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/
31 * lustre/lod/lod_dev.c
33 * Lustre Logical Object Device
35 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
36 * Author: Mikhail Pershin <mike.pershin@intel.com>
39 * The Logical Object Device (LOD) layer manages access to striped
40 * objects (both regular files and directories). It implements the DT
41 * device and object APIs and is responsible for creating, storing,
42 * and loading striping information as an extended attribute of the
43 * underlying OSD object. LOD is the server side analog of the LOV and
44 * LMV layers on the client side.
46 * Metadata LU object stack (layers of the same compound LU object,
47 * all have the same FID):
57 * During LOD object initialization the localness or remoteness of the
58 * object FID dictates the choice between OSD and OSP.
60 * An LOD object (file or directory) with N stripes (each has a
67 * S0 S1 S2 S3 S(N-1) OS[DP] objects, seen as DT objects by LOD
69 * When upper layers must access an object's stripes (which are
70 * themselves OST or MDT LU objects) LOD finds these objects by their
71 * FIDs and stores them as an array of DT object pointers on the
72 * object. Declarations and operations on LOD objects are received by
73 * LOD (as DT object operations) and performed on the underlying
74 * OS[DP] object and (as needed) on the stripes. From the perspective
75 * of LOD, a stripe-less file (created by mknod() or open with
76 * O_LOV_DELAY_CREATE) is an object which does not yet have stripes,
77 * while a non-striped directory (created by mkdir()) is an object
78 * which will never have stripes.
80 * The LOD layer also implements a small subset of the OBD device API
81 * to support MDT stack initialization and finalization (an MDD device
82 * connects and disconnects itself to and from the underlying LOD
83 * device), and pool management. In turn LOD uses the OBD device API
84 * to connect it self to the underlying OSD, and to connect itself to
85 * OSP devices representing the MDTs and OSTs that bear the stripes of
89 #define DEBUG_SUBSYSTEM S_MDS
91 #include <linux/kthread.h>
92 #include <obd_class.h>
93 #include <md_object.h>
94 #include <lustre_fid.h>
95 #include <uapi/linux/lustre/lustre_param.h>
96 #include <lustre_update.h>
97 #include <lustre_log.h>
98 #include <lustre_lmv.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;
278 static bool lod_recovery_abort(struct obd_device *top)
280 return (top->obd_stopping || top->obd_abort_recovery ||
281 top->obd_abort_recov_mdt);
285 * process update recovery record
287 * Add the update recovery recode to the update recovery list in
288 * lod_recovery_data. Then the recovery thread (target_recovery_thread)
289 * will redo these updates.
291 * \param[in]env execution environment
292 * \param[in]llh log handle of update record
293 * \param[in]rec update record to be replayed
294 * \param[in]data update recovery data which holds the necessary
295 * arguments for recovery (see struct lod_recovery_data)
297 * \retval 0 if the record is processed successfully.
298 * \retval negative errno if the record processing fails.
300 static int lod_process_recovery_updates(const struct lu_env *env,
301 struct llog_handle *llh,
302 struct llog_rec_hdr *rec,
305 struct lod_recovery_data *lrd = data;
306 struct llog_cookie *cookie = &lod_env_info(env)->lti_cookie;
307 struct lu_target *lut;
315 rc = lodname2mdt_index(lod2obd(lrd->lrd_lod)->obd_name, &index);
319 index = lrd->lrd_ltd->ltd_index;
323 llog_update_record_size((struct llog_update_record *)rec)) {
324 CERROR("%s: broken update record! index %u "DFID".%u: rc = %d\n",
325 lod2obd(lrd->lrd_lod)->obd_name, index,
326 PFID(&llh->lgh_id.lgl_oi.oi_fid), rec->lrh_index, -EIO);
330 cookie->lgc_lgl = llh->lgh_id;
331 cookie->lgc_index = rec->lrh_index;
332 cookie->lgc_subsys = LLOG_UPDATELOG_ORIG_CTXT;
334 CDEBUG(D_HA, "%s: process recovery updates "DFID".%u\n",
335 lod2obd(lrd->lrd_lod)->obd_name,
336 PFID(&llh->lgh_id.lgl_oi.oi_fid), rec->lrh_index);
337 lut = lod2lu_dev(lrd->lrd_lod)->ld_site->ls_tgt;
339 if (lod_recovery_abort(lut->lut_obd))
342 return insert_update_records_to_replay_list(lut->lut_tdtd,
343 (struct llog_update_record *)rec,
348 * recovery thread for update log
350 * Start recovery thread and prepare the sub llog, then it will retrieve
351 * the update records from the correpondent MDT and do recovery.
353 * \param[in] arg pointer to the recovery data
355 * \retval 0 if recovery succeeds
356 * \retval negative errno if recovery failed.
358 static int lod_sub_recovery_thread(void *arg)
360 struct lod_recovery_data *lrd = arg;
361 struct lod_device *lod = lrd->lrd_lod;
362 struct dt_device *dt;
363 struct llog_ctxt *ctxt = NULL;
364 struct lu_env *env = &lrd->lrd_env;
365 struct lu_target *lut;
366 struct lu_tgt_desc *mdt = NULL;
367 struct lu_device *top_device;
374 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
375 atomic_inc(&lut->lut_tdtd->tdtd_recovery_threads_count);
379 dt = lrd->lrd_ltd->ltd_tgt;
381 start = ktime_get_real_seconds();
382 complete(lrd->lrd_started);
386 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_TGT_RECOVERY_CONNECT)) &&
388 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_RECOVERY_CONNECT, cfs_fail_val);
391 rc = lod_sub_prep_llog(env, lod, dt, lrd->lrd_idx);
394 if (!rc && !lod->lod_child->dd_rdonly) {
395 /* Process the recovery record */
396 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
397 LLOG_UPDATELOG_ORIG_CTXT);
398 LASSERT(ctxt != NULL);
399 LASSERT(ctxt->loc_handle != NULL);
401 rc = llog_cat_process(env, ctxt->loc_handle,
402 lod_process_recovery_updates, lrd, 0, 0);
405 top_device = lod->lod_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
406 if (rc < 0 && dt != lod->lod_child &&
407 !lod_recovery_abort(top_device->ld_obd)) {
409 /* remote update llog is shorter than expected from
410 * local header. Cached copy could be de-synced during
411 * recovery, trust remote llog data
413 CDEBUG(D_HA, "%s update log data de-sync\n",
414 dt->dd_lu_dev.ld_obd->obd_name);
416 } else if (rc == -ETIMEDOUT || rc == -EAGAIN || rc == -EIO) {
418 * the remote target might failover at the same time,
422 if (ctxt->loc_handle)
423 llog_cat_close(env, ctxt->loc_handle);
428 CDEBUG(D_HA, "%s get update log failed %d, retry\n",
429 dt->dd_lu_dev.ld_obd->obd_name, rc);
436 CERROR("%s get update log failed: rc = %d\n",
437 dt->dd_lu_dev.ld_obd->obd_name, rc);
438 spin_lock(&top_device->ld_obd->obd_dev_lock);
439 if (!lod_recovery_abort(top_device->ld_obd))
440 /* abort just MDT-MDT recovery */
441 top_device->ld_obd->obd_abort_recov_mdt = 1;
442 spin_unlock(&top_device->ld_obd->obd_dev_lock);
446 CDEBUG(D_HA, "%s retrieved update log, duration %lld, retries %d\n",
447 dt->dd_lu_dev.ld_obd->obd_name, ktime_get_real_seconds() - start,
450 spin_lock(&lod->lod_lock);
452 lod->lod_child_got_update_log = 1;
454 lrd->lrd_ltd->ltd_got_update_log = 1;
456 if (!lod->lod_child_got_update_log) {
457 spin_unlock(&lod->lod_lock);
461 lod_foreach_mdt(lod, mdt) {
462 if (!mdt->ltd_got_update_log) {
463 spin_unlock(&lod->lod_lock);
467 lut->lut_tdtd->tdtd_replay_ready = 1;
468 spin_unlock(&lod->lod_lock);
470 CDEBUG(D_HA, "%s got update logs from all MDTs.\n",
471 lut->lut_obd->obd_name);
472 wake_up(&lut->lut_obd->obd_next_transno_waitq);
476 atomic_dec(&lut->lut_tdtd->tdtd_recovery_threads_count);
477 wake_up(&lut->lut_tdtd->tdtd_recovery_threads_waitq);
478 if (xchg(lrd->lrd_task, NULL) == NULL)
479 /* Someone is waiting for us to finish, need
480 * to synchronize cleanly.
482 wait_var_event(lrd, kthread_should_stop());
489 * finish sub llog context
491 * Stop update recovery thread for the sub device, then cleanup the
492 * correspondent llog ctxt.
494 * \param[in] env execution environment
495 * \param[in] lod lod device to do update recovery
496 * \param[in] thread recovery thread on this sub device
498 void lod_sub_fini_llog(const struct lu_env *env,
499 struct dt_device *dt, struct task_struct **thread)
501 struct obd_device *obd;
502 struct llog_ctxt *ctxt;
503 struct task_struct *task = NULL;
507 obd = dt->dd_lu_dev.ld_obd;
508 CDEBUG(D_INFO, "%s: finish sub llog\n", obd->obd_name);
509 /* Wait for recovery thread to complete */
511 task = xchg(thread, NULL);
515 ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
519 if (ctxt->loc_handle)
520 llog_cat_close(env, ctxt->loc_handle);
522 llog_cleanup(env, ctxt);
528 * Extract MDT target index from a device name.
530 * a helper function to extract index from the given device name
531 * like "fsname-MDTxxxx-mdtlov"
533 * \param[in] lodname device name
534 * \param[out] mdt_index extracted index
536 * \retval 0 on success
537 * \retval -EINVAL if the name is invalid
539 int lodname2mdt_index(char *lodname, u32 *mdt_index)
542 const char *ptr, *tmp;
545 /* 1.8 configs don't have "-MDT0000" at the end */
546 ptr = strstr(lodname, "-MDT");
552 ptr = strrchr(lodname, '-');
555 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
559 if (strncmp(ptr, "-mdtlov", 7) != 0) {
561 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
565 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
567 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
571 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
573 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
577 rc = target_name2index(ptr - 7, &index, &tmp);
578 if (rc < 0 || rc & LDD_F_SV_ALL || *tmp != '-') {
580 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
588 * Init sub llog context
590 * Setup update llog ctxt for update recovery threads, then start the
591 * recovery thread (lod_sub_recovery_thread) to read update llog from
592 * the correspondent MDT to do update recovery.
594 * \param[in] env execution environment
595 * \param[in] lod lod device to do update recovery
596 * \param[in] dt sub dt device for which the recovery thread is
598 * \retval 0 if initialization succeeds.
599 * \retval negative errno if initialization fails.
601 int lod_sub_init_llog(const struct lu_env *env, struct lod_device *lod,
602 struct dt_device *dt)
604 struct obd_device *obd;
605 struct lod_recovery_data *lrd = NULL;
606 DECLARE_COMPLETION_ONSTACK(started);
607 struct task_struct **taskp;
608 struct task_struct *task;
609 struct lod_tgt_desc *subtgt = NULL;
616 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &master_index);
624 if (lod->lod_child == dt) {
625 taskp = &lod->lod_child_recovery_task;
626 index = master_index;
628 struct lu_tgt_desc *mdt;
630 lod_foreach_mdt(lod, mdt) {
631 if (mdt->ltd_tgt == dt) {
632 index = mdt->ltd_index;
637 LASSERT(subtgt != NULL);
638 taskp = &subtgt->ltd_recovery_task;
641 CDEBUG(D_INFO, "%s init sub log %s\n", lod2obd(lod)->obd_name,
642 dt->dd_lu_dev.ld_obd->obd_name);
644 lrd->lrd_ltd = subtgt;
645 lrd->lrd_task = taskp;
646 lrd->lrd_idx = index;
647 lrd->lrd_started = &started;
649 obd = dt->dd_lu_dev.ld_obd;
650 obd->obd_lvfs_ctxt.dt = dt;
651 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_UPDATELOG_ORIG_CTXT,
652 NULL, &llog_common_cat_ops);
654 CERROR("%s: cannot setup updatelog llog: rc = %d\n",
659 rc = lu_env_init(&lrd->lrd_env, LCT_LOCAL | LCT_MD_THREAD);
661 CERROR("%s: can't initialize env: rc = %d\n",
662 lod2obd(lod)->obd_name, rc);
666 /* Start the recovery thread */
667 task = kthread_create(lod_sub_recovery_thread, lrd, "lod%04x_rec%04x",
668 master_index, index);
671 CERROR("%s: cannot start recovery thread: rc = %d\n",
673 lu_env_fini(&lrd->lrd_env);
677 wake_up_process(task);
678 wait_for_completion(&started);
682 lod_sub_fini_llog(env, dt, taskp);
689 * Stop sub recovery thread
691 * Stop sub recovery thread on all subs.
693 * \param[in] env execution environment
694 * \param[in] lod lod device to do update recovery
696 static void lod_sub_stop_recovery_threads(const struct lu_env *env,
697 struct lod_device *lod)
699 struct task_struct *task;
700 struct lu_tgt_desc *mdt;
703 * Stop the update log commit cancel threads and finish master
706 task = xchg(&lod->lod_child_recovery_task, NULL);
710 lod_getref(&lod->lod_mdt_descs);
711 lod_foreach_mdt(lod, mdt) {
712 task = xchg(&mdt->ltd_recovery_task, NULL);
716 lod_putref(lod, &lod->lod_mdt_descs);
720 * finish all sub llog
722 * cleanup all of sub llog ctxt on the LOD.
724 * \param[in] env execution environment
725 * \param[in] lod lod device to do update recovery
727 static void lod_sub_fini_all_llogs(const struct lu_env *env,
728 struct lod_device *lod)
730 struct lu_tgt_desc *mdt;
733 * Stop the update log commit cancel threads and finish master
736 lod_sub_fini_llog(env, lod->lod_child,
737 &lod->lod_child_recovery_task);
738 lod_getref(&lod->lod_mdt_descs);
739 lod_foreach_mdt(lod, mdt)
740 lod_sub_fini_llog(env, mdt->ltd_tgt,
741 &mdt->ltd_recovery_task);
742 lod_putref(lod, &lod->lod_mdt_descs);
745 static char *lod_show_update_logs_retrievers(void *data, int *size, int *count)
747 struct lod_device *lod = (struct lod_device *)data;
748 struct lu_target *lut = lod2lu_dev(lod)->ld_site->ls_tgt;
749 struct lu_tgt_desc *mdt = NULL;
755 *count = atomic_read(&lut->lut_tdtd->tdtd_recovery_threads_count);
761 *size = 5 * *count + 1;
762 OBD_ALLOC(buf, *size);
767 memset(buf, 0, *size);
769 if (!lod->lod_child_got_update_log) {
770 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &i);
771 LASSERTF(rc == 0, "Fail to parse target index: rc = %d\n", rc);
773 rc = scnprintf(buf + len, *size - len, " %04x", i);
780 lod_foreach_mdt(lod, mdt) {
781 if (!mdt->ltd_got_update_log) {
782 rc = scnprintf(buf + len, *size - len, " %04x",
784 if (unlikely(rc <= 0))
796 * Prepare distribute txn
798 * Prepare distribute txn structure for LOD
800 * \param[in] env execution environment
801 * \param[in] lod_device LOD device
803 * \retval 0 if preparation succeeds.
804 * \retval negative errno if preparation fails.
806 static int lod_prepare_distribute_txn(const struct lu_env *env,
807 struct lod_device *lod)
809 struct target_distribute_txn_data *tdtd;
810 struct lu_target *lut;
815 /* Init update recovery data */
820 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
821 tdtd->tdtd_dt = &lod->lod_dt_dev;
822 rc = distribute_txn_init(env, lut, tdtd,
823 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id);
826 CERROR("%s: cannot init distribute txn: rc = %d\n",
827 lod2obd(lod)->obd_name, rc);
832 tdtd->tdtd_show_update_logs_retrievers =
833 lod_show_update_logs_retrievers;
834 tdtd->tdtd_show_retrievers_cbdata = lod;
836 lut->lut_tdtd = tdtd;
842 * Finish distribute txn
844 * Release the resource holding by distribute txn, i.e. stop distribute
847 * \param[in] env execution environment
848 * \param[in] lod lod device
850 static void lod_fini_distribute_txn(const struct lu_env *env,
851 struct lod_device *lod)
853 struct lu_target *lut;
855 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
856 target_recovery_fini(lut->lut_obd);
860 distribute_txn_fini(env, lut->lut_tdtd);
862 OBD_FREE_PTR(lut->lut_tdtd);
863 lut->lut_tdtd = NULL;
867 * Implementation of lu_device_operations::ldo_process_config() for LOD
869 * The method is called by the configuration subsystem during setup,
870 * cleanup and when the configuration changes. The method processes
871 * few specific commands like adding/removing the targets, changing
872 * the runtime parameters.
874 * \param[in] env LU environment provided by the caller
875 * \param[in] dev lod device
876 * \param[in] lcfg configuration command to apply
878 * \retval 0 on success
879 * \retval negative negated errno on error
881 * The examples are below.
883 * Add osc config log:
884 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
885 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
886 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
887 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
888 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
889 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
891 * Add mdc config log:
892 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
893 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
894 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
895 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
896 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
897 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
899 static int lod_process_config(const struct lu_env *env,
900 struct lu_device *dev,
901 struct lustre_cfg *lcfg)
903 struct lod_device *lod = lu2lod_dev(dev);
904 struct lu_device *next = &lod->lod_child->dd_lu_dev;
910 switch (lcfg->lcfg_command) {
911 case LCFG_LOV_DEL_OBD:
912 case LCFG_LOV_ADD_INA:
913 case LCFG_LOV_ADD_OBD:
919 * lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
920 * modify_mdc_tgts add 0:lustre-MDT0001
921 * 1:lustre-MDT0001-mdc0002
924 arg1 = lustre_cfg_string(lcfg, 1);
926 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
927 GOTO(out, rc = -EINVAL);
928 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
929 GOTO(out, rc = -EINVAL);
931 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
934 rc = lodname2mdt_index(lustre_cfg_string(lcfg, 0),
939 rc = lod_add_device(env, lod, arg1, index, gen,
940 mdt_index, LUSTRE_OSC_NAME, 1);
941 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
943 rc = lod_add_device(env, lod, arg1, index, gen,
944 mdt_index, LUSTRE_MDC_NAME, 1);
945 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
946 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
948 rc = lod_add_device(env, lod, arg1, index, gen,
949 mdt_index, LUSTRE_OSC_NAME, 0);
951 rc = lod_del_device(env, lod, &lod->lod_ost_descs,
959 struct obd_device *obd;
964 * Check if it is activate/deactivate mdc
965 * lustre-MDTXXXX-osp-MDTXXXX.active=1
967 param = lustre_cfg_buf(lcfg, 1);
968 if (strstr(param, "osp") && strstr(param, ".active=")) {
969 struct lod_tgt_desc *sub_tgt = NULL;
970 struct lu_tgt_desc *mdt;
974 ptr = strstr(param, ".");
976 obd = class_name2obd(param);
978 CERROR("%s: can not find %s: rc = %d\n",
979 lod2obd(lod)->obd_name, param, -EINVAL);
984 lod_foreach_mdt(lod, mdt) {
985 if (mdt->ltd_tgt->dd_lu_dev.ld_obd == obd) {
992 CERROR("%s: can not find %s: rc = %d\n",
993 lod2obd(lod)->obd_name, param, -EINVAL);
999 tmp = strstr(param, "=");
1002 struct llog_ctxt *ctxt;
1004 obd = sub_tgt->ltd_tgt->dd_lu_dev.ld_obd;
1005 ctxt = llog_get_context(obd,
1006 LLOG_UPDATELOG_ORIG_CTXT);
1008 rc = llog_setup(env, obd, &obd->obd_olg,
1009 LLOG_UPDATELOG_ORIG_CTXT,
1010 NULL, &llog_common_cat_ops);
1014 llog_ctxt_put(ctxt);
1016 rc = lod_sub_prep_llog(env, lod,
1018 sub_tgt->ltd_index);
1020 sub_tgt->ltd_active = 1;
1022 lod_sub_fini_llog(env, sub_tgt->ltd_tgt,
1024 sub_tgt->ltd_active = 0;
1030 if (strstr(param, PARAM_LOD) != NULL)
1031 count = class_modify_config(lcfg, PARAM_LOD,
1032 &lod->lod_dt_dev.dd_kobj);
1034 count = class_modify_config(lcfg, PARAM_LOV,
1035 &lod->lod_dt_dev.dd_kobj);
1036 rc = count > 0 ? 0 : count;
1039 case LCFG_PRE_CLEANUP: {
1040 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1041 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1042 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_RECOVERY_CONNECT, cfs_fail_val * 2);
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);
1242 * Implementation of lu_device_operations::ldo_fid_alloc() for LOD
1244 * Find corresponding device by passed parent and name, and allocate FID from
1247 * see include/lu_object.h for the details.
1249 static int lod_fid_alloc(const struct lu_env *env, struct lu_device *d,
1250 struct lu_fid *fid, struct lu_object *parent,
1251 const struct lu_name *name)
1253 struct lod_device *lod = lu2lod_dev(d);
1254 struct lod_object *lo = lu2lod_obj(parent);
1255 struct dt_device *next;
1260 /* if @parent is remote, we don't know whether its layout was changed,
1261 * always reload layout.
1263 if (lu_object_remote(parent))
1264 lod_striping_free(env, lo);
1266 rc = lod_striping_load(env, lo);
1270 if (lo->ldo_dir_stripe_count > 0 && name) {
1271 struct dt_object *stripe;
1274 idx = __lmv_name_to_stripe_index(lo->ldo_dir_hash_type,
1275 lo->ldo_dir_stripe_count,
1276 lo->ldo_dir_migrate_hash,
1277 lo->ldo_dir_migrate_offset,
1279 name->ln_namelen, true);
1283 stripe = lo->ldo_stripe[idx];
1284 if (!stripe || !dt_object_exists(stripe))
1287 next = lu2dt_dev(stripe->do_lu.lo_dev);
1289 next = lod->lod_child;
1292 rc = dt_fid_alloc(env, next, fid, parent, name);
1297 const struct lu_device_operations lod_lu_ops = {
1298 .ldo_object_alloc = lod_object_alloc,
1299 .ldo_process_config = lod_process_config,
1300 .ldo_recovery_complete = lod_recovery_complete,
1301 .ldo_prepare = lod_prepare,
1302 .ldo_fid_alloc = lod_fid_alloc,
1306 * Implementation of dt_device_operations::dt_root_get() for LOD
1308 * see include/dt_object.h for the details.
1310 static int lod_root_get(const struct lu_env *env,
1311 struct dt_device *dev, struct lu_fid *f)
1313 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
1316 static void lod_statfs_sum(struct obd_statfs *sfs,
1317 struct obd_statfs *ost_sfs, int *bs)
1319 while (ost_sfs->os_bsize < *bs) {
1321 sfs->os_bsize >>= 1;
1322 sfs->os_bavail <<= 1;
1323 sfs->os_blocks <<= 1;
1324 sfs->os_bfree <<= 1;
1325 sfs->os_granted <<= 1;
1327 while (ost_sfs->os_bsize > *bs) {
1328 ost_sfs->os_bsize >>= 1;
1329 ost_sfs->os_bavail <<= 1;
1330 ost_sfs->os_blocks <<= 1;
1331 ost_sfs->os_bfree <<= 1;
1332 ost_sfs->os_granted <<= 1;
1334 sfs->os_bavail += ost_sfs->os_bavail;
1335 sfs->os_blocks += ost_sfs->os_blocks;
1336 sfs->os_bfree += ost_sfs->os_bfree;
1337 sfs->os_granted += ost_sfs->os_granted;
1341 * Implementation of dt_device_operations::dt_statfs() for LOD
1343 * see include/dt_object.h for the details.
1345 static int lod_statfs(const struct lu_env *env, struct dt_device *dev,
1346 struct obd_statfs *sfs, struct obd_statfs_info *info)
1348 struct lod_device *lod = dt2lod_dev(dev);
1349 struct lu_tgt_desc *tgt;
1350 struct obd_statfs ost_sfs;
1355 rc = dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
1364 sfs->os_granted = 0;
1366 lod_getref(&lod->lod_mdt_descs);
1367 lod_foreach_mdt(lod, tgt) {
1368 rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
1372 sfs->os_files += ost_sfs.os_files;
1373 sfs->os_ffree += ost_sfs.os_ffree;
1374 lod_statfs_sum(sfs, &ost_sfs, &bs);
1376 lod_putref(lod, &lod->lod_mdt_descs);
1379 * at some point we can check whether DoM is enabled and
1380 * decide how to account MDT space. for simplicity let's
1381 * just fallback to pre-DoM policy if any OST is alive
1383 lod_getref(&lod->lod_ost_descs);
1384 lod_foreach_ost(lod, tgt) {
1385 rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
1387 if (rc || ost_sfs.os_bsize == 0)
1391 * if only MDTs with DoM then report only MDT blocks,
1392 * otherwise show only OST blocks, and DoM is "free"
1397 sfs->os_granted = 0;
1399 ost_files += ost_sfs.os_files;
1400 ost_ffree += ost_sfs.os_ffree;
1401 ost_sfs.os_bavail += ost_sfs.os_granted;
1402 lod_statfs_sum(sfs, &ost_sfs, &bs);
1403 LASSERTF(bs == ost_sfs.os_bsize, "%d != %d\n",
1404 (int)sfs->os_bsize, (int)ost_sfs.os_bsize);
1406 lod_putref(lod, &lod->lod_ost_descs);
1407 sfs->os_state |= OS_STATFS_SUM;
1409 /* If we have _some_ OSTs, but don't have as many free objects on the
1410 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1411 * to compensate, so that the "inodes in use" number is correct.
1412 * This should be kept in sync with ll_statfs_internal().
1414 if (ost_files && ost_ffree < sfs->os_ffree) {
1415 sfs->os_files = (sfs->os_files - sfs->os_ffree) + ost_ffree;
1416 sfs->os_ffree = ost_ffree;
1419 /* a single successful statfs should be enough */
1427 * Implementation of dt_device_operations::dt_trans_create() for LOD
1429 * Creates a transaction using local (to this node) OSD.
1431 * see include/dt_object.h for the details.
1433 static struct thandle *lod_trans_create(const struct lu_env *env,
1434 struct dt_device *dt)
1438 th = top_trans_create(env, dt2lod_dev(dt)->lod_child);
1448 * Implementation of dt_device_operations::dt_trans_start() for LOD
1450 * Starts the set of local transactions using the targets involved
1451 * in declare phase. Initial support for the distributed transactions.
1453 * see include/dt_object.h for the details.
1455 static int lod_trans_start(const struct lu_env *env, struct dt_device *dt,
1458 return top_trans_start(env, dt2lod_dev(dt)->lod_child, th);
1461 static int lod_trans_cb_add(struct thandle *th,
1462 struct dt_txn_commit_cb *dcb)
1464 struct top_thandle *top_th = container_of(th, struct top_thandle,
1466 return dt_trans_cb_add(top_th->tt_master_sub_thandle, dcb);
1470 * add noop update to the update records
1472 * Add noop updates to the update records, which is only used in
1475 * \param[in] env execution environment
1476 * \param[in] dt dt device of lod
1477 * \param[in] th thandle
1478 * \param[in] count the count of update records to be added.
1480 * \retval 0 if adding succeeds.
1481 * \retval negative errno if adding fails.
1483 static int lod_add_noop_records(const struct lu_env *env,
1484 struct dt_device *dt, struct thandle *th,
1487 struct top_thandle *top_th;
1488 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1492 top_th = container_of(th, struct top_thandle, tt_super);
1493 if (!top_th->tt_multiple_thandle)
1497 for (i = 0; i < count; i++) {
1498 rc = update_record_pack(noop, th, fid);
1506 * Implementation of dt_device_operations::dt_trans_stop() for LOD
1508 * Stops the set of local transactions using the targets involved
1509 * in declare phase. Initial support for the distributed transactions.
1511 * see include/dt_object.h for the details.
1513 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
1516 if (OBD_FAIL_CHECK(OBD_FAIL_SPLIT_UPDATE_REC)) {
1519 rc = lod_add_noop_records(env, dt, th, 5000);
1523 return top_trans_stop(env, dt2lod_dev(dt)->lod_child, th);
1527 * Implementation of dt_device_operations::dt_conf_get() for LOD
1529 * Currently returns the configuration provided by the local OSD.
1531 * see include/dt_object.h for the details.
1533 static void lod_conf_get(const struct lu_env *env,
1534 const struct dt_device *dev,
1535 struct dt_device_param *param)
1537 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
1541 * Implementation of dt_device_operations::dt_sync() for LOD
1543 * Syncs all known OST targets. Very very expensive and used
1544 * rarely by LFSCK now. Should not be used in general.
1546 * see include/dt_object.h for the details.
1548 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
1550 struct lod_device *lod = dt2lod_dev(dev);
1551 struct lu_tgt_desc *tgt;
1556 lod_getref(&lod->lod_ost_descs);
1557 lod_foreach_ost(lod, tgt) {
1558 if (!tgt->ltd_active)
1560 rc = dt_sync(env, tgt->ltd_tgt);
1562 if (rc != -ENOTCONN) {
1563 CERROR("%s: can't sync ost %u: rc = %d\n",
1564 lod2obd(lod)->obd_name, tgt->ltd_index,
1571 lod_putref(lod, &lod->lod_ost_descs);
1576 lod_getref(&lod->lod_mdt_descs);
1577 lod_foreach_mdt(lod, tgt) {
1578 if (!tgt->ltd_active)
1580 rc = dt_sync(env, tgt->ltd_tgt);
1582 if (rc != -ENOTCONN) {
1583 CERROR("%s: can't sync mdt %u: rc = %d\n",
1584 lod2obd(lod)->obd_name, tgt->ltd_index,
1591 lod_putref(lod, &lod->lod_mdt_descs);
1594 rc = dt_sync(env, lod->lod_child);
1600 * Implementation of dt_device_operations::dt_ro() for LOD
1602 * Turns local OSD read-only, used for the testing only.
1604 * see include/dt_object.h for the details.
1606 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
1608 return dt_ro(env, dt2lod_dev(dev)->lod_child);
1612 * Implementation of dt_device_operations::dt_commit_async() for LOD
1614 * Asks local OSD to commit sooner.
1616 * see include/dt_object.h for the details.
1618 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
1620 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
1623 static const struct dt_device_operations lod_dt_ops = {
1624 .dt_root_get = lod_root_get,
1625 .dt_statfs = lod_statfs,
1626 .dt_trans_create = lod_trans_create,
1627 .dt_trans_start = lod_trans_start,
1628 .dt_trans_stop = lod_trans_stop,
1629 .dt_conf_get = lod_conf_get,
1630 .dt_sync = lod_sync,
1632 .dt_commit_async = lod_commit_async,
1633 .dt_trans_cb_add = lod_trans_cb_add,
1637 * Connect to a local OSD.
1639 * Used to connect to the local OSD at mount. OSD name is taken from the
1640 * configuration command passed. This connection is used to identify LU
1641 * site and pin the OSD from early removal.
1643 * \param[in] env LU environment provided by the caller
1644 * \param[in] lod lod device
1645 * \param[in] cfg configuration command to apply
1647 * \retval 0 on success
1648 * \retval negative negated errno on error
1650 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
1651 struct lustre_cfg *cfg)
1653 struct obd_connect_data *data = NULL;
1654 struct obd_device *obd;
1655 char *nextdev = NULL, *p, *s;
1661 LASSERT(lod->lod_child_exp == NULL);
1664 * compatibility hack: we still use old config logs
1665 * which specify LOV, but we need to learn underlying
1666 * OSD device, which is supposed to be:
1667 * <fsname>-MDTxxxx-osd
1669 * 2.x MGS generates lines like the following:
1670 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
1671 * 1.8 MGS generates lines like the following:
1672 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
1674 * we use "-MDT" to differentiate 2.x from 1.8
1676 p = lustre_cfg_string(cfg, 0);
1677 if (p && strstr(p, "-mdtlov")) {
1678 len = strlen(p) + 6;
1679 OBD_ALLOC(nextdev, len);
1681 GOTO(out, rc = -ENOMEM);
1684 s = strstr(nextdev, "-mdtlov");
1686 CERROR("%s: unable to parse device name: rc = %d\n",
1687 lustre_cfg_string(cfg, 0), -EINVAL);
1688 GOTO(out, rc = -EINVAL);
1691 if (strstr(nextdev, "-MDT")) {
1696 strcpy(s, "-MDT0000-osd");
1699 CERROR("%s: unable to parse device name: rc = %d\n",
1700 lustre_cfg_string(cfg, 0), -EINVAL);
1701 GOTO(out, rc = -EINVAL);
1704 OBD_ALLOC_PTR(data);
1706 GOTO(out, rc = -ENOMEM);
1708 obd = class_name2obd(nextdev);
1710 CERROR("%s: can not locate next device: rc = %d\n",
1711 nextdev, -ENOTCONN);
1712 GOTO(out, rc = -ENOTCONN);
1715 data->ocd_connect_flags = OBD_CONNECT_VERSION;
1716 data->ocd_version = LUSTRE_VERSION_CODE;
1718 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
1721 CERROR("%s: cannot connect to next dev: rc = %d\n",
1726 lod->lod_dt_dev.dd_lu_dev.ld_site =
1727 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
1728 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
1729 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
1735 OBD_FREE(nextdev, len);
1739 static int lod_lsfs_init(const struct lu_env *env, struct lod_device *d)
1741 struct obd_statfs sfs;
1744 rc = dt_statfs(env, d->lod_child, &sfs);
1746 CDEBUG(D_LAYOUT, "%s: failed to get OSD statfs, rc = %d\n",
1747 lod2obd(d)->obd_name, rc);
1751 /* udpate local OSD cached statfs data */
1752 spin_lock_init(&d->lod_lsfs_lock);
1753 d->lod_lsfs_age = ktime_get_seconds();
1754 d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
1755 d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
1760 * Initialize LOD device at setup.
1762 * Initializes the given LOD device using the original configuration command.
1763 * The function initiates a connection to the local OSD and initializes few
1764 * internal structures like pools, target tables, etc.
1766 * \param[in] env LU environment provided by the caller
1767 * \param[in] lod lod device
1768 * \param[in] ldt not used
1769 * \param[in] cfg configuration command
1771 * \retval 0 on success
1772 * \retval negative negated errno on error
1774 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
1775 struct lu_device_type *ldt, struct lustre_cfg *cfg)
1777 struct dt_device_param ddp;
1778 struct obd_device *obd;
1783 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1786 CERROR("Cannot find obd with name '%s': rc = %d\n",
1787 lustre_cfg_string(cfg, 0), rc);
1791 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1792 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
1793 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
1794 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
1796 rc = lod_connect_to_osd(env, lod, cfg);
1800 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
1801 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
1802 lod->lod_dom_stripesize_max_kb = (1ULL << 10); /* 1Mb is default */
1803 lod->lod_max_stripecount = 0;
1805 /* initialize local statfs cached values */
1806 rc = lod_lsfs_init(env, lod);
1808 GOTO(out_disconnect, rc);
1810 /* default threshold as half of total space, in MiB */
1811 lod->lod_dom_threshold_free_mb = lod->lod_lsfs_total_mb / 2;
1812 /* set default DoM stripe size based on free space amount */
1813 lod_dom_stripesize_recalc(lod);
1815 /* setup obd to be used with old lov code */
1816 rc = lod_pools_init(lod, cfg);
1818 GOTO(out_disconnect, rc);
1820 rc = lod_procfs_init(lod);
1822 GOTO(out_pools, rc);
1824 spin_lock_init(&lod->lod_lock);
1825 spin_lock_init(&lod->lod_connects_lock);
1826 lu_tgt_descs_init(&lod->lod_mdt_descs, true);
1827 lu_tgt_descs_init(&lod->lod_ost_descs, false);
1828 lu_qos_rr_init(&lod->lod_mdt_descs.ltd_qos.lq_rr);
1829 lu_qos_rr_init(&lod->lod_ost_descs.ltd_qos.lq_rr);
1834 lod_pools_fini(lod);
1836 obd_disconnect(lod->lod_child_exp);
1841 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
1843 * Releases the memory allocated for LOD device.
1845 * see include/lu_object.h for the details.
1847 static struct lu_device *lod_device_free(const struct lu_env *env,
1848 struct lu_device *lu)
1850 struct lod_device *lod = lu2lod_dev(lu);
1851 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1855 if (atomic_read(&lu->ld_site->ls_obj_hash.nelems)) {
1856 lu_site_print(env, lu->ld_site, &lu->ld_ref, D_ERROR,
1859 LASSERTF(atomic_read(&lu->ld_ref) == 0, "lu is %p\n", lu);
1860 dt_device_fini(&lod->lod_dt_dev);
1866 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
1868 * Allocates LOD device and calls the helpers to initialize it.
1870 * see include/lu_object.h for the details.
1872 static struct lu_device *lod_device_alloc(const struct lu_env *env,
1873 struct lu_device_type *type,
1874 struct lustre_cfg *lcfg)
1876 struct lod_device *lod;
1877 struct lu_device *lu_dev;
1881 lu_dev = ERR_PTR(-ENOMEM);
1885 lu_dev = lod2lu_dev(lod);
1886 dt_device_init(&lod->lod_dt_dev, type);
1887 rc = lod_init0(env, lod, type, lcfg);
1889 lod_device_free(env, lu_dev);
1890 lu_dev = ERR_PTR(rc);
1897 static void lod_avoid_guide_fini(struct lod_avoid_guide *lag)
1899 if (lag->lag_oss_avoid_array)
1900 OBD_FREE_PTR_ARRAY(lag->lag_oss_avoid_array,
1902 bitmap_free(lag->lag_ost_avoid_bitmap);
1906 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
1908 * Releases the internal resources used by LOD device.
1910 * see include/lu_object.h for the details.
1912 static struct lu_device *lod_device_fini(const struct lu_env *env,
1913 struct lu_device *d)
1915 struct lod_device *lod = lu2lod_dev(d);
1920 lod_pools_fini(lod);
1922 lod_procfs_fini(lod);
1924 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs);
1926 CERROR("%s: can not fini ost descriptors: rc = %d\n",
1927 lod2obd(lod)->obd_name, rc);
1929 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs);
1931 CERROR("%s: can not fini mdt descriptors: rc = %d\n",
1932 lod2obd(lod)->obd_name, rc);
1938 * Implementation of obd_ops::o_connect() for LOD
1940 * Used to track all the users of this specific LOD device,
1941 * so the device stays up until the last user disconnected.
1943 * \param[in] env LU environment provided by the caller
1944 * \param[out] exp export the caller will be using to access LOD
1945 * \param[in] obd OBD device representing LOD device
1946 * \param[in] cluuid unique identifier of the caller
1947 * \param[in] data not used
1948 * \param[in] localdata not used
1950 * \retval 0 on success
1951 * \retval negative negated errno on error
1953 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
1954 struct obd_device *obd, struct obd_uuid *cluuid,
1955 struct obd_connect_data *data, void *localdata)
1957 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1958 struct lustre_handle conn;
1963 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1965 rc = class_connect(&conn, obd, cluuid);
1969 *exp = class_conn2export(&conn);
1971 spin_lock(&lod->lod_connects_lock);
1972 lod->lod_connects++;
1973 /* at the moment we expect the only user */
1974 LASSERT(lod->lod_connects == 1);
1975 spin_unlock(&lod->lod_connects_lock);
1982 * Implementation of obd_ops::o_disconnect() for LOD
1984 * When the caller doesn't need to use this LOD instance, it calls
1985 * obd_disconnect() and LOD releases corresponding export/reference count.
1986 * Once all the users gone, LOD device is released.
1988 * \param[in] exp export provided to the caller in obd_connect()
1990 * \retval 0 on success
1991 * \retval negative negated errno on error
1993 static int lod_obd_disconnect(struct obd_export *exp)
1995 struct obd_device *obd = exp->exp_obd;
1996 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1997 int rc, release = 0;
2001 /* Only disconnect the underlying layers on the final disconnect. */
2002 spin_lock(&lod->lod_connects_lock);
2003 lod->lod_connects--;
2004 if (lod->lod_connects != 0) {
2005 /* why should there be more than 1 connect? */
2006 spin_unlock(&lod->lod_connects_lock);
2007 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
2011 spin_unlock(&lod->lod_connects_lock);
2013 /* the last user of lod has gone, let's release the device */
2017 rc = class_disconnect(exp); /* bz 9811 */
2019 if (rc == 0 && release)
2020 class_manual_cleanup(obd);
2024 LU_KEY_INIT(lod, struct lod_thread_info);
2026 static void lod_key_fini(const struct lu_context *ctx,
2027 struct lu_context_key *key, void *data)
2029 struct lod_thread_info *info = data;
2030 struct lod_layout_component *lds =
2031 info->lti_def_striping.lds_def_comp_entries;
2034 * allocated in lod_get_lov_ea
2035 * XXX: this is overload, a tread may have such store but used only
2036 * once. Probably better would be pool of such stores per LOD.
2038 if (info->lti_ea_store) {
2039 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
2040 info->lti_ea_store = NULL;
2041 info->lti_ea_store_size = 0;
2043 lu_buf_free(&info->lti_linkea_buf);
2046 lod_free_def_comp_entries(&info->lti_def_striping);
2048 if (info->lti_comp_size > 0)
2049 OBD_FREE_PTR_ARRAY(info->lti_comp_idx,
2050 info->lti_comp_size);
2052 lod_avoid_guide_fini(&info->lti_avoid);
2057 /* context key: lod_thread_key */
2058 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
2060 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
2062 static const struct lu_device_type_operations lod_device_type_ops = {
2063 .ldto_init = lod_type_init,
2064 .ldto_fini = lod_type_fini,
2066 .ldto_start = lod_type_start,
2067 .ldto_stop = lod_type_stop,
2069 .ldto_device_alloc = lod_device_alloc,
2070 .ldto_device_free = lod_device_free,
2072 .ldto_device_fini = lod_device_fini
2075 static struct lu_device_type lod_device_type = {
2076 .ldt_tags = LU_DEVICE_DT,
2077 .ldt_name = LUSTRE_LOD_NAME,
2078 .ldt_ops = &lod_device_type_ops,
2079 .ldt_ctx_tags = LCT_MD_THREAD,
2083 * Implementation of obd_ops::o_get_info() for LOD
2085 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
2086 * the caller binary status whether LOD has seen connection to any OST target.
2087 * It will also check if the MDT update log context being initialized (if
2090 * \param[in] env LU environment provided by the caller
2091 * \param[in] exp export of the caller
2092 * \param[in] keylen len of the key
2093 * \param[in] key the key
2094 * \param[in] vallen not used
2095 * \param[in] val not used
2097 * \retval 0 if a connection was seen
2098 * \retval -EAGAIN if LOD isn't running yet or no
2099 * connection has been seen yet
2100 * \retval -EINVAL if not supported key is requested
2102 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
2103 u32 keylen, void *key, u32 *vallen, void *val)
2107 if (KEY_IS(KEY_OSP_CONNECTED)) {
2108 struct obd_device *obd = exp->exp_obd;
2109 struct lod_device *d;
2110 struct lod_tgt_desc *tgt;
2113 if (!obd->obd_set_up || obd->obd_stopping)
2116 d = lu2lod_dev(obd->obd_lu_dev);
2117 lod_getref(&d->lod_ost_descs);
2118 lod_foreach_ost(d, tgt) {
2119 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2121 /* one healthy device is enough */
2125 lod_putref(d, &d->lod_ost_descs);
2127 lod_getref(&d->lod_mdt_descs);
2128 lod_foreach_mdt(d, tgt) {
2129 struct llog_ctxt *ctxt;
2131 if (!tgt->ltd_active)
2134 ctxt = llog_get_context(tgt->ltd_tgt->dd_lu_dev.ld_obd,
2135 LLOG_UPDATELOG_ORIG_CTXT);
2137 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2139 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2143 if (!ctxt->loc_handle) {
2144 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2146 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2148 llog_ctxt_put(ctxt);
2151 llog_ctxt_put(ctxt);
2153 lod_putref(d, &d->lod_mdt_descs);
2161 static int lod_obd_set_info_async(const struct lu_env *env,
2162 struct obd_export *exp,
2163 u32 keylen, void *key,
2164 u32 vallen, void *val,
2165 struct ptlrpc_request_set *set)
2167 struct obd_device *obd = class_exp2obd(exp);
2168 struct lod_device *d;
2169 struct lod_tgt_desc *tgt;
2177 set = ptlrpc_prep_set();
2182 d = lu2lod_dev(obd->obd_lu_dev);
2183 lod_getref(&d->lod_ost_descs);
2184 lod_foreach_ost(d, tgt) {
2185 if (!tgt->ltd_active)
2188 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2190 if (rc2 != 0 && rc == 0)
2193 lod_putref(d, &d->lod_ost_descs);
2195 lod_getref(&d->lod_mdt_descs);
2196 lod_foreach_mdt(d, tgt) {
2197 if (!tgt->ltd_active)
2199 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2201 if (rc2 != 0 && rc == 0)
2204 lod_putref(d, &d->lod_mdt_descs);
2208 rc2 = ptlrpc_set_wait(env, set);
2209 if (rc2 == 0 && rc == 0)
2211 ptlrpc_set_destroy(set);
2217 #define QMT0_DEV_NAME_LEN (LUSTRE_MAXFSNAME + sizeof("-QMT0000"))
2218 static struct obd_device *obd_find_qmt0(char *obd_name)
2220 char qmt_name[QMT0_DEV_NAME_LEN];
2221 struct obd_device *qmt = NULL;
2223 if (!server_name2fsname(obd_name, qmt_name, NULL)) {
2224 strlcat(qmt_name, "-QMT0000", QMT0_DEV_NAME_LEN);
2225 qmt = class_name2obd(qmt_name);
2231 /* Run QMT0000 pool operations only for MDT0000 */
2232 static inline bool lod_pool_need_qmt0(const char *obd_name)
2237 type = server_name2index(obd_name, &idx, NULL);
2239 return type == LDD_F_SV_TYPE_MDT && idx == 0;
2242 static int lod_pool_new_q(struct obd_device *obd, char *poolname)
2244 int err = lod_pool_new(obd, poolname);
2246 if (!err && lod_pool_need_qmt0(obd->obd_name)) {
2247 obd = obd_find_qmt0(obd->obd_name);
2249 obd_pool_new(obd, poolname);
2255 static int lod_pool_remove_q(struct obd_device *obd, char *poolname,
2258 int err = lod_pool_remove(obd, poolname, ostname);
2260 if (!err && lod_pool_need_qmt0(obd->obd_name)) {
2261 obd = obd_find_qmt0(obd->obd_name);
2263 obd_pool_rem(obd, poolname, ostname);
2269 static int lod_pool_add_q(struct obd_device *obd, char *poolname, char *ostname)
2271 int err = lod_pool_add(obd, poolname, ostname);
2273 if (!err && lod_pool_need_qmt0(obd->obd_name)) {
2274 obd = obd_find_qmt0(obd->obd_name);
2276 obd_pool_add(obd, poolname, ostname);
2282 static int lod_pool_del_q(struct obd_device *obd, char *poolname)
2284 int err = lod_pool_del(obd, poolname);
2286 if (!err && lod_pool_need_qmt0(obd->obd_name)) {
2287 obd = obd_find_qmt0(obd->obd_name);
2289 obd_pool_del(obd, poolname);
2295 static const struct obd_ops lod_obd_device_ops = {
2296 .o_owner = THIS_MODULE,
2297 .o_connect = lod_obd_connect,
2298 .o_disconnect = lod_obd_disconnect,
2299 .o_get_info = lod_obd_get_info,
2300 .o_set_info_async = lod_obd_set_info_async,
2301 .o_pool_new = lod_pool_new_q,
2302 .o_pool_rem = lod_pool_remove_q,
2303 .o_pool_add = lod_pool_add_q,
2304 .o_pool_del = lod_pool_del_q,
2307 static int __init lod_init(void)
2309 struct obd_type *sym;
2312 rc = lu_kmem_init(lod_caches);
2316 rc = class_register_type(&lod_obd_device_ops, NULL, true,
2317 LUSTRE_LOD_NAME, &lod_device_type);
2319 lu_kmem_fini(lod_caches);
2323 /* create "lov" entry for compatibility purposes */
2324 sym = class_add_symlinks(LUSTRE_LOV_NAME, true);
2327 /* does real "lov" already exist ? */
2335 static void __exit lod_exit(void)
2337 struct obd_type *sym = class_search_type(LUSTRE_LOV_NAME);
2339 /* if this was never fully initialized by the lov layer
2340 * then we are responsible for freeing this obd_type
2343 /* final put if we manage this obd type */
2344 if (sym->typ_sym_filter)
2345 kobject_put(&sym->typ_kobj);
2346 /* put reference taken by class_search_type */
2347 kobject_put(&sym->typ_kobj);
2350 class_unregister_type(LUSTRE_LOD_NAME);
2351 lu_kmem_fini(lod_caches);
2354 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2355 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
2356 MODULE_VERSION(LUSTRE_VERSION_STRING);
2357 MODULE_LICENSE("GPL");
2359 module_init(lod_init);
2360 module_exit(lod_exit);