
2085 d = lu2lod_dev(obd->obd_lu_dev);
2086 lod_getref(&d->lod_ost_descs);
2087 lod_foreach_ost(d, tgt) {
2088 if (!tgt->ltd_active)
2091 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2093 if (rc2 != 0 && rc == 0)
2096 lod_putref(d, &d->lod_ost_descs);
2098 lod_getref(&d->lod_mdt_descs);
2099 lod_foreach_mdt(d, tgt) {
2100 if (!tgt->ltd_active)
2102 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2104 if (rc2 != 0 && rc == 0)
2107 lod_putref(d, &d->lod_mdt_descs);
2111 rc2 = ptlrpc_set_wait(env, set);
2112 if (rc2 == 0 && rc == 0)
2114 ptlrpc_set_destroy(set);
2119 static const struct obd_ops lod_obd_device_ops = {
2120 .o_owner = THIS_MODULE,
2121 .o_connect = lod_obd_connect,
2122 .o_disconnect = lod_obd_disconnect,
2123 .o_get_info = lod_obd_get_info,
2124 .o_set_info_async = lod_obd_set_info_async,
2125 .o_pool_new = lod_pool_new,
2126 .o_pool_rem = lod_pool_remove,
2127 .o_pool_add = lod_pool_add,
2128 .o_pool_del = lod_pool_del,
2131 static int __init lod_init(void)
2133 struct obd_type *sym;
2136 rc = lu_kmem_init(lod_caches);
2140 rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
2141 LUSTRE_LOD_NAME, &lod_device_type);
2143 lu_kmem_fini(lod_caches);
2147 /* create "lov" entry for compatibility purposes */
2148 sym = class_add_symlinks(LUSTRE_LOV_NAME, true);
2151 /* does real "lov" already exist ? */
2159 static void __exit lod_exit(void)
2161 struct obd_type *sym = class_search_type(LUSTRE_LOV_NAME);
2163 /* if this was never fully initialized by the lov layer
2164 * then we are responsible for freeing this obd_type
2167 /* final put if we manage this obd type */
2168 if (sym->typ_sym_filter)
2169 kobject_put(&sym->typ_kobj);
2170 /* put reference taken by class_search_type */
2171 kobject_put(&sym->typ_kobj);
2174 class_unregister_type(LUSTRE_LOD_NAME);
2175 lu_kmem_fini(lod_caches);
2178 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2179 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
2180 MODULE_VERSION(LUSTRE_VERSION_STRING);
2181 MODULE_LICENSE("GPL");
2183 module_init(lod_init);
2184 module_exit(lod_exit);